Intel used number their CPU models, until they lost some sort of trademark case and decided to use actual names. That's when they first introduced Pentium, as the successor to the i486. With Penta being the Greek prefix for 5, those who didn't know their Latin from their Greek would joke that the following generation CPU should be called the Sexium.
Of course, Intel probably didn't much fancy Hexium either, since they stayed stuck on Pentium-based names until the somehow even worse-named "Core" branding. And that would later be outdone by their Xe branding for GPUs, which is the chemical symbol for the element Xenon -- far too easily confused with their Xeon branding of server/workstation CPUs. I digress...
Not to mention that xenon was once classified under ‘inert’ — which reminds of the apocryphal tale of the ‘no go’ Chevrolet Nova.
Perhaps worse yet is that the compounds made with Xenon are all much too unstable to survive outside of very specific controlled conditions — as with Intel’s industrial chiller it used to demo its 28 core 14nm 5 GHz ‘Oops, we forgot’ CPU.
"In practice, I expect the 5800X3D to live (or die) based on pricing. AMD’s 12-core Ryzen 5900X is also $449, so buyers who are willing to part with that money will be choosing between 4 more CPU cores or 64MB more L3 cache."
It's just 32MB more L3 cache as it's 64MB for the 5800X and 96MB for the 5800X3D since the 5900X is a dual (6 core) die setup each with the full 32MB of L3 where the 5800X3D is a single 8 core die with 32MB of L3 and 64MB of V-Cache (or whatever they're calling it).
It really makes the 5900X look like the better choice. The 5800X3D is clocked more like a 5700X after all.
In practice, I expect the 5800X3D to perform better than the 5900X on all tasks that do not peg all 12 cores to 100%. And will be noticably better on all lightly threaded tasks. So a better fit for at least 90% of people that can spare that kind of dosh.
Also, it would be interesting to see how it overclocks. If it OCs at least as well as the 5900X it's an instant purchase for me (coming from a 3600X). Otherwise I might look at 5700X/5800X as better bang for buck options.
Signs so far point to the 5800X3D being non-overclockable (or at least not officially). Not a surprise with having the manage clocks across a bus that spans multiple dies without an explicit interconnect interface.
That whole "leak" strikes me as a bit of a joke, for a very basic reason, CPU abilities are enabled/disabled based on the BIOS. If the current AGESA versions do not allow for overclocking due to AMD wanting to keep the leaks from revealing too much, then that would make a lot more sense than not allowing overclocking at all. At CES and the interviews after it, it was strongly hinted that if you put better cooling on the 5800X3D, it will overclock up to the same speeds as the regular 5800X, but that will obviously go over the TDP rating of the CPU.
AMD has actually played with the leak community over the years by releasing different samples to different places, just to see where the leaks come from. If the leaks line up with the version that went to Asus for example, then bingo, the leak comes from there, and AMD can respond properly. That is why the true leaks about Zen4 have been very few and far between.
It has been confirmed. The 3d cache is the part that prevents OC in this cpu. The cache does not handle overclocking at all, so PBO and other kind of OC is prevented to protect the cache.
That's really going to depend on the boosts and heat island effect vs. the extra cache. For lightly threaded/CPU-bound tasks, on a 5900X two cores on separate chips can boost fairly high as neither significantly heats the other. Of course, if they're the same app using the same memory space, they effectively share 32MB of cache and that's cache-bound..
> It's just 32MB more L3 cache as it's 64MB for the 5800X and 96MB for the 5800X3D > since the 5900X is a dual (6 core) die setup each with the full 32MB of L3
Here's where mere sums don't tell the whole story. I'll just quote from the original Zen 3 review:
"The Ryzen 7 5800X ... uses a single eight-core chiplet with a total 32 MB of L3 cache. The single .. chiplet (design) has some small benefits over a dual chiplet design where some cross-CPU communication is needed, and that comes across in some of our very CPU-limited gaming benchmarks."
So, that explains AMD's rationale. Having 64 MB split across two chiplets means having to go through the I/O die for cores to get at half of it. Whereas AMD has quoted vastly higher throughput numbers for accessing L3 stacked on the same chiplet, which I'll quote from another article:
"AMD claims that the total bandwidth of the L3 cache increases to beyond 2 TB/sec, which would technically be faster than the L1 cache on the die"
12 is alot of cores for normal workloads. Unless you are doing something very specific that needs that many cores... Most people would be better off with more cache. Thus the expected gaming improvements. - "AMD is specifically pitching this 8 core processor as the ultimate chip for gaming"
I thought this was a pretty good write up with one small exception...taking yesterday's high-end CPU/GPUs and cycling them back to the lower/bottom end of markets is nothing new or unprecedented. That's sort of routine, really. Zen 2 will be three years old this July, and to see it cycle back to the bottom end is hardly surprising, especially when the debut for Zen 4 is only a few months away, still slated for 2022.
@WaltC, Cycling back, good point, part of the AMD sales objective building a graded by gen and SKU performance CPU price rung stack in the secondary market supports every next generation price attempting to reach a new (by category and segment) price ceiling. Intel's been doing this in performance workstation / gaming mobile for a long time. Not as much as in desktop although did with Comet deca otherwise desktop price rungs have been fairly stable.
At the back in time opposing product price floor, the foundation, also creates what is known as a carpet and both AMD and Intel and Nvidia have carpets designed to keep low cost : price commodity entrants out, coming up from under the floorboards as it were.
On CPU side ARM and Apple are being blocked with a carpet where Apple attempts to come in from the top but then think Snapdragon on Windows 10 and the world market for ARM consumer mobile that are generally devices priced less than $300. The carpet is a defensive move just like (over) segmentation.
In the Xeon market AMD Epyc and ARM both came in from over the top of Intel core grades because there were no performance price rung voids in the highly segmented, at the time, Xeon 4 to 24 core market.
Production of surplus, overage, also plays a part in this game because the overage holds channels financially on their inventory accumulated capital values. More surplus means more capital values that every broker holds and hopes they can flip by finding every next home for them before the component becomes obsolete. There were three times when this became a severe issue on Intel over acceleration of primary production sticking the channel with a prematurely obsolete inventory; in 2001, 2005, 2007/8. Intel did this to take back cartel (PC dealer) price making control from the channel. Today FTC Docket 9341 guards and monitors for this economically debilitating practice destroying accumulated capital value, and Intel has been reducing surplus production since Rocket because of the unnecessary costs involved both in terms of Intel production added cost. Less unwanted product every generation means cost optimization and lowers the risk of channel financial losses.
Intel has a mandate to reconfigure the enterprise from producing for supply to producing for demand that is a demand and supply balancing. This move supports industry commerce by reducing limits placed on competitors by any dominant manufacturer / producer attempting to hold a market through over production on surplus volume.
On the Vermeer Refresh front, AMD averaged 13 M units of Ryzen performance desktop per quarter in 2021. Intel Rocket Lake was around 27 M units most of it deadweight that should never have been produced and is affecting the channel financially in terms of inventory value cost : price margin losses that were primarily made up taxing buyers clandestinely on inflated sGPU pricing paid for a lot of channel's Rocket, Comet down bin and Coffee Refresh single thread mid shelf losses so it does still happen but improving. I estimate Alder as a 30 M unit run that will be mirrored by Raptor.
I'm quite disappointed with the announcements. There's not much performance uplift or price decrease really to get hyped about.
Ever since the Zen2 release, where AMD has matched and surpassed Intel. Well they've put on the brakes. Zen3 was an upgrade but not huge, yet prices were much higher. And the chip shortage/pandemic merely secured their place for an extra two years.
Even after the debut of the Apple M1, Pro, Max, and Ultra, which put Apple ahead of the industry giants (albeit on macOS) nothing much changed at AMD.
I'm hoping that has all been on purpose. That Zen4 is a proper upgrade over Zen3 (not just a +refresh), and that RDNA-3 is also substantial. And that AMD is going to slowly discontinue their chips which have Vega Graphics, and Zen2 architecture. That this is merely TSMC and AMD getting rid of potential stock.
You know start fresh. With a new mobo platform, new pcie standard, and new DDR5 memory. And they're all Zen4, RDNA-3, and APUs. Hopefully they will come at a more reasonable price, with no nasty surprises from security or availability.
Here's what Ian and Andrei wrote, in their conclusion of the Ryzen 5000 review:
"In real world benchmarks, we saw an average +24% performance gain"
"All things considered, we’re really impressed with what AMD has achieved here."
"Zen3 gets a gold award. No question."
But you know better? M'kay. It's just a pity they're not here to show you what's what.
> Even after the debut of the Apple M1, Pro, Max, and Ultra, ... nothing much changed at AMD.
AMD reacts to its market, its customers, and its partners. I doubt the M1 has caused huge disruptions in these areas, but maybe you can provide evidence demonstrating otherwise.
Even if you're right, what would you *expect* them to do? Pretty much the only knob they can twiddle, in the short term, is pricing vs. features. That's it. They can't turn-around a whole new CPU in even the time since the first M1 hit. The pipeline for new silicon is far longer than that.
In the current world where AMD is selling basically all the silicon it can get from TSMC at the current price structure, AMD is doing just fine. It doesn't need to do anything big... yet.
> I'm hoping ... AMD is going to slowly discontinue their chips which have > Vega Graphics, and Zen2 architecture.
Okay, for a refresh, but that's about it. Doesn't really deserve 6000-series branding, but then it *is* a meaningful step beyond the 5000-series stack.
> You know start fresh. With a new mobo platform, new pcie standard, and new DDR5 memory.
That's AM5. I'm not 100% sure it'll adopt PCIe 5.0, but the other stuff will be there. So far, PCIe 5.0 on the desktop is just Intel playing the specsmanship game.
> Hopefully they will come at a more reasonable price
Pricing will be competitive. Expect nothing more and nothing less.
To be clear, I mean Intel's desktop support for PCIe 5.0 seems early, by at least a couple years. And, to the extent it's responsible for higher ADL motherboard prices, it could actually be hurting them. It wouldn't bother me at all, if first gen AM5 boards and Zen4 CPUs are still PCIe 4.0.
Regarding pricing, you can expect leading performance *or* discounted prices, but not both.
Not "as" huge. I didn't word it correctly. From Bulldozer to Zen1 was the biggest upgrade, from Zen1 to Zen+ it was an overclock, from Zen1/1+ to Zen2 it was a huge upgrade, from Zen2 to Zen3 it was a good upgrade. ...I was postulating if Zen4 will be another huge upgrade, or merely a decent upgrade.
I do like Zen3, but if I was confined to using Zen2 in a laptop or handheld like the Valve SD... well, I wouldn't really mind too much. Heck, even on a desktop it's still pretty good, just look at the r9-3950x compared to Intel's latest and greatest.
It is on the graphics side where I feel like AMD needs to make the biggest change. This is especially true with the problems going on with PC Gaming for the past 3 years. RDNA-1 was good, but it's RDNA-2 feels revolutionary, and AMD needs to capitalise on that properly. Apple's Graphics are still more powerful AND more efficient, again that is only macOS but indirect competition is still good for the market.
I agree with the entire Pcie lane and standard issue. I disagree with the pricing sentiment. AMD is not your friend, and they will rip you off given the chance. We need ARM, Apple, and Intel to pressure them and compete. Prices went up significantly with Zen3 cpu's to the point where it was Intel providing more bang for less buck. Modern AMD were the most competitive, from a value point, when they released Zen2 CPU, also their RDNA-1 dGPUs (since they again raised prices significantly for their RDNA-2 dGPUs).
AMD, Nvidia, Intel GPU had to support multiple API like DirectX, OpenGL, Vulkan in their GPU pipeline thus increasing architecture baggage & reducing efficiency. Unlike Apple GPU which only support Metal, Apple sacrifice 99% developer support for efficiency. What the point anyway having the most efficient & powerful GPU and yet only <1% developer in the world want to program on your GPU, especially game developer.
> had to support multiple API like DirectX, OpenGL, Vulkan in their GPU pipeline > thus increasing architecture baggage & reducing efficiency.
Nonsense. The complexity involved in supporting those APIs is almost entirely contained within the software stack. The only points where they affect the hardware are in aspects like texture formats and tessellation details, where you need native hardware support to get good performance. However, that doesn't tend to involve a lot of effort or silicon.
> I disagree with the pricing sentiment. > AMD is not your friend, and they will rip you off given the chance.
I'm not sure we *do* disagree on that. They're trying to charge as much as they can. No doubt about that.
The practical implication being that you only get the bargain prices when they're lagging on performance (and there's enough supply to meet demand). Otherwise, they're going to charge a premium.
I don't regard any company as my friend, but I think companies do differ in terms of their honesty and fairness. However, this can change over time.
Anyway, I think the pricing anomalies you're worked up over wouldn't be happening if there wasn't a supply crunch.
Sorry. I wanted to say that, even if Zen4 will be PCI-E 4.0 only (and not 5.0), as there's usually more PCI-E lanes there's less pressure to provide higher bandwidth per lane.
It looks like they are cleaning out the closet in the face of Alder Lake competition and in advance of Zen 4. I won't be the first to buy Zen 4 anyway, what with DDR5 shortages and graphics card turmoil. I upgraded to a 5600X two weeks ago for $23x and am happy with that. The 5600 might have been an acceptable cheaper choice, but the 5500 has a noticeably reduced L3 cache and only PCIe 3.0(presumably it is a mobile chip without the graphics?).
maybe their point is: for 4000 series, it's just the only one APU available. And if you can get it for $154, that's a nice price, versus ~$230 for 5600G
It wouldn't surprise me, since die-stacking capabilities in Zen 3 seem to be there mostly to enable AMD's own prototyping. Messing with the clocks without knowing what you're doing might make it spit out all kinds of fun little errors.
My B350 got 3000 series support and an upgrade to my 1500x would be nice. A 3000 series wasn't quite big enough to really be interesting outside of more cores and a new MB made 5000 series less appealing but a drop in CPU upgrade to a 5000 series would be nice. I'll have to keep my eye's open for an updated Bios and a deal.
My Asus prime pro x370 just got the bios update for 5000 CPU support (2022/03/20) witch includes the 5800x3d
making the decision to go from my 3800x to the 5800x3D significantly easier as I don't have to pull the motherboard out and replace it, does mean I am still stuck at 2933 ram speeds as that's a my motherboard issue
I have gone from 1800x to 3800x (best thing I eclver did ans the 1000/2000 CPUs suck for single threaded loads) and now looking at 5800x3d as its a significant jump from the 3800x (and an improvement over the 5800x)
Surprised there's no note on the 15% being an IPC gain - the 15% charts they seem to be showing are based on a 4GHz fixed frequency, but the 5800X3D is already clocked about 4-5% below the base 5900X, with no OC support. That seems like it can eat into those gains by an appreciable percent.
And those graphs showing the 15% average are for a dual-CCX processor, not a single CCX one like the 5800X3D. Not sure if that will help or hurt. If the remote L4 is accessable to both CCX, then you may be see some crazy single threaded code with up to 196MB of cache available to it. Halving that to 96MB that we're going to see in the 5800X3D and 15% might not be an accurate value.
The modern edition of old EE Intel cpus. They have nothing to offer (a new core with an IPC in line with Golden Lake) so they add cache, doing an unbalanced, low margin SKU. Curious to know how much they pay for the cache stacking to TSMC. A big stop for AMD, in pratics they willl stay with an old core lineup since Intel will be out with Meteor Lake on Intel 4 node. Where is TSMC 5nm ? lost in action ? never really high volume? This situation is also a disaster on Mobile segment for all 2022. How much are changed the things, AMD is again second line in X86 cpus world.
I see only this in 2022----->an old and slower core------->very few 5800x3d available----->Intel doors open, since AMD have only one (rare) SKU competitive and the rest of the line is useless. Zen 4 ? too late, there is Meteor Lake, done for kill it.
Well meteor will fall into intel fabs and they decided to reuse 14nm++++ again. Intel executive might become gone and bald (gondalf) due to execessive heat.
They will market it as a CPU that *outperforms* the i9-12900K in gaming. Worse multithreaded performance, but it's your last best upgrade for AM4 (for gaming, so no 5950X), and $50 to $150 cheaper than the 12900K (which is currently $500 at Micro Center). And the supply of the 5800X3D will probably be low.
The problem is this CPU more than any recent one needs a comprehensive review before purchase. If reviews are positive hopefully I'll be able to buy one. I won't pre-order though.
It's actually surprisingly affordable for a die-stacked CPU*, given that it's value is just a tad worse than the other CPU's - that is, if you disregard the fact that it's essentially a prototype, and you should never buy a prototype.
They're just price-matching the 12 core to guarantee that they'll get enough CPU's out of the door to give the design a little field test.
*Remember Lakefield, Intel's first 1+4 SoC using Foveros? It's price/performance ratio was astonishingly bad, but that's because today this kind of tech simply isn't cheap.
5800X3D will be the final upgrade chip for AM4, so I think the $450 price is fair. If you are building new, get the 12700K, but if you are upgrading an AM4, then 5800X3D should be the go to chip.
For gaming that is.......maybe........let's wait for the reviews. If you don't care about the maximum possible fps under non-GPU limited cases then the cheaper non X3D models are definitely a better deal.
3d cache is expensive to make! Even Zen4 does not use 3d cache because of that reason and 3d cache does not like OC. There will be some expensive halo Zen4 cpus with 3d cache like we now have one for Zen3, but these are not for customers who care bang for the buck. Those people buy cpus without 3d cache also in the future!
To be fair, the tone of Alistair's comment was more like "I think it needs to be priced at this level to be competitive", which is really a statement about performance expectations.
However, I agree with you on the posts complaining like "AMD too greedy."
Yeah, I think it people are used to Intel being the greedy evil empire, but AMD long had underdog status and was the bargain option. So, there's been some backlash as they've had to start seeing AMD in a slightly different light.
It's still the underdog, in spite of its recent wins, and has quite an uphill climb to really change that. If Zen 4 can solidly defend its recent gains and AMD can make some well-timed strategic non-x86 moves, it might actually have the potential to turn the tables on Intel. But AMD is going to have to pull some moves we haven't seen, yet. Innovations like their cache-stacking, because we know Intel is going big on chiplets/tiles and 3D stacking.
To bring us back on-topic, I guess it's worth considering that this 1st gen of 3D cache CPUs is a real learning opportunity for AMD to try something like this at production-scale. I'm sure their next generation of 3D-stacked dies will benefit from the experience in numerous ways.
I'm liking the 5700X. That might be my next CPU, since I wanted 8 cores without going above 65 W. Price is still more than I'd like to pay for it, so I'll probably still sit back and see if it gets discounted when Zen 4 launches.
Is the 15% boost estimate based on use with a particular dGPU? Seems to me all the extra memory on the dGPUs has more impact on the graphic performance than any CPU fancies.
I mean, we all know that the 5800X3D will absolutely crush the 10900k/11900k. But the 12900k? Intel has improved quite a bit, and might be worth the extra $100.
*Finally, the Ryzen 3 4100 offers just 4 Zen 2 cores, which are clocked at 3.8GHz base and 4.0GHz turbo. With one of Renoir’s two CCXes disabled, it offers just 4MB of L3 cache.*
Is it just 4MB though? The AMD slides would indicate 6MB.
The 5800X3D: “The World’s Fastest Gaming CPU?” Indeed this is a million dollar question. Will indeed Intel’s uncontested Alder Lake product dominance in gaming finally come to an end in early April? I love sheer competition which maintains to be a huge benefit for the masses. I cannot wait until all of the relevant 'tech-channels' (including those abroad) getting their hands on a complementary 5800X3D for stringent testing, observation, compatibilities and final commentary. With that I guess that by early May the JURY will be having the final verdict ready for us? To be or not to be!
Its Windows 10 late market stage V5x refresh think 286 20, 386 40, 486 133, Athlon 22xx entering Alder adaptive adoptive (thread directed) unless you're on Linux
Sticking with Windows 10 = AMD 5/3x, Comet top bin and 9900K. Moving to Windows 11 = Alder or wait for Zen 4 compatibility validation. It's that simple hard to understand why PC Tech press hasn't annunciated.
Wholesale price guide so you know when you're getting a price deal;
AMD is doing the same thing as Nvidia which is flooding the market with end run volume to drag down inventoried CPU and GPU price to clear them.
Here's how you know you're getting a good AMD V5x price and even a deal. NOW WATCH OUT FOR Vermeer 5x priced by retail as a loss leader. If V5x is priced as a loss leader you can be gouged on compliment purchases; board, memory, dGPU, PSU whatever to cost : price / cancel out the retail margin loser.
So, pick up the CPU at the loss leader price then buy everything else somewhere else.
V5X3D cost high volume OEMs $315 and volume retail no less than $420. BEST V5x wholesale to OEMs buying MILLIONS of units is $256 to $330 BUT that OEM has to buy an equal weight percent of every SKU in the whole product stack in a million's of unit's procurement to get this price.
On lower volume procurements, 5950X wholesale price to the reseller is range $412 to $532 so any price under this range is a loss leader to pull you in to compliment purchases.
5900X wholesale price to the reseller is range $284 to $366 so any price under this range is a loss leader to pull you in to compliment purchases.
5800X wholesale price to the reseller is range $232 to $299 so any price under this range is a loss leader to pull you in to compliment purchases BEWARE of inflated pricing on the extra's.
5600X wholesale price to the reseller is range $155 to $199 so any price under this range is a loss leader to pull you in to compliment purchases.
The range of prices given still requires hundreds and tens of thousands of unit's procurements from AMD so there are higher levels of wholesale pricing at the 100's of unit's and 10's of units (tray) level.
AMD sales strategy at mass market etail and retail is to replace inventory product sales out. So that retailer gets to purchase as many NEW AMD product grade SKUs as they are moving out; think first in first out and in this example sales out defines the level of new product in.
This means retail that wants access to a large NEW procurement will be more aggressive on their inventory sales out pricing, and there are likely AMD concessions and spiffs involved off-setting CPU cost : price / margin loss leaders on retailer's signing a new procurement contract and in this example with AMD.
On the Intel side Alder desktop pricing has been kept artificially low. The wholesale price is easily calculated dividing the $1K price by 2. Comet desktop is on clearance sale; HEDT SL 79xxX_ sold well six months ago primarily systems and now channel sales of XCL 99xxX_ are strong but XCLr 10x00X_ are sitting on the shelf (they're next to get flushed on clearance sale pricing).
On the Intel mobile side there's an extreme overage of Tiger Lake Quads that will be dumped at fire sale pricing. Adding Comet U and Ice U mobile there's a lot all clearance sale priced and a way to get a 3060 mobile at a reasonable price if the MXM module has not been pulled from the system.
On AMD mobile side Renoir has to go and Cezanne refresh 58/56/54U are sub $1000 compared to their Rembrandt cousins where 6980HX entered the channel this week with 3080/70ti for $6.5K and $7.5K respectively and at that price will move a lot of prior generation product and systems priced to move.
It's channel clearance sale time into summer the deals on late market and back gen will only get better before the into October Refresh. I expect Zen 4 to launch on November 1 and OEM builds to begin in q3.
I'd like to build a NAS/Homeserver (with containers and VMs) based on an open-box Asus B550 board that I could get for $45.
Whats the cheapest pick of out these if I want functional ECC and idle power efficiency? It's my understanding that APUs consume less power at idle due to being monolithic and mobile optimized, but the Pro versions are expensive.
My options then are:
R5 Pro 4650G R5 Pro 5650G R5 5600 or something cheap and used with questionable provenance like a R3 1200.
@mulot, AMD Ryzen 4650G, 5650G and 5600 do not support EEC protected memory. For ECC protected turnkey home network server on eBay search Xeon D NAS. Then you put it together, the whole NAS hardware and software solution ala PC begins with Intel E3 12xx v3/v4/v5/v6 quad or just search E3 12xx v3, v4, v5, v6 server or workstation they'll all do for a start and the mini tower servers should have it all already. Application stacks may or may not be wiped on the resale ask but a word of caution if you don't want to do that yourself the reseller will try to sell it to you whatever works . . . mb
Intel Core i3's typically support ECC memory, if you put them in the proper motherboard. Check motherboard documentation, to be sure. They typically tell you even which models they've validated. Sadly, the motherboard is where you end up spending most of the money (besides actual storage).
Xeon D is really about more heavily-threaded workloads than the typical NAS, in my opinion. Unless you're doing enterprise-grade NAS, even a dual-core, quad-thread CPU is fine for NAS.
Still can't believe the deal on the HP MicroServer Gen8; they were pushing them out the door for ~$250 in the UK. And yep, ECC with the dual-core i3 I put in - and the Celeron they came with.
ECC enabled i3, for the industrial market, really? Consumer in disguise as industrial? Check the motherboard, got it an industrial chip set. thanks. mb
Mike, check ark.intel.com You'll find plenty of i3's that are retail-boxed, desktop CPUs with advertised ECC support. Intel has been doing that for a long time. The NAS community definitely knows about it.
mode_13th, thanks for the pointer; minimally since Ivy/Haswell. Core i3 ECC support also explains why Pentium D 1508, 1517 NAS however none are currently listed on ebay but were and set on the sales shelf for some time. Thanks for your clarification on my knowledge gap. I originally brought up E3 12xx (not Comet of Coffee Refresh E) because they're highly discounted and the channel appears very negotiable on inventory volume to find a home. The 1U servers still trade regularly but the mini tower group servers and workstations sit although top bin 70/75/80 sales have picked up recently. mb
I think E3 12xx v3/5/6 unless if you want the Crystal Well v4 is the right cost optimized solution for a home built storage server or workstation featured server acting in double duty. mb
Xeon E-series is the new name for what used to be called "E3".
Once Intel bumped up i3's to 4 cores, there's less reason than ever to use a Xeon E-series for any kind of NAS. If you haven't read, their latest Alder Lake i3's now have hyperthreading enabled.
Regulators should put a stop to the selective disabling of ECC in CPUs for segmentation.
Unfortunately, what Intel likes, Intel gets.
ECC should be considered a minimum spec for CPUs. People who don't care about their data can use non-ECC RAM but it should be a BIOS option.
If Apple hadn't botched the Lisa platform... (It had ECC in 1983, for goodness' sake.) Of course, IBM is also culpable for not mandating it as part of the original PC spec — and for never mandating it after that. A lot of brainless people. The Lisa team had it right.
> ECC should be considered a minimum spec for CPUs.
I think what would be more reasonable is to have a labeling requirement indicating the likelihood of memory errors. This still gives consumers choice not to use it for something like a gaming machine, as well as creating pressures for manufacturers to make higher quality DRAM.
> It had ECC in 1983, for goodness' sake.
You don't know how 1983's DRAM compares with today's. Maybe it had quality problems that made ECC more of a practical necessity, for the amount that machine had.
AMD releasing a Zen 2 quadcore processor in 2022 is just mocking consumers at this point, i3-12100F will eat Ryzen 3 4100 for breakfast. Yes, I know AMD wants to reserve Zen 3 chips for more profitable CPUs, but it's still frustrating. As a consumers, I wish AMD still had its factories :(
Since yields are high, they should have just put the worst binned 6-core imaginable at the $99 price point to go against the i3-12100F, and forget about the quad-core. Instead they have it at $129. Maybe they can't make enough of either one long term, like the 3300X/3100 situation.
AMD will continue to use high yield 8-core chiplets for Zen 4 and probably Zen 5, and disabling 50% of the chiplet is a waste of good silicon.
@nandnandnand Back in time hexa are entering the used market in droves on top of a large quad surplus. The only new hexa and quads are fall out from sort because it's not best practice industrially or economically to disable a good octa or any core grade component. In fact it can be illegal under the antitrust laws if relied to disable a competitor by flooding the market with disabled components to fill some channel product volume space. I see you recognize that. "waste of good silicon". AMD does not disable the quads have been relied for 1 x 8 ccx + 1 x 4 ccx = 12 . . . its not always 2 x 6 ccx. if quads are available to meet 12 core count. mb
AMD is in no position to hurt Intel in that regard. There is zero possibility that the company would meet legal trouble for fusing cores to hit price points in volume. Unlike Intel, AMD doesn't have its own fabs. AMD also lacks Intel's financial resources. There is inadequate competition in the x86 CPU space but Intel and AMD are hardly on equal footing.
You're probably right about the yield, still disappointing that it will be another generation where i3 will have no competition. Also, I think 3300X and 3100 existed only on paper, rarely could find them on stock back when they launched.
@ksio89, its all sort, fallout from full line production saved up for the end whether V5x quad or Alder quad. i3 12000F represents 0.33% of full run production and 12900K/700K represent 72.98% and all of Alder i9 and i7 are 89.01% of all Alder Lake available in the WW channel. If you truly want a quad there are 25 units of 3300X available on eBay and E3 12xx v3/4/5/6 are highly available. One step up Coffee Lake 87xx hexa is returning to the market in droves. Haswell EE also was highly traded in in the last four weeks and quad, hexa, octa all available and the resellers are negotiable on price. mb
I don't know about you, but for me 25 units of a quad core processor that use a 2019 architecture is not exactly proper competition for i3-12100F. And you're forgetting that are also still a lot of i3-10100F and 10105F on the retail market. Consumers shouldn't have too choose between getting a quad core processor or a modern one, they want both like 12th gen i3.
If my Asus prime pro x370 gets a bios update I might just get the 5800x3d (even thought I still be hobbled with 2933 ram limitation of x370 motherboard) currently have the 3800x
I also didn't notice that it was 32mb+64mb of L3 cache,, so is the 64mb 3d cache more like a L4 cache or is it directly addressable same as the built in 32mb L3 cache
I hope anandtech is going to be doing some detailed work on this (be interesting details)
OK it did on the 2022/03/20 (it got 5000 CPU support including the 5800x3d) yay don't have to replace Motherboard (only downside is limited to 2933 due to my motherboard)
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
98 Comments
Back to Article
nikaldro - Tuesday, March 15, 2022 - link
Hex-core, the processor that will turn you into a frogmode_13h - Tuesday, March 15, 2022 - link
Intel used number their CPU models, until they lost some sort of trademark case and decided to use actual names. That's when they first introduced Pentium, as the successor to the i486. With Penta being the Greek prefix for 5, those who didn't know their Latin from their Greek would joke that the following generation CPU should be called the Sexium.Of course, Intel probably didn't much fancy Hexium either, since they stayed stuck on Pentium-based names until the somehow even worse-named "Core" branding. And that would later be outdone by their Xe branding for GPUs, which is the chemical symbol for the element Xenon -- far too easily confused with their Xeon branding of server/workstation CPUs. I digress...
Jasonovich - Wednesday, March 16, 2022 - link
I had a good chuckle, thanks!Oxford Guy - Wednesday, March 16, 2022 - link
Not to mention that xenon was once classified under ‘inert’ — which reminds of the apocryphal tale of the ‘no go’ Chevrolet Nova.Perhaps worse yet is that the compounds made with Xenon are all much too unstable to survive outside of very specific controlled conditions — as with Intel’s industrial chiller it used to demo its 28 core 14nm 5 GHz ‘Oops, we forgot’ CPU.
Jasonovich - Wednesday, March 16, 2022 - link
Let's hope the princess isn't a toad :)dwillmore - Tuesday, March 15, 2022 - link
"In practice, I expect the 5800X3D to live (or die) based on pricing. AMD’s 12-core Ryzen 5900X is also $449, so buyers who are willing to part with that money will be choosing between 4 more CPU cores or 64MB more L3 cache."It's just 32MB more L3 cache as it's 64MB for the 5800X and 96MB for the 5800X3D since the 5900X is a dual (6 core) die setup each with the full 32MB of L3 where the 5800X3D is a single 8 core die with 32MB of L3 and 64MB of V-Cache (or whatever they're calling it).
It really makes the 5900X look like the better choice. The 5800X3D is clocked more like a 5700X after all.
kobblestown - Tuesday, March 15, 2022 - link
In practice, I expect the 5800X3D to perform better than the 5900X on all tasks that do not peg all 12 cores to 100%. And will be noticably better on all lightly threaded tasks. So a better fit for at least 90% of people that can spare that kind of dosh.Also, it would be interesting to see how it overclocks. If it OCs at least as well as the 5900X it's an instant purchase for me (coming from a 3600X). Otherwise I might look at 5700X/5800X as better bang for buck options.
edzieba - Tuesday, March 15, 2022 - link
Signs so far point to the 5800X3D being non-overclockable (or at least not officially). Not a surprise with having the manage clocks across a bus that spans multiple dies without an explicit interconnect interface.Targon - Tuesday, March 15, 2022 - link
That whole "leak" strikes me as a bit of a joke, for a very basic reason, CPU abilities are enabled/disabled based on the BIOS. If the current AGESA versions do not allow for overclocking due to AMD wanting to keep the leaks from revealing too much, then that would make a lot more sense than not allowing overclocking at all. At CES and the interviews after it, it was strongly hinted that if you put better cooling on the 5800X3D, it will overclock up to the same speeds as the regular 5800X, but that will obviously go over the TDP rating of the CPU.AMD has actually played with the leak community over the years by releasing different samples to different places, just to see where the leaks come from. If the leaks line up with the version that went to Asus for example, then bingo, the leak comes from there, and AMD can respond properly. That is why the true leaks about Zen4 have been very few and far between.
haukionkannel - Thursday, March 17, 2022 - link
It has been confirmed. The 3d cache is the part that prevents OC in this cpu. The cache does not handle overclocking at all, so PBO and other kind of OC is prevented to protect the cache.Wrs - Tuesday, March 15, 2022 - link
That's really going to depend on the boosts and heat island effect vs. the extra cache. For lightly threaded/CPU-bound tasks, on a 5900X two cores on separate chips can boost fairly high as neither significantly heats the other. Of course, if they're the same app using the same memory space, they effectively share 32MB of cache and that's cache-bound..Ryan Smith - Tuesday, March 15, 2022 - link
Thanks! I was meaning to make a single chiplet comparison there, but fumbled it.dwillmore - Tuesday, March 15, 2022 - link
No worries, I put in 5800x then I meant 5900x, so we all make mistakes. Thanks for the article.mode_13h - Tuesday, March 15, 2022 - link
> It's just 32MB more L3 cache as it's 64MB for the 5800X and 96MB for the 5800X3D> since the 5900X is a dual (6 core) die setup each with the full 32MB of L3
Here's where mere sums don't tell the whole story. I'll just quote from the original Zen 3 review:
"The Ryzen 7 5800X ... uses a single eight-core chiplet with a total 32 MB of L3 cache. The single .. chiplet (design) has some small benefits over a dual chiplet design where some cross-CPU communication is needed, and that comes across in some of our very CPU-limited gaming benchmarks."
https://www.anandtech.com/show/16214/amd-zen-3-ryz...
So, that explains AMD's rationale. Having 64 MB split across two chiplets means having to go through the I/O die for cores to get at half of it. Whereas AMD has quoted vastly higher throughput numbers for accessing L3 stacked on the same chiplet, which I'll quote from another article:
"AMD claims that the total bandwidth of the L3 cache increases to beyond 2 TB/sec, which would technically be faster than the L1 cache on the die"
https://www.anandtech.com/show/16725/amd-demonstra...
goatfajitas - Monday, April 11, 2022 - link
12 is alot of cores for normal workloads. Unless you are doing something very specific that needs that many cores... Most people would be better off with more cache. Thus the expected gaming improvements. - "AMD is specifically pitching this 8 core processor as the ultimate chip for gaming"Tilmitt - Tuesday, March 15, 2022 - link
Hype train!WaltC - Tuesday, March 15, 2022 - link
I thought this was a pretty good write up with one small exception...taking yesterday's high-end CPU/GPUs and cycling them back to the lower/bottom end of markets is nothing new or unprecedented. That's sort of routine, really. Zen 2 will be three years old this July, and to see it cycle back to the bottom end is hardly surprising, especially when the debut for Zen 4 is only a few months away, still slated for 2022.Mike Bruzzone - Friday, March 18, 2022 - link
@WaltC, Cycling back, good point, part of the AMD sales objective building a graded by gen and SKU performance CPU price rung stack in the secondary market supports every next generation price attempting to reach a new (by category and segment) price ceiling. Intel's been doing this in performance workstation / gaming mobile for a long time. Not as much as in desktop although did with Comet deca otherwise desktop price rungs have been fairly stable.At the back in time opposing product price floor, the foundation, also creates what is known as a carpet and both AMD and Intel and Nvidia have carpets designed to keep low cost : price commodity entrants out, coming up from under the floorboards as it were.
On CPU side ARM and Apple are being blocked with a carpet where Apple attempts to come in from the top but then think Snapdragon on Windows 10 and the world market for ARM consumer mobile that are generally devices priced less than $300. The carpet is a defensive move just like (over) segmentation.
In the Xeon market AMD Epyc and ARM both came in from over the top of Intel core grades because there were no performance price rung voids in the highly segmented, at the time, Xeon 4 to 24 core market.
Production of surplus, overage, also plays a part in this game because the overage holds channels financially on their inventory accumulated capital values. More surplus means more capital values that every broker holds and hopes they can flip by finding every next home for them before the component becomes obsolete. There were three times when this became a severe issue on Intel over acceleration of primary production sticking the channel with a prematurely obsolete inventory; in 2001, 2005, 2007/8. Intel did this to take back cartel (PC dealer) price making control from the channel. Today FTC Docket 9341 guards and monitors for this economically debilitating practice destroying accumulated capital value, and Intel has been reducing surplus production since Rocket because of the unnecessary costs involved both in terms of Intel production added cost. Less unwanted product every generation means cost optimization and lowers the risk of channel financial losses.
Intel has a mandate to reconfigure the enterprise from producing for supply to producing for demand that is a demand and supply balancing. This move supports industry commerce by reducing limits placed on competitors by any dominant manufacturer / producer attempting to hold a market through over production on surplus volume.
On the Vermeer Refresh front, AMD averaged 13 M units of Ryzen performance desktop per quarter in 2021. Intel Rocket Lake was around 27 M units most of it deadweight that should never have been produced and is affecting the channel financially in terms of inventory value cost : price margin losses that were primarily made up taxing buyers clandestinely on inflated sGPU pricing paid for a lot of channel's Rocket, Comet down bin and Coffee Refresh single thread mid shelf losses so it does still happen but improving. I estimate Alder as a 30 M unit run that will be mirrored by Raptor.
Mike Bruzzone, Camp Marketing
Kangal - Tuesday, March 15, 2022 - link
I'm quite disappointed with the announcements.There's not much performance uplift or price decrease really to get hyped about.
Ever since the Zen2 release, where AMD has matched and surpassed Intel. Well they've put on the brakes. Zen3 was an upgrade but not huge, yet prices were much higher. And the chip shortage/pandemic merely secured their place for an extra two years.
Even after the debut of the Apple M1, Pro, Max, and Ultra, which put Apple ahead of the industry giants (albeit on macOS) nothing much changed at AMD.
I'm hoping that has all been on purpose. That Zen4 is a proper upgrade over Zen3 (not just a +refresh), and that RDNA-3 is also substantial. And that AMD is going to slowly discontinue their chips which have Vega Graphics, and Zen2 architecture. That this is merely TSMC and AMD getting rid of potential stock.
You know start fresh. With a new mobo platform, new pcie standard, and new DDR5 memory. And they're all Zen4, RDNA-3, and APUs. Hopefully they will come at a more reasonable price, with no nasty surprises from security or availability.
mode_13h - Wednesday, March 16, 2022 - link
> Zen3 was an upgrade but not hugeLOL, wut?
Here's what Ian and Andrei wrote, in their conclusion of the Ryzen 5000 review:
"In real world benchmarks, we saw an average +24% performance gain"
"All things considered, we’re really impressed with what AMD has achieved here."
"Zen3 gets a gold award. No question."
But you know better? M'kay. It's just a pity they're not here to show you what's what.
> Even after the debut of the Apple M1, Pro, Max, and Ultra, ... nothing much changed at AMD.
AMD reacts to its market, its customers, and its partners. I doubt the M1 has caused huge disruptions in these areas, but maybe you can provide evidence demonstrating otherwise.
Even if you're right, what would you *expect* them to do? Pretty much the only knob they can twiddle, in the short term, is pricing vs. features. That's it. They can't turn-around a whole new CPU in even the time since the first M1 hit. The pipeline for new silicon is far longer than that.
In the current world where AMD is selling basically all the silicon it can get from TSMC at the current price structure, AMD is doing just fine. It doesn't need to do anything big... yet.
> I'm hoping ... AMD is going to slowly discontinue their chips which have
> Vega Graphics, and Zen2 architecture.
https://www.anandtech.com/show/17276/amd-ryzen-9-6...
TSMC 6 nm + Zen 3 + RDNA 2.
Okay, for a refresh, but that's about it. Doesn't really deserve 6000-series branding, but then it *is* a meaningful step beyond the 5000-series stack.
> You know start fresh. With a new mobo platform, new pcie standard, and new DDR5 memory.
That's AM5. I'm not 100% sure it'll adopt PCIe 5.0, but the other stuff will be there. So far, PCIe 5.0 on the desktop is just Intel playing the specsmanship game.
> Hopefully they will come at a more reasonable price
Pricing will be competitive. Expect nothing more and nothing less.
mode_13h - Wednesday, March 16, 2022 - link
To be clear, I mean Intel's desktop support for PCIe 5.0 seems early, by at least a couple years. And, to the extent it's responsible for higher ADL motherboard prices, it could actually be hurting them. It wouldn't bother me at all, if first gen AM5 boards and Zen4 CPUs are still PCIe 4.0.Regarding pricing, you can expect leading performance *or* discounted prices, but not both.
Calin - Wednesday, March 16, 2022 - link
"Zen4 CPUs are still PCIe 4.0."Also, AMD usually has more PCIE lanes than Intel, so there's less pressure on increasing the number of PCIE lanes.
mode_13h - Wednesday, March 16, 2022 - link
Eh, please watch the quoting. I'm not saying they *will* be PCIe 4.0, just that I think it'll make no practical difference for a couple more years.Kangal - Wednesday, March 16, 2022 - link
Not "as" huge. I didn't word it correctly.From Bulldozer to Zen1 was the biggest upgrade, from Zen1 to Zen+ it was an overclock, from Zen1/1+ to Zen2 it was a huge upgrade, from Zen2 to Zen3 it was a good upgrade.
...I was postulating if Zen4 will be another huge upgrade, or merely a decent upgrade.
I do like Zen3, but if I was confined to using Zen2 in a laptop or handheld like the Valve SD... well, I wouldn't really mind too much. Heck, even on a desktop it's still pretty good, just look at the r9-3950x compared to Intel's latest and greatest.
It is on the graphics side where I feel like AMD needs to make the biggest change. This is especially true with the problems going on with PC Gaming for the past 3 years. RDNA-1 was good, but it's RDNA-2 feels revolutionary, and AMD needs to capitalise on that properly. Apple's Graphics are still more powerful AND more efficient, again that is only macOS but indirect competition is still good for the market.
I agree with the entire Pcie lane and standard issue.
I disagree with the pricing sentiment.
AMD is not your friend, and they will rip you off given the chance. We need ARM, Apple, and Intel to pressure them and compete. Prices went up significantly with Zen3 cpu's to the point where it was Intel providing more bang for less buck. Modern AMD were the most competitive, from a value point, when they released Zen2 CPU, also their RDNA-1 dGPUs (since they again raised prices significantly for their RDNA-2 dGPUs).
wira6444 - Wednesday, March 16, 2022 - link
AMD, Nvidia, Intel GPU had to support multiple API like DirectX, OpenGL, Vulkan in their GPU pipeline thus increasing architecture baggage & reducing efficiency. Unlike Apple GPU which only support Metal, Apple sacrifice 99% developer support for efficiency. What the point anyway having the most efficient & powerful GPU and yet only <1% developer in the world want to program on your GPU, especially game developer.mode_13h - Friday, March 18, 2022 - link
> had to support multiple API like DirectX, OpenGL, Vulkan in their GPU pipeline> thus increasing architecture baggage & reducing efficiency.
Nonsense. The complexity involved in supporting those APIs is almost entirely contained within the software stack. The only points where they affect the hardware are in aspects like texture formats and tessellation details, where you need native hardware support to get good performance. However, that doesn't tend to involve a lot of effort or silicon.
mode_13h - Friday, March 18, 2022 - link
> I disagree with the pricing sentiment.> AMD is not your friend, and they will rip you off given the chance.
I'm not sure we *do* disagree on that. They're trying to charge as much as they can. No doubt about that.
The practical implication being that you only get the bargain prices when they're lagging on performance (and there's enough supply to meet demand). Otherwise, they're going to charge a premium.
I don't regard any company as my friend, but I think companies do differ in terms of their honesty and fairness. However, this can change over time.
Anyway, I think the pricing anomalies you're worked up over wouldn't be happening if there wasn't a supply crunch.
Calin - Thursday, March 17, 2022 - link
Sorry.I wanted to say that, even if Zen4 will be PCI-E 4.0 only (and not 5.0), as there's usually more PCI-E lanes there's less pressure to provide higher bandwidth per lane.
yetanotherhuman - Thursday, March 17, 2022 - link
Zen 3 was a huge upgrade over Zen 2, not sure where you were when the reviews hitshabby - Tuesday, March 15, 2022 - link
That 5500 looks enticing, too bad they released it so late.CrystalCowboy - Tuesday, March 15, 2022 - link
It looks like they are cleaning out the closet in the face of Alder Lake competition and in advance of Zen 4. I won't be the first to buy Zen 4 anyway, what with DDR5 shortages and graphics card turmoil.I upgraded to a 5600X two weeks ago for $23x and am happy with that. The 5600 might have been an acceptable cheaper choice, but the 5500 has a noticeably reduced L3 cache and only PCIe 3.0(presumably it is a mobile chip without the graphics?).
Spunjji - Friday, March 18, 2022 - link
"presumably it is a mobile chip without the graphics"Yup!
nandnandnand - Tuesday, March 15, 2022 - link
"Still, if you need a retail Ryzen chip with integrated graphics, then the 4600G will be your one and only option for an AMD APU with graphics."Well, no. You could also pick up a 5700G or 5600G.
t.s - Tuesday, March 15, 2022 - link
maybe their point is: for 4000 series, it's just the only one APU available. And if you can get it for $154, that's a nice price, versus ~$230 for 5600GCrystalCowboy - Tuesday, March 15, 2022 - link
I don't know why you have the 5600G listed as "OEM." It is available and in stock at my favorite retailer for $219.Ryan Smith - Tuesday, March 15, 2022 - link
Thanks! Fixed.leo_sk - Tuesday, March 15, 2022 - link
Is it true that amd is disabling overclocking for 5600X3D through BIOS? Why are they doing so?meacupla - Tuesday, March 15, 2022 - link
speculation is the 3D cache doesn't work well with overclockingWereweeb - Tuesday, March 15, 2022 - link
It wouldn't surprise me, since die-stacking capabilities in Zen 3 seem to be there mostly to enable AMD's own prototyping. Messing with the clocks without knowing what you're doing might make it spit out all kinds of fun little errors.Targon - Tuesday, March 15, 2022 - link
The leakers don't have access to the final AGESA is why they think the chip won't overclock.kpb321 - Tuesday, March 15, 2022 - link
My B350 got 3000 series support and an upgrade to my 1500x would be nice. A 3000 series wasn't quite big enough to really be interesting outside of more cores and a new MB made 5000 series less appealing but a drop in CPU upgrade to a 5000 series would be nice. I'll have to keep my eye's open for an updated Bios and a deal.leexgx - Wednesday, April 20, 2022 - link
My Asus prime pro x370 just got the bios update for 5000 CPU support (2022/03/20) witch includes the 5800x3dmaking the decision to go from my 3800x to the 5800x3D significantly easier as I don't have to pull the motherboard out and replace it, does mean I am still stuck at 2933 ram speeds as that's a my motherboard issue
I have gone from 1800x to 3800x (best thing I eclver did ans the 1000/2000 CPUs suck for single threaded loads) and now looking at 5800x3d as its a significant jump from the 3800x (and an improvement over the 5800x)
Drumsticks - Tuesday, March 15, 2022 - link
Surprised there's no note on the 15% being an IPC gain - the 15% charts they seem to be showing are based on a 4GHz fixed frequency, but the 5800X3D is already clocked about 4-5% below the base 5900X, with no OC support. That seems like it can eat into those gains by an appreciable percent.dwillmore - Tuesday, March 15, 2022 - link
And those graphs showing the 15% average are for a dual-CCX processor, not a single CCX one like the 5800X3D. Not sure if that will help or hurt. If the remote L4 is accessable to both CCX, then you may be see some crazy single threaded code with up to 196MB of cache available to it. Halving that to 96MB that we're going to see in the 5800X3D and 15% might not be an accurate value.Spunjji - Friday, March 18, 2022 - link
AMD are currently claiming that the 5800X3D is "up to" 15% faster in their tested games than the 5900X, stock to stock.In practice that's going to vary a lot depending on the game, resolution and GPU capability.
Gondalf - Tuesday, March 15, 2022 - link
The modern edition of old EE Intel cpus. They have nothing to offer (a new core with an IPC in line with Golden Lake) so they add cache, doing an unbalanced, low margin SKU. Curious to know how much they pay for the cache stacking to TSMC.A big stop for AMD, in pratics they willl stay with an old core lineup since Intel will be out with Meteor Lake on Intel 4 node. Where is TSMC 5nm ? lost in action ? never really high volume?
This situation is also a disaster on Mobile segment for all 2022.
How much are changed the things, AMD is again second line in X86 cpus world.
Thunder 57 - Tuesday, March 15, 2022 - link
I see you've come to troll as usual and spew your AMD hate. I can't wait to see what you come up with when Zen 4 launches.Gondalf - Wednesday, March 16, 2022 - link
I see only this in 2022----->an old and slower core------->very few 5800x3d available----->Intel doors open, since AMD have only one (rare) SKU competitive and the rest of the line is useless.Zen 4 ? too late, there is Meteor Lake, done for kill it.
Qasar - Wednesday, March 16, 2022 - link
yea ok there gondaft. mr pro intel, i hate amd to no end.wira6444 - Wednesday, March 16, 2022 - link
Well meteor will fall into intel fabs and they decided to reuse 14nm++++ again. Intel executive might become gone and bald (gondalf) due to execessive heat.yankeeDDL - Wednesday, March 16, 2022 - link
Let him be. He clearly has issues with fact and reality.Spunjji - Friday, March 18, 2022 - link
Hey everybody look, it's Gandaft! Like Radagast the brown but instead of bird poop on his head, it's his own faeces.Alistair - Tuesday, March 15, 2022 - link
5800X3D should be $350, not $450. I don't think this is going to look good against the 5900X and i7-12700K. We'll see. Exciting though!nandnandnand - Tuesday, March 15, 2022 - link
They will market it as a CPU that *outperforms* the i9-12900K in gaming. Worse multithreaded performance, but it's your last best upgrade for AM4 (for gaming, so no 5950X), and $50 to $150 cheaper than the 12900K (which is currently $500 at Micro Center). And the supply of the 5800X3D will probably be low.Alistair - Wednesday, March 16, 2022 - link
The problem is this CPU more than any recent one needs a comprehensive review before purchase. If reviews are positive hopefully I'll be able to buy one. I won't pre-order though.nandnandnand - Wednesday, March 16, 2022 - link
It's a highly anticipated review, that's for sure. It will be fascinating to see where it improves and falls short.Alistair - Thursday, March 17, 2022 - link
right! it is unprecedented, I really can't predict the review's results beforehandWereweeb - Tuesday, March 15, 2022 - link
It's actually surprisingly affordable for a die-stacked CPU*, given that it's value is just a tad worse than the other CPU's - that is, if you disregard the fact that it's essentially a prototype, and you should never buy a prototype.They're just price-matching the 12 core to guarantee that they'll get enough CPU's out of the door to give the design a little field test.
*Remember Lakefield, Intel's first 1+4 SoC using Foveros? It's price/performance ratio was astonishingly bad, but that's because today this kind of tech simply isn't cheap.
Oxford Guy - Wednesday, March 16, 2022 - link
Considering how much it was delayed it may have been very well-tested. Some first-generation products equal or exceed later versions in quality.meacupla - Tuesday, March 15, 2022 - link
5800X3D will be the final upgrade chip for AM4, so I think the $450 price is fair.If you are building new, get the 12700K, but if you are upgrading an AM4, then 5800X3D should be the go to chip.
techjunkie123 - Tuesday, March 15, 2022 - link
For gaming that is.......maybe........let's wait for the reviews. If you don't care about the maximum possible fps under non-GPU limited cases then the cheaper non X3D models are definitely a better deal.haukionkannel - Thursday, March 17, 2022 - link
Yes. buy 5900x or 5950x if you need serious cpu power, but for gamers... 5800X3D may be the end game solution for AM4 socket!pun intended ;)
haukionkannel - Thursday, March 17, 2022 - link
3d cache is expensive to make!Even Zen4 does not use 3d cache because of that reason and 3d cache does not like OC.
There will be some expensive halo Zen4 cpus with 3d cache like we now have one for Zen3, but these are not for customers who care bang for the buck. Those people buy cpus without 3d cache also in the future!
Spunjji - Friday, March 18, 2022 - link
Every time I see one of these "AMD should not make any profit on their high-performing products" posts, I cringe.mode_13h - Friday, March 18, 2022 - link
To be fair, the tone of Alistair's comment was more like "I think it needs to be priced at this level to be competitive", which is really a statement about performance expectations.However, I agree with you on the posts complaining like "AMD too greedy."
Qasar - Friday, March 18, 2022 - link
it amazes me how quite a few seem to have forgotten just how greedy intel has been in the past.mode_13h - Saturday, March 19, 2022 - link
Yeah, I think it people are used to Intel being the greedy evil empire, but AMD long had underdog status and was the bargain option. So, there's been some backlash as they've had to start seeing AMD in a slightly different light.It's still the underdog, in spite of its recent wins, and has quite an uphill climb to really change that. If Zen 4 can solidly defend its recent gains and AMD can make some well-timed strategic non-x86 moves, it might actually have the potential to turn the tables on Intel. But AMD is going to have to pull some moves we haven't seen, yet. Innovations like their cache-stacking, because we know Intel is going big on chiplets/tiles and 3D stacking.
To bring us back on-topic, I guess it's worth considering that this 1st gen of 3D cache CPUs is a real learning opportunity for AMD to try something like this at production-scale. I'm sure their next generation of 3D-stacked dies will benefit from the experience in numerous ways.
mode_13h - Tuesday, March 15, 2022 - link
I'm liking the 5700X. That might be my next CPU, since I wanted 8 cores without going above 65 W. Price is still more than I'd like to pay for it, so I'll probably still sit back and see if it gets discounted when Zen 4 launches.JayNor - Wednesday, March 16, 2022 - link
Is the 15% boost estimate based on use with a particular dGPU? Seems to me all the extra memory on the dGPUs has more impact on the graphic performance than any CPU fancies.mode_13h - Wednesday, March 16, 2022 - link
In a month, I'm sure we'll have all the details about the precise performance impact.I think Intel doesn't have too much to worry about, but we'll see.
Alistair - Thursday, March 17, 2022 - link
I mean, we all know that the 5800X3D will absolutely crush the 10900k/11900k. But the 12900k? Intel has improved quite a bit, and might be worth the extra $100.Mr Perfect - Wednesday, March 16, 2022 - link
*Finally, the Ryzen 3 4100 offers just 4 Zen 2 cores, which are clocked at 3.8GHz base and 4.0GHz turbo. With one of Renoir’s two CCXes disabled, it offers just 4MB of L3 cache.*Is it just 4MB though? The AMD slides would indicate 6MB.
jjem002 - Thursday, March 17, 2022 - link
AMD includes both L2 and L3 cache in their marketing slides, so its the 4MB L3, and the 512KB L2 per coreTom Sunday - Thursday, March 17, 2022 - link
The 5800X3D: “The World’s Fastest Gaming CPU?” Indeed this is a million dollar question. Will indeed Intel’s uncontested Alder Lake product dominance in gaming finally come to an end in early April? I love sheer competition which maintains to be a huge benefit for the masses. I cannot wait until all of the relevant 'tech-channels' (including those abroad) getting their hands on a complementary 5800X3D for stringent testing, observation, compatibilities and final commentary. With that I guess that by early May the JURY will be having the final verdict ready for us? To be or not to be!nandnandnand - Saturday, March 19, 2022 - link
Will 5800X3D become the ultimate Dwarf Fortress CPU?Mike Bruzzone - Friday, March 18, 2022 - link
Camp Marketing into 2022 buying tips;Its Windows 10 late market stage V5x refresh think 286 20, 386 40, 486 133, Athlon 22xx entering Alder adaptive adoptive (thread directed) unless you're on Linux
Sticking with Windows 10 = AMD 5/3x, Comet top bin and 9900K.
Moving to Windows 11 = Alder or wait for Zen 4 compatibility validation.
It's that simple hard to understand why PC Tech press hasn't annunciated.
Wholesale price guide so you know when you're getting a price deal;
AMD is doing the same thing as Nvidia which is flooding the market with end run volume to drag down inventoried CPU and GPU price to clear them.
Here's how you know you're getting a good AMD V5x price and even a deal. NOW WATCH OUT FOR Vermeer 5x priced by retail as a loss leader. If V5x is priced as a loss leader you can be gouged on compliment purchases; board, memory, dGPU, PSU whatever to cost : price / cancel out the retail margin loser.
So, pick up the CPU at the loss leader price then buy everything else somewhere else.
V5X3D cost high volume OEMs $315 and volume retail no less than $420. BEST V5x wholesale to OEMs buying MILLIONS of units is $256 to $330 BUT that OEM has to buy an equal weight percent of every SKU in the whole product stack in a million's of unit's procurement to get this price.
On lower volume procurements, 5950X wholesale price to the reseller is range $412 to $532 so any price under this range is a loss leader to pull you in to compliment purchases.
5900X wholesale price to the reseller is range $284 to $366 so any price under this range is a loss leader to pull you in to compliment purchases.
5800X wholesale price to the reseller is range $232 to $299 so any price under this range is a loss leader to pull you in to compliment purchases BEWARE of inflated pricing on the extra's.
5600X wholesale price to the reseller is range $155 to $199 so any price under this range is a loss leader to pull you in to compliment purchases.
The range of prices given still requires hundreds and tens of thousands of unit's procurements from AMD so there are higher levels of wholesale pricing at the 100's of unit's and 10's of units (tray) level.
AMD sales strategy at mass market etail and retail is to replace inventory product sales out. So that retailer gets to purchase as many NEW AMD product grade SKUs as they are moving out; think first in first out and in this example sales out defines the level of new product in.
This means retail that wants access to a large NEW procurement will be more aggressive on their inventory sales out pricing, and there are likely AMD concessions and spiffs involved off-setting CPU cost : price / margin loss leaders on retailer's signing a new procurement contract and in this example with AMD.
On the Intel side Alder desktop pricing has been kept artificially low. The wholesale price is easily calculated dividing the $1K price by 2. Comet desktop is on clearance sale; HEDT SL 79xxX_ sold well six months ago primarily systems and now channel sales of XCL 99xxX_ are strong but XCLr 10x00X_ are sitting on the shelf (they're next to get flushed on clearance sale pricing).
On the Intel mobile side there's an extreme overage of Tiger Lake Quads that will be dumped at fire sale pricing. Adding Comet U and Ice U mobile there's a lot all clearance sale priced and a way to get a 3060 mobile at a reasonable price if the MXM module has not been pulled from the system.
On AMD mobile side Renoir has to go and Cezanne refresh 58/56/54U are sub $1000 compared to their Rembrandt cousins where 6980HX entered the channel this week with 3080/70ti for $6.5K and $7.5K respectively and at that price will move a lot of prior generation product and systems priced to move.
It's channel clearance sale time into summer the deals on late market and back gen will only get better before the into October Refresh. I expect Zen 4 to launch on November 1 and OEM builds to begin in q3.
Mike Bruzzone, Camp Marketing
mulot - Friday, March 18, 2022 - link
I'd like to build a NAS/Homeserver (with containers and VMs) based on an open-box Asus B550 board that I could get for $45.Whats the cheapest pick of out these if I want functional ECC and idle power efficiency? It's my understanding that APUs consume less power at idle due to being monolithic and mobile optimized, but the Pro versions are expensive.
My options then are:
R5 Pro 4650G
R5 Pro 5650G
R5 5600
or something cheap and used with questionable provenance like a R3 1200.
Mike Bruzzone - Sunday, March 20, 2022 - link
@mulot, AMD Ryzen 4650G, 5650G and 5600 do not support EEC protected memory. For ECC protected turnkey home network server on eBay search Xeon D NAS. Then you put it together, the whole NAS hardware and software solution ala PC begins with Intel E3 12xx v3/v4/v5/v6 quad or just search E3 12xx v3, v4, v5, v6 server or workstation they'll all do for a start and the mini tower servers should have it all already. Application stacks may or may not be wiped on the resale ask but a word of caution if you don't want to do that yourself the reseller will try to sell it to you whatever works . . . mbmode_13h - Sunday, March 20, 2022 - link
Intel Core i3's typically support ECC memory, if you put them in the proper motherboard. Check motherboard documentation, to be sure. They typically tell you even which models they've validated. Sadly, the motherboard is where you end up spending most of the money (besides actual storage).Xeon D is really about more heavily-threaded workloads than the typical NAS, in my opinion. Unless you're doing enterprise-grade NAS, even a dual-core, quad-thread CPU is fine for NAS.
GreenReaper - Sunday, March 20, 2022 - link
Still can't believe the deal on the HP MicroServer Gen8; they were pushing them out the door for ~$250 in the UK. And yep, ECC with the dual-core i3 I put in - and the Celeron they came with.Mike Bruzzone - Sunday, March 20, 2022 - link
ECC enabled i3, for the industrial market, really? Consumer in disguise as industrial? Check the motherboard, got it an industrial chip set. thanks. mbmode_13h - Monday, March 21, 2022 - link
Mike, check ark.intel.com You'll find plenty of i3's that are retail-boxed, desktop CPUs with advertised ECC support. Intel has been doing that for a long time. The NAS community definitely knows about it.https://ark.intel.com/content/www/us/en/ark/search...
Mine is a i3-4370 in a homebuilt machine.
Mike Bruzzone - Monday, March 21, 2022 - link
mode_13th, thanks for the pointer; minimally since Ivy/Haswell. Core i3 ECC support also explains why Pentium D 1508, 1517 NAS however none are currently listed on ebay but were and set on the sales shelf for some time. Thanks for your clarification on my knowledge gap. I originally brought up E3 12xx (not Comet of Coffee Refresh E) because they're highly discounted and the channel appears very negotiable on inventory volume to find a home. The 1U servers still trade regularly but the mini tower group servers and workstations sit although top bin 70/75/80 sales have picked up recently. mbMike Bruzzone - Sunday, March 20, 2022 - link
I think E3 12xx v3/5/6 unless if you want the Crystal Well v4 is the right cost optimized solution for a home built storage server or workstation featured server acting in double duty. mbmode_13h - Monday, March 21, 2022 - link
Xeon E-series is the new name for what used to be called "E3".Once Intel bumped up i3's to 4 cores, there's less reason than ever to use a Xeon E-series for any kind of NAS. If you haven't read, their latest Alder Lake i3's now have hyperthreading enabled.
https://ark.intel.com/content/www/us/en/ark/produc...
And although that page doesn't mention anything about ECC, this does:
https://www.anandtech.com/show/17308/the-intel-w68...
Oxford Guy - Monday, March 21, 2022 - link
Regulators should put a stop to the selective disabling of ECC in CPUs for segmentation.Unfortunately, what Intel likes, Intel gets.
ECC should be considered a minimum spec for CPUs. People who don't care about their data can use non-ECC RAM but it should be a BIOS option.
If Apple hadn't botched the Lisa platform... (It had ECC in 1983, for goodness' sake.) Of course, IBM is also culpable for not mandating it as part of the original PC spec — and for never mandating it after that. A lot of brainless people. The Lisa team had it right.
mode_13h - Monday, March 21, 2022 - link
> ECC should be considered a minimum spec for CPUs.I think what would be more reasonable is to have a labeling requirement indicating the likelihood of memory errors. This still gives consumers choice not to use it for something like a gaming machine, as well as creating pressures for manufacturers to make higher quality DRAM.
> It had ECC in 1983, for goodness' sake.
You don't know how 1983's DRAM compares with today's. Maybe it had quality problems that made ECC more of a practical necessity, for the amount that machine had.
ksio89 - Saturday, March 19, 2022 - link
AMD releasing a Zen 2 quadcore processor in 2022 is just mocking consumers at this point, i3-12100F will eat Ryzen 3 4100 for breakfast. Yes, I know AMD wants to reserve Zen 3 chips for more profitable CPUs, but it's still frustrating. As a consumers, I wish AMD still had its factories :(nandnandnand - Saturday, March 19, 2022 - link
Since yields are high, they should have just put the worst binned 6-core imaginable at the $99 price point to go against the i3-12100F, and forget about the quad-core. Instead they have it at $129. Maybe they can't make enough of either one long term, like the 3300X/3100 situation.AMD will continue to use high yield 8-core chiplets for Zen 4 and probably Zen 5, and disabling 50% of the chiplet is a waste of good silicon.
Mike Bruzzone - Sunday, March 20, 2022 - link
@nandnandnand Back in time hexa are entering the used market in droves on top of a large quad surplus. The only new hexa and quads are fall out from sort because it's not best practice industrially or economically to disable a good octa or any core grade component. In fact it can be illegal under the antitrust laws if relied to disable a competitor by flooding the market with disabled components to fill some channel product volume space. I see you recognize that. "waste of good silicon". AMD does not disable the quads have been relied for 1 x 8 ccx + 1 x 4 ccx = 12 . . . its not always 2 x 6 ccx. if quads are available to meet 12 core count. mbOxford Guy - Monday, March 21, 2022 - link
AMD is in no position to hurt Intel in that regard. There is zero possibility that the company would meet legal trouble for fusing cores to hit price points in volume. Unlike Intel, AMD doesn't have its own fabs. AMD also lacks Intel's financial resources. There is inadequate competition in the x86 CPU space but Intel and AMD are hardly on equal footing.ksio89 - Tuesday, March 22, 2022 - link
You're probably right about the yield, still disappointing that it will be another generation where i3 will have no competition. Also, I think 3300X and 3100 existed only on paper, rarely could find them on stock back when they launched.Mike Bruzzone - Sunday, March 20, 2022 - link
@ksio89, its all sort, fallout from full line production saved up for the end whether V5x quad or Alder quad. i3 12000F represents 0.33% of full run production and 12900K/700K represent 72.98% and all of Alder i9 and i7 are 89.01% of all Alder Lake available in the WW channel. If you truly want a quad there are 25 units of 3300X available on eBay and E3 12xx v3/4/5/6 are highly available. One step up Coffee Lake 87xx hexa is returning to the market in droves. Haswell EE also was highly traded in in the last four weeks and quad, hexa, octa all available and the resellers are negotiable on price. mbksio89 - Tuesday, March 22, 2022 - link
I don't know about you, but for me 25 units of a quad core processor that use a 2019 architecture is not exactly proper competition for i3-12100F. And you're forgetting that are also still a lot of i3-10100F and 10105F on the retail market. Consumers shouldn't have too choose between getting a quad core processor or a modern one, they want both like 12th gen i3.And what do you mean by "V5X"?
leexgx - Wednesday, April 20, 2022 - link
If my Asus prime pro x370 gets a bios update I might just get the 5800x3d (even thought I still be hobbled with 2933 ram limitation of x370 motherboard) currently have the 3800xI also didn't notice that it was 32mb+64mb of L3 cache,, so is the 64mb 3d cache more like a L4 cache or is it directly addressable same as the built in 32mb L3 cache
I hope anandtech is going to be doing some detailed work on this (be interesting details)
leexgx - Wednesday, April 20, 2022 - link
OK it did on the 2022/03/20 (it got 5000 CPU support including the 5800x3d) yay don't have to replace Motherboard (only downside is limited to 2933 due to my motherboard)leexgx - Wednesday, April 20, 2022 - link
The 96mb L3 cache is all one blobhaplo602 - Friday, April 22, 2022 - link
so where is the review ?