I'm curious about the CSGO results (tied with 5900x iirc), at launch of 5000 series we saw ~%45 uplift compared to 3000. Does this suggest the cache pressure was mostly relieved in the jump from 16MB to 32MB?
It was already understood that not all games would benefit from a tripling of cache. CS:GO is a very old game that is easy to run. The drop in clock speed could explain the decrease.
“The cache does add a few watts to the power both in terms of idle and load. Rather than bin a stricter chiplet, the decision was made to reduce the frequency a little.”
Come on AMD. Intel shamelessly asks for 190 watt on their i7 12700K. Why not do the same for what is already a niche product? Let loose the dogs of war! I wanna know what that 5800X3D can do at 4.5-4.7 GHz all core clock speed!
By you selecting modes of operation that exceed the design power. AMD allow people to do that, but they don't (yet?) ship chips that do it out of the box.
As someone with a 5800X, I know it has issues dissipating more than 150-160W with a high-end air cooler (the D15) while staying within 90C Tdie. That said, many workloads already run at 4.5 all-core on a mediocre bin and that D15 - but forget about Linpack or optimized AVX2. The question is how much harder it'll be to clock the cores on the same instructions with the additional silicon sliver on top.
Eh, no thank you. The draw of Ryzen for gaming is partly because it can do so well without a ludicrous power budget. Remember how Intel's been ridiculed for it's high TDPs the last few years? Let's not loose the one sane vendor here.
just as amd was be fore zen, but the difference i have been seeing, while most reviews show intel uses more power, there are alot of people that are fine with it, and keep saying its fine, and intel and amd use the same amount of power.
Not necessarily. If CSGO is one of the titles where performance increases by 1.4x vs the 5900X, then there is likely still significant cache pressure for that application. It is always possible to write a program that will have significant cache pressure for any given amount of cache. So, you cannot look at the average application performance improvement/bottleneck to make conclusions about a specific applications performance improvement/bottleneck.
However, it looks like in general the cache pressure has been mostly eliminated. A 3x cache increase for an average of 1.15x performance, while still a good increase, indicates that on average we are very likely at diminishing returns or close to it.
I'm imagining that there should be compilers that aren't cache size agnostic. Would one really need to do it at the design/code level? or would there be optimizing compilers?
" it's really not that impressive " like intel was before zen ? they werent that inpressive either. only because intel was getting beaten by amd, did intel finally wake up and started doing something again. some i know, are still not impressed with intel, and have no interest in their cpus, and will be going with amd when they upgrade in the next few months.
" increase the cost." like intel kept doing before zen ?
Why are you arguing, when based on your comments, you agree with what he said? Just because Intel did something similar in the past does not mean AMD should not be criticized for the same practices. They put something out to try and take headlines away from Alder Lake because they do not have anything of note until Zen 4 releases later this year.
i wasnt agreeing with him, quite the opposite. lemurbutton was saying he isnt impressed with v cache, and i was countering that with how unimpressive intel was before zen was released. the fact is, most people are complaining about amd raising its cpu prices, while i would bet none of them, complained about intel doing the same. i dont recall ANY posts complaining about intels prices at all before zen, and intel would raise their prices by hundreds, and raises theirs by 50 bucks, and most jumped off the deep end. i can still remember seeing intel dropping their prices across the board when zen came out, and by at least 1k at the top end.
" hey put something out to try and take headlines away from Alder Lake" um you DO know that CES is for the most part, a few days away, regardless if intel said anything about alder lake, amd would of still announced something about their products that they have in the pipe to be released this year.
The 2600K-7700K 4c/8t processors all carried roughly the same retail price, 8700K added 2 cores and ~$50, 9900K was a bigger jump of around $120 with 2 more cores and then the 10900K close to the same with the 11900K being the first confusingly priced one and that covers 10 years so I'm not sure where your "Intel raising prices by hundreds" comes from. HEDT was the only place with stupid prices (though they were always increasing core count here too), and that was always true for Intel until AMD brought the competition before flying past.
This part is AMD's halo product for gaming, and it doesn't appear to be all that impressive being limited to a 5800X running lower clocks. Perhaps AMD will release others, but it really seems like they picked the highest single ccx core count and ran with it. We likely won't ever know why the decision was made for sure, but it could be money, power consumption, heat or something else I haven't thought of entirely.
This will cost something like $550... Most people would not be buing 5950X3D for $1100 anyway. So they looked what 5900x and intel 12900k cost and did see that there is a slot in between where this product fits! 5850X3D would not be faster in the games, and when extra cache is useful mainly in the games, this was the only sensible place where they can use expensive 3D cache while maintain sensible price point.
thestryker " so I'm not sure where your "Intel raising prices by hundreds" comes from. " your are probably thinking in US $, every where else, the increases gen over gen where much higher.
" you might be paying more they're not charging more. " yes they are. intel kept raising prices in the US, so every where else prices also had to go up, in some places,by quite a lot, and all we got out of it, was 10% or less performance gains gen on gen, amd does 20%+ gen on gen, raise theirs by 50 ( here i think it was closer 100) and quite a few complained and cried.
Intel *never* charged more without increasing the core count until the 11th gen series. That's the reality since the desktop/HEDT split with Sandy Bridge. AMD raised prices due to their market position without changing core counts which is something that again never happened from Intel until 11th gen. AMD was smart to do it, but that doesn't make it look less bad to consumers.
thestryker, maybe where YOU are, but that wasnt the case every where. thats the reality here. gen on gen, intels cpu prices kept going up, increased core counts or not.
Intel created, stacked performance mobile 'gaming' and 'workstation' NEW price rungs every generation since Haswell raising the price ceiling. AMD enabled and allowed the channel to set AMD MSRP to offset channels AMD inventory losses on all products before Zen 1x made obsolete by Zen. The practice has become inflationary. mb
Intel component and mobile U stuck on the shelf is the primary reason for regional pricing problems. Rocket deadweight, Comet S down bin, Coffee Refresh single thread, Tiger, Ice, Comet U quad laptops are on fire sale. Inflated dGPU card price is paying for channel's Intel inventory financial write down and/or loss. Worst is retail CPU + GPU and board kit deal where inflated GPU card price offsets the CPU loss leader and the board add its own mark up. There is no x86 or GPU card shortage data can validate it's called capital creation on channel Intel surplus inventory financial risk management. mb
Package area for pulling off the heat and material composition is not TR and Epyc Fujitsu composite? I think so. 59003D was always a lid off lab demo? I think so. mb
"They put something out to try and take headlines away from Alder Lake because they do not have anything of note until Zen 4 releases later this year." You might have missed how this was previewed months back? AMD and TSMC don't go through months and months of design and validation to chuck out something to mess with publicity for one of Intel's products. The technology will mostly benefit AMD's server division, but they promised something for the gamers and here it is, for better or worse.
Rembrandt appears AMD 2020'mainstay 'consumer' and mass market gaming product between the two suspension bridge towers of 3D and Zen 4. Rembrandt is the deck. Then imagine an arch from tower 1 abutment reaching height at mid span and then back down to tower two abutment is commercial gross margin lift; TR 5K, Epyc where Genoa is already shiping on AMD loss of 7 nm cost : price / margin advantage vis a vis Intel on TSMC foundry market up and commercial GPU which AMD is now focusing getting out from in between Intel and Nvidia in the consumer market. mb
To be completely fair, given timelines of CPU design, Alder Lake design began almost immediately after Zen 1 was released. Every CPU between Zen1 and now that Intel released felt like a half-assed retort because all they could do until Alder Lake was finished was convert their HEDT line to their consumer socket, cut prices, and crank the wattage to extreme levels.
Began after Zen 1???? You're crazy. The Alder Lake plans started way before that. Intel has been iterating their Atom cores for this moment literally since 2013. The Apple A7 was Intel's wakeup call, architecturally. And yes, it takes that long to turn around a giant in an industry where design to tapeout is usually ~5 years, and Atom was way beyond just one product cycle from reaching the necessary performance characteristics.
Up until the A7, Intel assumed it could meet Apple on Apple's newfound turf. With the A7, Intel canned their smartphone product plans almost instantly and pivoted Atom toward its current trajectory. The only reason Alder Lake took quite this long is because of Intel's foundry slips.
"The Alder Lake plans started way before that" Is there any evidence for that? I could believe they had some blue-sky ideas floating around about it, but given your quoted 5 years from design to tape-out the idea that Alder Lake is a direct response to Zen makes perfect sense.
"The trade-off is more frequency or more cache" Unless you're already at the limit for frequency, then you'd just get more cache (and a little more power draw).
"Also, I'm sure the stacked die reduces the ability to dissipate heat" Sure - from the cache. It's not stacked on top of the cores where most of the heat comes from.
@ lemurbutton: Agree. After all the hypes it seem to be a performance regression for many workloads. Not everything depend on cache size. Majority of the apps will fit in the 32MB cache. The performance in these apps will actually go down due to the frequency regression. That and the added cost will make it meme worthy.
"Meh, it's really not that impressive." Then you're not paying attention. The first time a CPU was launched with a stacked die was Intel's little Lakefield mayfly. It performed like crap (even compared with low expectations) and was available in a grand total of two products at an unreasonable price.
This will be the first mainstream high-performance desktop processor to launch with a stacked die. That it appears to improve performance in games - where performance improvements are hard to find - is no small feat.
Between Lakefield i5 L16G7nd i3 L13G4 only i5 was supplied. All Lakefield production as software validation and 'development' platform is in the same period of production June 2020 to date is 1.7% of Tiger U quad. mb
I wonder if this will be more expensive than the 5900X. I honestly thought this would make more sense on a 16-core 5950X, where the CPU cores are more bandwidth limited to DDR4. It should lead to a more decent Prosumer product.
Everyone has their mid-cycle "Whoops, need another performance SKU" products, but AMD's actually put in some more effort. Intel's "KS" (or "EE" back in the day): just ramp the TDP, volts, and clocks. NVIDIA's "Super" and "Ti", AMD's "XT", etc.
>I’m looking forward to getting it in hand for review – let us know what sort of tests you want to see.
I imagine this is already part of the review, but I'm interested in the 200 to 400 MHz loss vs 64 MB L3 cache gain in non-gaming applications. Can SPEC or other applications claw back that much with the extra L3 cache? Conventional wisdom would say "no", but I'd be curious.
200 MHz lower boost (-4%) 400 MHz lower base (-11%)
That lower base clock is an unexpectedly large reduction for a high-end CPU, especially w/ longer workloads (if I understand PPT correctly).
I'm also liking the Zen4 IHS: likely just the render (lighting / angle) that makes it look like the center bulges (i.e., a convex IHS), right?
Or is it actually convex? Some Z690 motherboards tried LGA1200 "fallback mounts", but they've not been received that well, I think, as the IHS height differences need the proper LGA1700 sockets.
All in all, I'm still really enjoying proper competition again in x86 desktop / enterprise. God, can we imagine another year of zero pressure against Intel again?
Will be eager to see the Zen4 vs Rocket Lake fight. May be the first time in years that they both have big launches in the same quarter, no?!
The *realistic* answer is, "fast enough for current 3D games: yes," just like any remotely recent desktop CPU. But if you want to know how it compares to other CPUs and how long it is likely to remain a viable gaming CPU, it needs to be tested in a situation where the rest of the system is gotten out of the way as much as it can be. That means overpowered graphics VGA and low resolution.
intel 11600K and amd 3800X are fast enough at 1440K so that is why. But yeah... Both intel 12+++ and amd 5+++ are useless at 1440p and 4K gaming! Just get the fastest GPU you can afford and you are good to go!
AMD is really into much more greed now. They are only issuing the 3DV refresh for a single SKU because of that BS shortage, wtf ? Console junk got a ton of Silicon allocated for those APUs which are high volume and the BGA dumpster technology got USB4 and others but that TSMC 6nm node. While the Desktop AM4 gets a single damn CPU with 200MHz clock rate cut off plus a single CPU for the hyped 3D-V refresh.
A big bummer, this is just like Intel. Intel did Rocket Lake backport to 14nm and screwed IMC on top with 2Cores and 4Thread loss to make Alder Lake appear good. AMD is now delivering on that hype of 3DV with a single damn CPU. Because they want to save their Zen 4 sales, if all the Ryzen 5000 processors got 3DV refresh, Zen 4 would make zero sense and who the fuck will even buy Zen 4 from Zen 3/3D. What a shame.
I expected this to blow Alder Lake but I understood these all companies are there to screw us only. EPYC-X Milan rejects where are they going ? Threadripper, ah that Single Chipset HEDT LOL. Fuck you AMD honestly.
Also there's no IOD refresh, so expect same buggy WHEA, USB drops again since it's the same silicon and AMD processors out of box demonstrate those issues, on top lower clock rate for that gaming junk performance.
Everyone on this site knows how Businesses operate. Even Ian expected AMD to deliver 16C32T 3DV refresh. It's not like AMD has to run business we do not get anything BS. Add anything better next time.
Oh I do not even own AMD processor nor even was willing to buy as I mentioned how the IOD has issues with the Ryzen 5000 series processors I expected a full stack 6000 series and addressing the issues with the 5000 from Firmware to HW. I just wanted new Technology on Desktops, more power in people's hand not BGA dumpster. And to do compute and all things including gaming.
The 3DV on AM4 is obviously a stopgap to appear more competitive against Alder Lake. It appears to me a classic case of AMD putting more of its effort towards Zen4 than the short term solution.
This is not a stop gap. It's a failure. No one should buy AM4 now, because it's dead end. Only ones who will are either needing AMD machine or are not interested in Big Little technology of Intel. Raptor Lake will be a drop in upgrade so this 5800X3D will be dead in the hands of 12900KS already as Intel is binning them at 5.5GHz, so RPL will easily destroy that.
As for Zen 4 vs RPL we do not know anything yet, I doubt Intel can beat AMD. But also coming to the effort ? What effort they already have EPYC-X bins of Chiplets. The rejected ones can be fabbed on that 3D Fabric and got a nice 6000 series desktop refresh. For the whole stack that would be massive. Plus they could bin it for Threadripper, which AMD killed pathetically and shamelessly just a fucking single generation on TRX boards is gutter trash.
AMD is clearly maximizing money on BGA dumpster and Consoles. Not on AM4.
I agree, "Only ones who will are either needing AMD machine or are not interested in Big Little technology of Intel. Raptor Lake will be a drop in upgrade so this 5800X3D will be dead". But look at this in a slightly different way. V5x is a late market technology at WIN 10. Alder/Raptor is adoptive adaptive at Win 11 optimized applications I suspect Intel will begin training how to at associated BIOS class for example at Systems Integration Conference and IDF.
AMD has always done well with late market product; 286-20 and 20 LP, 386 40, 486 133, 5k86, Athlon 2200 and the list goes on. Late market bridges through to adoptive adaptive while in this scenario 1) thread directed gets ironed out and 2) AMD figures out how to fit Raphael into the new Intel + MS reconfigured 'sand box' that is Win 11. Between the two V5x on AM4 is actually the 'more standard x86 platform'. My workstation is Office and basic business and I tend to buy late market on proven and stable, so, I'm the laggard referred but that does not mean I don't buy for stretch that will be no 3D 5800X; aim is 10 year life and on depreciating platform price basis and AMD channel clearance sale my aim is achievable fir any simulation I might want to run.
Yes "stop gap", failure [?] let's see if 3D can be cooled if when OC'd. Like Studebaker Hawk slap some fins on and add a blower. In summary, If the user sticks with Win 10 V5x is fine. If you're a developer Win 11 becomes mandate for that sort of application development then as suggested see how Raptor which is just a more core Alder v Zen 4 sort out.
I anticipate AMD will offer out the gate 15 M V5x 3D on q3 10Q financial reconciliation in relation channel supply data. Pursuant TR 5K, I suspect on the same q3 analysis around 1.194 M sit in a finished goods cage on an AMD silver bullet play waiting Alder Lake missing $749, $999 Extreme and $1189 to $1979 missing Xeon HEDT SKUs. Pursuant TR 5K, AMD needs the q2 = q3 price margin support over the Rembrandt APU deck that exists between V5x refresh tower 1 and Zen 4 tower 2, think suspension bridge.
"Maximizing money", As Alder ramp and Raptor run down volumes play out, through summer Rembrandt is the mass market mainstay not V5x 3D or Zen 4 yet on the ease of OEM APU integration on cost : price /margin. In the moment AMD runs back to its APU stronghold in a design scramble getting Zen 4 to fit within the NEW Intel tied to MS Win 11 thread directed applications optimized play ground.
"Because they want to save their Zen 4 sales, if all the Ryzen 5000 processors got 3DV refresh, Zen 4 would make zero sense and who the fuck will even buy Zen 4 from Zen 3/3D. What a shame." This is a bizarre hypothesis. Zen 4 is a much larger improvement - there's no way a 16-core Zen3D product would make that look bad.
"I expected this to blow Alder Lake" You expected too much.
Heat on package (area) thermals and cost, 32 Mb SRAM sliver + $45 per CCX to AMD before mark up to OEMs and their mark up and channel mark up. I estimate 5800X3D MSRP $553 on bottom-up cost : price / margin assessment following haukionkannel down com string observation a price rung void between 5900 and 12900K and then I did the estimate. There is no x86 CPU or compliment GPU shortage data can validate. mb
The thing that strikes me is AMD are only getting to 5nm two years after Apple ramped to volume production.
I get there are difference in LP vs HPC process, die size and yield challenges (tho don’t chillers help this). But it’s crazy they have taken so long. I mean what’s the point of having a three year process advantage over the competition if it takes two years to get the parts out.
Is it just being parked behind Apple in the queue that’s the issue, or am I missing something here? I mean if it takes another two years to get to N3 then Intel will CNN e ahead next year with 4nm Meteor Lake, no?
it's because it's expensive. According to TSMC they expected 8% of all revenue to be 5nm in 2020, and at that point only Apple was on 5nm, so Apple fronted 3~4 billion dollars for 1~2 quarters worth of chips. They also invest in TSMC's expansion directly.
Hard to know how much of the wait is because they wouldn't get enough volume due to Apple hogging the process, how much of it is AMD waiting for the process to be mature enough for a large chip to be sold to OEMs (Apple can swallow losses like nobody else because they own everything), and how much of it is AMD's design schedule not lining up neatly with N5 availability.
One other factor -- maybe the most important: Because AMD's direct competition is Intel (not Apple) and they simply haven't needed TSMC 5nm to beat Intel. It only makes sense for AMD to move to 5nm when Intel starts offering competitive products. Looks like Intel is finally doing that and so, what da ya know, AMD is now going to move to 5nm.
It's not hard to say how much is the latter -- AMD is unwilling to redo an already-working IO die, which would be needed to support N5. There's a reason mobile SoCs are getting Zen 3+ and desktop aren't -- even that jump would probably need a notable reinvestment into an IO die that'll just be deprecated with Zen 4 anyway.
From what we’ve seen in older Zen2 & later Zen3 chiplets the contact points for 3D stacking has been there for a while. My take on this is that this was supposed to be out a long time ago. I reckon they’ve had issues getting this working properly and manufacturable, now I think they’ve finally cracked it but AM4 is soon to be legacy, so that’s why we’ll see just one SKU on AM4 with stacking. However I do believe all the ‘learning’ for this has been done, and Zen4 will see lots of stacking options.
Expensive to produce... so no point of releasing it until you need expensive "flagship" product. When 5900 is $540 and 12900KF is $590, there is marginal slot in between those. 5800X3D at $550 would be price competitive Anywhere else it would not make sense.
Thank you for that observation thinking about the missing price rung between $540 5900 and $590 12900KF; $550. 32K of SRAM costs TSMC $5.40 in sand and on the pads x2 so $11.80 plus 100% mark up on extra package and test steps = $21.60 and TSMC price to AMD x2 = $43.20 (less if x1.5). On V5x full run volume all SKUs the average price from TSMC to AMD is $165 before AMD mark up to the OEM. So $165 + lets round up to $45 for 32 MB SRAM = $210 and AMD will mark up x1.5 to x2 on high to lower volume procurement means the OEM price is range $316 to $421 and then the OEM adds x1.5 = range $474 to $632 or if we average between that range $553. mb
For someone with a R5-3600 or below, this CPU might make sense. For someone building a system from scratch the choice is not so obvious, DDR4 board for 12th gen Intel or wait a bit for better options?
Gonna be a wait on my end for multiple reasons: - I always keep my build for a very long time and only ever update the GPU and SSD so I really want to have bleeding edge support at MOBO level. It's is not the case with Zen 3 and I find the latest Intel to be too power hungry. - The chip shortage makes prices ridiculous, I hope that it gets better by next fall - I have a 4K screen so I'd like to wait for the next GPU gen (please let mining be close to dead by then with most crypto using proof of stake) - Hopefully 4Tb M.2 SSD at a reasonable price
5800X3D might not be an interesting product, but it does fill in while we wait for 7000 series, and it adds an option for the final AM4 upgrade.
However, at this point, I'm waiting to see how much 5600X and 5800X come down in price, because i5 12400 is an absolute beast for cost/performance in games.
Ex post, their choice seems very sensible. We have yet to see AMD doing stupid or desperate things under Lisa Su, even if it takes a little longer to understand the wisdom of its choices.
Very few games really profit from high core counts yet and this is a chip specifically designed to offer the best gaming experience on AM4. I'm pretty sure they've tested extensively and somehow those numbers must have reached Intel, too, because otherwise team blue wouldn't bin to 5.3/5.0 GHz to contest the gaming crown.
In terms of pricing, I'd expect them to stay pretty close to actual extra cost. The stacked cache only exists, because they could meet a reasonable price and that cost could be very similar to an extra CCD--a little less for the silicon, a little extra for the tricky packaging. This isn't like Intel using GP3 iGPUs and e-DRAM to meet Apple's performance demands.
Of course I would have liked to own a 5950X with the extra cache, much less sure I'd have been ready to pay the premium, given that gaming is only a very minor side benefit in my use cases *and* the fact, that the 'ordinary' 5950X is doing quite well enough, thank you.
I am rather sure that in a couple of months review data will point out that AMD again was able to deliver performance with a lot less energy and brute force using a smarter and forward looking approach which will gain them market share in consoles and servers while adding nicely to their bottom line. They caught Chipzilla somewhat left-footed and I find that very entertaining.
Beyond the gaming crown volume production of these chips will build skills and confidence for future products of which we'll now see much more variety than without this stacking technology. It will keep this race much more interesting than cores, clocks and IPC alone.
Looking at the Zen 4 render, I'm afraid we could see tons of issues with excess thermal paste going where it shouldn't. "Liquid metat" doesn't sound like much of an insulator and without a side view there's no telling just how deep these caves for these die carrier SMD parts are...
Hmm. No TRX40 or sWRX80 news. No new HEDT processors. Suddenly, HEDT is a term that AMD cannot get its head around. Or, more likely, cannot engineer properly at the 7nm node. I can right now get a Xeon that has enough PCIe 4.0 lanes (64) and is faster than a 'leaked' zen 3 pro, and was launched in '21 Only 16 threads, but at 1299$ I COULD run two for cheaper than a 5xxx pro. or four, to get the max 64 threads....
Emulation tests, please! I appreciate Anandtech reviews including the Dolphin test. Emulation is one area where a low number of fast threads is very advantageous and in the more cutting edge emulators can make the difference between a smooth frame rate or a stuttery mess.
Strange, that no one is talking about PCI-E 5.0. What's the point in that? Barely anything moved to 4.0 and the ceiling of throughput wasn't hit. So, why?
I wonder what AMD has planned for 2023? Since DDR5 vendors won't have manufacturing capacity till Q4 2024+ there's going to be a DDR4 requirement on everything that isn't the high end market. Maybe they'll backport DDR4 support to Zen 4 and put Zen 4 on AM4??? AMD already said AM4 will be sticking around for quite a while yet.
Hi! this is nice article you shared with great information. <a href=https://tokomesinku.com/2019/09/30/jual-thermal-oi... thermal oil CPO kapal</a> Thanks for giving such a wonderful informative information. I hope you will publish again such type of post. Also, please check out about
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
99 Comments
Back to Article
Unashamed_unoriginal_username_x86 - Tuesday, January 4, 2022 - link
I'm curious about the CSGO results (tied with 5900x iirc), at launch of 5000 series we saw ~%45 uplift compared to 3000. Does this suggest the cache pressure was mostly relieved in the jump from 16MB to 32MB?nandnandnand - Tuesday, January 4, 2022 - link
It was already understood that not all games would benefit from a tripling of cache. CS:GO is a very old game that is easy to run. The drop in clock speed could explain the decrease.blanarahul - Tuesday, January 4, 2022 - link
“The cache does add a few watts to the power both in terms of idle and load. Rather than bin a stricter chiplet, the decision was made to reduce the frequency a little.”Come on AMD. Intel shamelessly asks for 190 watt on their i7 12700K. Why not do the same for what is already a niche product? Let loose the dogs of war! I wanna know what that 5800X3D can do at 4.5-4.7 GHz all core clock speed!
schujj07 - Tuesday, January 4, 2022 - link
Am4 max power is 142W. Therefore it is impossible for AMD to do what you ask.nullbyte - Wednesday, January 5, 2022 - link
Then how does my 5950x pull 230? 🙄Hinton - Wednesday, January 5, 2022 - link
It doesn't.Spunjji - Wednesday, January 5, 2022 - link
By you selecting modes of operation that exceed the design power. AMD allow people to do that, but they don't (yet?) ship chips that do it out of the box.Wrs - Thursday, January 6, 2022 - link
As someone with a 5800X, I know it has issues dissipating more than 150-160W with a high-end air cooler (the D15) while staying within 90C Tdie. That said, many workloads already run at 4.5 all-core on a mediocre bin and that D15 - but forget about Linpack or optimized AVX2. The question is how much harder it'll be to clock the cores on the same instructions with the additional silicon sliver on top.Mike Bruzzone - Sunday, January 16, 2022 - link
Thank you for the data 5800X 150 to 160W OC. My audit of 3rd party inputs recorded 147W so this data is useful. mbTomDev - Friday, February 18, 2022 - link
Your cooler is too weak. Get a better cooler e.g. a 360 AIO water cooler.Mr Perfect - Monday, January 10, 2022 - link
Eh, no thank you. The draw of Ryzen for gaming is partly because it can do so well without a ludicrous power budget. Remember how Intel's been ridiculed for it's high TDPs the last few years? Let's not loose the one sane vendor here.Qasar - Tuesday, January 11, 2022 - link
just as amd was be fore zen, but the difference i have been seeing, while most reviews show intel uses more power, there are alot of people that are fine with it, and keep saying its fine, and intel and amd use the same amount of power.Mike Bruzzone - Sunday, January 16, 2022 - link
3D TDP increase on 32 MB SRAM slice can be estimated on Milan F SRAM increase + 12.5W. mbvlad42 - Tuesday, January 4, 2022 - link
Not necessarily. If CSGO is one of the titles where performance increases by 1.4x vs the 5900X, then there is likely still significant cache pressure for that application. It is always possible to write a program that will have significant cache pressure for any given amount of cache. So, you cannot look at the average application performance improvement/bottleneck to make conclusions about a specific applications performance improvement/bottleneck.However, it looks like in general the cache pressure has been mostly eliminated. A 3x cache increase for an average of 1.15x performance, while still a good increase, indicates that on average we are very likely at diminishing returns or close to it.
icedeocampo - Friday, January 7, 2022 - link
I'm imagining that there should be compilers that aren't cache size agnostic. Would one really need to do it at the design/code level? or would there be optimizing compilers?lemurbutton - Tuesday, January 4, 2022 - link
Meh, it's really not that impressive.In order to accommodate the extra cache, it needs to lower clock speeds and increase the cost.
Seems like a gimmick or will have very limited use cases overall.
FreckledTrout - Tuesday, January 4, 2022 - link
The Zen4 version of this should be really good. I suspect the extra power savings from 5nm will make it no compromise in an 8 core chip.lemurbutton - Tuesday, January 4, 2022 - link
Doesn't matter what nm you use. The trade-off is more frequency or more cache. Also, I'm sure the stacked die reduces the ability to dissipate heat.Qasar - Tuesday, January 4, 2022 - link
" it's really not that impressive "like intel was before zen ? they werent that inpressive either. only because intel was getting beaten by amd, did intel finally wake up and started doing something again. some i know, are still not impressed with intel, and have no interest in their cpus, and will be going with amd when they upgrade in the next few months.
" increase the cost." like intel kept doing before zen ?
minisindad - Tuesday, January 4, 2022 - link
Why are you arguing, when based on your comments, you agree with what he said? Just because Intel did something similar in the past does not mean AMD should not be criticized for the same practices. They put something out to try and take headlines away from Alder Lake because they do not have anything of note until Zen 4 releases later this year.Qasar - Wednesday, January 5, 2022 - link
i wasnt agreeing with him, quite the opposite. lemurbutton was saying he isnt impressed with v cache, and i was countering that with how unimpressive intel was before zen was released. the fact is, most people are complaining about amd raising its cpu prices, while i would bet none of them, complained about intel doing the same. i dont recall ANY posts complaining about intels prices at all before zen, and intel would raise their prices by hundreds, and raises theirs by 50 bucks, and most jumped off the deep end. i can still remember seeing intel dropping their prices across the board when zen came out, and by at least 1k at the top end." hey put something out to try and take headlines away from Alder Lake" um you DO know that CES is for the most part, a few days away, regardless if intel said anything about alder lake, amd would of still announced something about their products that they have in the pipe to be released this year.
thestryker - Wednesday, January 5, 2022 - link
The 2600K-7700K 4c/8t processors all carried roughly the same retail price, 8700K added 2 cores and ~$50, 9900K was a bigger jump of around $120 with 2 more cores and then the 10900K close to the same with the 11900K being the first confusingly priced one and that covers 10 years so I'm not sure where your "Intel raising prices by hundreds" comes from. HEDT was the only place with stupid prices (though they were always increasing core count here too), and that was always true for Intel until AMD brought the competition before flying past.This part is AMD's halo product for gaming, and it doesn't appear to be all that impressive being limited to a 5800X running lower clocks. Perhaps AMD will release others, but it really seems like they picked the highest single ccx core count and ran with it. We likely won't ever know why the decision was made for sure, but it could be money, power consumption, heat or something else I haven't thought of entirely.
haukionkannel - Wednesday, January 5, 2022 - link
This will cost something like $550... Most people would not be buing 5950X3D for $1100 anyway.So they looked what 5900x and intel 12900k cost and did see that there is a slot in between where this product fits!
5850X3D would not be faster in the games, and when extra cache is useful mainly in the games, this was the only sensible place where they can use expensive 3D cache while maintain sensible price point.
Mike Bruzzone - Sunday, January 16, 2022 - link
5800X3D MSRP $550 on bottom up cost : price margin assessment range $474 to $632. mbQasar - Thursday, January 6, 2022 - link
thestryker " so I'm not sure where your "Intel raising prices by hundreds" comes from. " your are probably thinking in US $, every where else, the increases gen over gen where much higher.thestryker - Thursday, January 6, 2022 - link
Intel isn't to blame for regional pricing problems you may see so while you might be paying more they're not charging more.Qasar - Thursday, January 6, 2022 - link
" you might be paying more they're not charging more. " yes they are.intel kept raising prices in the US, so every where else prices also had to go up, in some places,by quite a lot, and all we got out of it, was 10% or less performance gains gen on gen, amd does 20%+ gen on gen, raise theirs by 50 ( here i think it was closer 100) and quite a few complained and cried.
thestryker - Friday, January 7, 2022 - link
Intel *never* charged more without increasing the core count until the 11th gen series. That's the reality since the desktop/HEDT split with Sandy Bridge. AMD raised prices due to their market position without changing core counts which is something that again never happened from Intel until 11th gen. AMD was smart to do it, but that doesn't make it look less bad to consumers.Qasar - Saturday, January 8, 2022 - link
thestryker, maybe where YOU are, but that wasnt the case every where. thats the reality here. gen on gen, intels cpu prices kept going up, increased core counts or not.Mike Bruzzone - Sunday, January 16, 2022 - link
Intel created, stacked performance mobile 'gaming' and 'workstation' NEW price rungs every generation since Haswell raising the price ceiling. AMD enabled and allowed the channel to set AMD MSRP to offset channels AMD inventory losses on all products before Zen 1x made obsolete by Zen. The practice has become inflationary. mbMike Bruzzone - Sunday, January 16, 2022 - link
Intel component and mobile U stuck on the shelf is the primary reason for regional pricing problems. Rocket deadweight, Comet S down bin, Coffee Refresh single thread, Tiger, Ice, Comet U quad laptops are on fire sale. Inflated dGPU card price is paying for channel's Intel inventory financial write down and/or loss. Worst is retail CPU + GPU and board kit deal where inflated GPU card price offsets the CPU loss leader and the board add its own mark up. There is no x86 or GPU card shortage data can validate it's called capital creation on channel Intel surplus inventory financial risk management. mbMike Bruzzone - Sunday, January 16, 2022 - link
Package area for pulling off the heat and material composition is not TR and Epyc Fujitsu composite? I think so. 59003D was always a lid off lab demo? I think so. mbSpunjji - Wednesday, January 5, 2022 - link
"They put something out to try and take headlines away from Alder Lake because they do not have anything of note until Zen 4 releases later this year."You might have missed how this was previewed months back? AMD and TSMC don't go through months and months of design and validation to chuck out something to mess with publicity for one of Intel's products. The technology will mostly benefit AMD's server division, but they promised something for the gamers and here it is, for better or worse.
Mike Bruzzone - Tuesday, January 18, 2022 - link
Rembrandt appears AMD 2020'mainstay 'consumer' and mass market gaming product between the two suspension bridge towers of 3D and Zen 4. Rembrandt is the deck. Then imagine an arch from tower 1 abutment reaching height at mid span and then back down to tower two abutment is commercial gross margin lift; TR 5K, Epyc where Genoa is already shiping on AMD loss of 7 nm cost : price / margin advantage vis a vis Intel on TSMC foundry market up and commercial GPU which AMD is now focusing getting out from in between Intel and Nvidia in the consumer market. mbkwohlt - Wednesday, January 5, 2022 - link
To be completely fair, given timelines of CPU design, Alder Lake design began almost immediately after Zen 1 was released.Every CPU between Zen1 and now that Intel released felt like a half-assed retort because all they could do until Alder Lake was finished was convert their HEDT line to their consumer socket, cut prices, and crank the wattage to extreme levels.
lmcd - Saturday, January 8, 2022 - link
Began after Zen 1???? You're crazy. The Alder Lake plans started way before that. Intel has been iterating their Atom cores for this moment literally since 2013. The Apple A7 was Intel's wakeup call, architecturally. And yes, it takes that long to turn around a giant in an industry where design to tapeout is usually ~5 years, and Atom was way beyond just one product cycle from reaching the necessary performance characteristics.Up until the A7, Intel assumed it could meet Apple on Apple's newfound turf. With the A7, Intel canned their smartphone product plans almost instantly and pivoted Atom toward its current trajectory. The only reason Alder Lake took quite this long is because of Intel's foundry slips.
Spunjji - Monday, January 10, 2022 - link
"The Alder Lake plans started way before that"Is there any evidence for that? I could believe they had some blue-sky ideas floating around about it, but given your quoted 5 years from design to tape-out the idea that Alder Lake is a direct response to Zen makes perfect sense.
Mike Bruzzone - Sunday, January 16, 2022 - link
Intel had disaggregate dice in desktop MCM parallel Core Haswell and Rad hard MCM SIP launched into space decades ago. mbSpunjji - Wednesday, January 5, 2022 - link
"The trade-off is more frequency or more cache"Unless you're already at the limit for frequency, then you'd just get more cache (and a little more power draw).
"Also, I'm sure the stacked die reduces the ability to dissipate heat"
Sure - from the cache. It's not stacked on top of the cores where most of the heat comes from.
nandnandnand - Tuesday, January 4, 2022 - link
If the Epyc Milan-X numbers are anything to go by, it's more impressive for non-gaming use cases.m53 - Tuesday, January 4, 2022 - link
@ lemurbutton: Agree. After all the hypes it seem to be a performance regression for many workloads. Not everything depend on cache size. Majority of the apps will fit in the 32MB cache. The performance in these apps will actually go down due to the frequency regression. That and the added cost will make it meme worthy.Spunjji - Wednesday, January 5, 2022 - link
"After all the hypes it seem to be a performance regression for many workloads."Based on what information?
Never mind, you're another of the FUDposters.
Spunjji - Wednesday, January 5, 2022 - link
"Meh, it's really not that impressive."Then you're not paying attention. The first time a CPU was launched with a stacked die was Intel's little Lakefield mayfly. It performed like crap (even compared with low expectations) and was available in a grand total of two products at an unreasonable price.
This will be the first mainstream high-performance desktop processor to launch with a stacked die. That it appears to improve performance in games - where performance improvements are hard to find - is no small feat.
Mike Bruzzone - Monday, January 17, 2022 - link
Between Lakefield i5 L16G7nd i3 L13G4 only i5 was supplied. All Lakefield production as software validation and 'development' platform is in the same period of production June 2020 to date is 1.7% of Tiger U quad. mbToTTenTranz - Tuesday, January 4, 2022 - link
I wonder if this will be more expensive than the 5900X.I honestly thought this would make more sense on a 16-core 5950X, where the CPU cores are more bandwidth limited to DDR4. It should lead to a more decent Prosumer product.
haukionkannel - Wednesday, January 5, 2022 - link
I expect this to become between 5900x and 12900K, so $550 MSRP seems most likely.ikjadoon - Tuesday, January 4, 2022 - link
Everyone has their mid-cycle "Whoops, need another performance SKU" products, but AMD's actually put in some more effort. Intel's "KS" (or "EE" back in the day): just ramp the TDP, volts, and clocks. NVIDIA's "Super" and "Ti", AMD's "XT", etc.>I’m looking forward to getting it in hand for review – let us know what sort of tests you want to see.
I imagine this is already part of the review, but I'm interested in the 200 to 400 MHz loss vs 64 MB L3 cache gain in non-gaming applications. Can SPEC or other applications claw back that much with the extra L3 cache? Conventional wisdom would say "no", but I'd be curious.
200 MHz lower boost (-4%)
400 MHz lower base (-11%)
That lower base clock is an unexpectedly large reduction for a high-end CPU, especially w/ longer workloads (if I understand PPT correctly).
kwohlt - Wednesday, January 5, 2022 - link
Intel KS is just silicon lottery sold for an upcharge.Ian Cutress - Friday, January 7, 2022 - link
Newsflash: that's what binning isikjadoon - Tuesday, January 4, 2022 - link
I'm also liking the Zen4 IHS: likely just the render (lighting / angle) that makes it look like the center bulges (i.e., a convex IHS), right?Or is it actually convex? Some Z690 motherboards tried LGA1200 "fallback mounts", but they've not been received that well, I think, as the IHS height differences need the proper LGA1700 sockets.
All in all, I'm still really enjoying proper competition again in x86 desktop / enterprise. God, can we imagine another year of zero pressure against Intel again?
Will be eager to see the Zen4 vs Rocket Lake fight. May be the first time in years that they both have big launches in the same quarter, no?!
ikjadoon - Tuesday, January 4, 2022 - link
*bracketsOxford Guy - Tuesday, January 4, 2022 - link
Duopoly is not ‘proper competition’.ikjadoon - Wednesday, January 5, 2022 - link
No arguments there!maroon1 - Tuesday, January 4, 2022 - link
Imagine having RTX 3080 and playing at 1080p LOLThe should tested the games at 1440p
Or use RTX 3060 if you want to test at 1080p
That would be more realistic
vegemeister - Tuesday, January 4, 2022 - link
The *realistic* answer is, "fast enough for current 3D games: yes," just like any remotely recent desktop CPU. But if you want to know how it compares to other CPUs and how long it is likely to remain a viable gaming CPU, it needs to be tested in a situation where the rest of the system is gotten out of the way as much as it can be. That means overpowered graphics VGA and low resolution.trenzterra - Tuesday, January 4, 2022 - link
Funny how they are using an RTX rather than a Radeon though.Spunjji - Wednesday, January 5, 2022 - link
RTX GPUs have a greater CPU dependency. It makes the comparisons more clear.haukionkannel - Wednesday, January 5, 2022 - link
intel 11600K and amd 3800X are fast enough at 1440K so that is why. But yeah... Both intel 12+++ and amd 5+++ are useless at 1440p and 4K gaming! Just get the fastest GPU you can afford and you are good to go!Spunjji - Wednesday, January 5, 2022 - link
Then you'll be hitting GPU limits, not CPU, and not seeing any real performance difference.And yes, that means buying a faster CPU for gaming is largely a pointless exercise.
It might be of interest to the 360Hz 1080p gamers, though.
Silver5urfer - Tuesday, January 4, 2022 - link
This is just sad.AMD is really into much more greed now. They are only issuing the 3DV refresh for a single SKU because of that BS shortage, wtf ? Console junk got a ton of Silicon allocated for those APUs which are high volume and the BGA dumpster technology got USB4 and others but that TSMC 6nm node. While the Desktop AM4 gets a single damn CPU with 200MHz clock rate cut off plus a single CPU for the hyped 3D-V refresh.
A big bummer, this is just like Intel. Intel did Rocket Lake backport to 14nm and screwed IMC on top with 2Cores and 4Thread loss to make Alder Lake appear good. AMD is now delivering on that hype of 3DV with a single damn CPU. Because they want to save their Zen 4 sales, if all the Ryzen 5000 processors got 3DV refresh, Zen 4 would make zero sense and who the fuck will even buy Zen 4 from Zen 3/3D. What a shame.
I expected this to blow Alder Lake but I understood these all companies are there to screw us only. EPYC-X Milan rejects where are they going ? Threadripper, ah that Single Chipset HEDT LOL. Fuck you AMD honestly.
Also there's no IOD refresh, so expect same buggy WHEA, USB drops again since it's the same silicon and AMD processors out of box demonstrate those issues, on top lower clock rate for that gaming junk performance.
Makaveli - Tuesday, January 4, 2022 - link
lol its like you just realized where you are in the pecking order.Shareholders and enterprise customers come first. Gamers are at the bottom of the totem pole, they have a business to run at all.
Silver5urfer - Tuesday, January 4, 2022 - link
Mr. Obvious is stating Obviousness.Everyone on this site knows how Businesses operate. Even Ian expected AMD to deliver 16C32T 3DV refresh. It's not like AMD has to run business we do not get anything BS. Add anything better next time.
Silver5urfer - Tuesday, January 4, 2022 - link
Oh I do not even own AMD processor nor even was willing to buy as I mentioned how the IOD has issues with the Ryzen 5000 series processors I expected a full stack 6000 series and addressing the issues with the 5000 from Firmware to HW. I just wanted new Technology on Desktops, more power in people's hand not BGA dumpster. And to do compute and all things including gaming.CrystalCowboy - Tuesday, January 4, 2022 - link
The 3DV on AM4 is obviously a stopgap to appear more competitive against Alder Lake. It appears to me a classic case of AMD putting more of its effort towards Zen4 than the short term solution.Silver5urfer - Tuesday, January 4, 2022 - link
This is not a stop gap. It's a failure. No one should buy AM4 now, because it's dead end. Only ones who will are either needing AMD machine or are not interested in Big Little technology of Intel. Raptor Lake will be a drop in upgrade so this 5800X3D will be dead in the hands of 12900KS already as Intel is binning them at 5.5GHz, so RPL will easily destroy that.As for Zen 4 vs RPL we do not know anything yet, I doubt Intel can beat AMD. But also coming to the effort ? What effort they already have EPYC-X bins of Chiplets. The rejected ones can be fabbed on that 3D Fabric and got a nice 6000 series desktop refresh. For the whole stack that would be massive. Plus they could bin it for Threadripper, which AMD killed pathetically and shamelessly just a fucking single generation on TRX boards is gutter trash.
AMD is clearly maximizing money on BGA dumpster and Consoles. Not on AM4.
SunMaster - Wednesday, January 5, 2022 - link
Anyone needing a computer now should buy AM4. The low DDR5 supply and price of Intel boards for the latest gen Intel makes it the only viable choice.Oxford Guy - Wednesday, January 12, 2022 - link
It also doesn’t have that Pluton downgrade.Mike Bruzzone - Wednesday, January 5, 2022 - link
I agree, "Only ones who will are either needing AMD machine or are not interested in Big Little technology of Intel. Raptor Lake will be a drop in upgrade so this 5800X3D will be dead". But look at this in a slightly different way. V5x is a late market technology at WIN 10. Alder/Raptor is adoptive adaptive at Win 11 optimized applications I suspect Intel will begin training how to at associated BIOS class for example at Systems Integration Conference and IDF.AMD has always done well with late market product; 286-20 and 20 LP, 386 40, 486 133, 5k86, Athlon 2200 and the list goes on. Late market bridges through to adoptive adaptive while in this scenario 1) thread directed gets ironed out and 2) AMD figures out how to fit Raphael into the new Intel + MS reconfigured 'sand box' that is Win 11. Between the two V5x on AM4 is actually the 'more standard x86 platform'. My workstation is Office and basic business and I tend to buy late market on proven and stable, so, I'm the laggard referred but that does not mean I don't buy for stretch that will be no 3D 5800X; aim is 10 year life and on depreciating platform price basis and AMD channel clearance sale my aim is achievable fir any simulation I might want to run.
Yes "stop gap", failure [?] let's see if 3D can be cooled if when OC'd. Like Studebaker Hawk slap some fins on and add a blower. In summary, If the user sticks with Win 10 V5x is fine. If you're a developer Win 11 becomes mandate for that sort of application development then as suggested see how Raptor which is just a more core Alder v Zen 4 sort out.
I anticipate AMD will offer out the gate 15 M V5x 3D on q3 10Q financial reconciliation in relation channel supply data. Pursuant TR 5K, I suspect on the same q3 analysis around 1.194 M sit in a finished goods cage on an AMD silver bullet play waiting Alder Lake missing $749, $999 Extreme and $1189 to $1979 missing Xeon HEDT SKUs. Pursuant TR 5K, AMD needs the q2 = q3 price margin support over the Rembrandt APU deck that exists between V5x refresh tower 1 and Zen 4 tower 2, think suspension bridge.
"Maximizing money", As Alder ramp and Raptor run down volumes play out, through summer Rembrandt is the mass market mainstay not V5x 3D or Zen 4 yet on the ease of OEM APU integration on cost : price /margin. In the moment AMD runs back to its APU stronghold in a design scramble getting Zen 4 to fit within the NEW Intel tied to MS Win 11 thread directed applications optimized play ground.
mb
Spunjji - Wednesday, January 5, 2022 - link
"Because they want to save their Zen 4 sales, if all the Ryzen 5000 processors got 3DV refresh, Zen 4 would make zero sense and who the fuck will even buy Zen 4 from Zen 3/3D. What a shame."This is a bizarre hypothesis. Zen 4 is a much larger improvement - there's no way a 16-core Zen3D product would make that look bad.
"I expected this to blow Alder Lake"
You expected too much.
Mike Bruzzone - Wednesday, January 5, 2022 - link
Heat on package (area) thermals and cost, 32 Mb SRAM sliver + $45 per CCX to AMD before mark up to OEMs and their mark up and channel mark up. I estimate 5800X3D MSRP $553 on bottom-up cost : price / margin assessment following haukionkannel down com string observation a price rung void between 5900 and 12900K and then I did the estimate. There is no x86 CPU or compliment GPU shortage data can validate. mbesoel_ - Tuesday, January 4, 2022 - link
My wallet is shaking in terror…Jon Tseng - Tuesday, January 4, 2022 - link
The thing that strikes me is AMD are only getting to 5nm two years after Apple ramped to volume production.I get there are difference in LP vs HPC process, die size and yield challenges (tho don’t chillers help this). But it’s crazy they have taken so long. I mean what’s the point of having a three year process advantage over the competition if it takes two years to get the parts out.
Is it just being parked behind Apple in the queue that’s the issue, or am I missing something here? I mean if it takes another two years to get to N3 then Intel will CNN e ahead next year with 4nm Meteor Lake, no?
whatthe123 - Tuesday, January 4, 2022 - link
it's because it's expensive. According to TSMC they expected 8% of all revenue to be 5nm in 2020, and at that point only Apple was on 5nm, so Apple fronted 3~4 billion dollars for 1~2 quarters worth of chips. They also invest in TSMC's expansion directly.Spunjji - Wednesday, January 5, 2022 - link
Hard to know how much of the wait is because they wouldn't get enough volume due to Apple hogging the process, how much of it is AMD waiting for the process to be mature enough for a large chip to be sold to OEMs (Apple can swallow losses like nobody else because they own everything), and how much of it is AMD's design schedule not lining up neatly with N5 availability.Blastdoor - Thursday, January 6, 2022 - link
One other factor -- maybe the most important:Because AMD's direct competition is Intel (not Apple) and they simply haven't needed TSMC 5nm to beat Intel.
It only makes sense for AMD to move to 5nm when Intel starts offering competitive products. Looks like Intel is finally doing that and so, what da ya know, AMD is now going to move to 5nm.
lmcd - Saturday, January 8, 2022 - link
It's not hard to say how much is the latter -- AMD is unwilling to redo an already-working IO die, which would be needed to support N5. There's a reason mobile SoCs are getting Zen 3+ and desktop aren't -- even that jump would probably need a notable reinvestment into an IO die that'll just be deprecated with Zen 4 anyway.Mike Bruzzone - Monday, January 17, 2022 - link
good point end of run and end of runway. mbDannyH246 - Tuesday, January 4, 2022 - link
From what we’ve seen in older Zen2 & later Zen3 chiplets the contact points for 3D stacking has been there for a while. My take on this is that this was supposed to be out a long time ago. I reckon they’ve had issues getting this working properly and manufacturable, now I think they’ve finally cracked it but AM4 is soon to be legacy, so that’s why we’ll see just one SKU on AM4 with stacking. However I do believe all the ‘learning’ for this has been done, and Zen4 will see lots of stacking options.haukionkannel - Wednesday, January 5, 2022 - link
Expensive to produce... so no point of releasing it until you need expensive "flagship" product.When 5900 is $540 and 12900KF is $590, there is marginal slot in between those.
5800X3D at $550 would be price competitive Anywhere else it would not make sense.
Mike Bruzzone - Wednesday, January 5, 2022 - link
Thank you for that observation thinking about the missing price rung between $540 5900 and $590 12900KF; $550. 32K of SRAM costs TSMC $5.40 in sand and on the pads x2 so $11.80 plus 100% mark up on extra package and test steps = $21.60 and TSMC price to AMD x2 = $43.20 (less if x1.5). On V5x full run volume all SKUs the average price from TSMC to AMD is $165 before AMD mark up to the OEM. So $165 + lets round up to $45 for 32 MB SRAM = $210 and AMD will mark up x1.5 to x2 on high to lower volume procurement means the OEM price is range $316 to $421 and then the OEM adds x1.5 = range $474 to $632 or if we average between that range $553. mbMike Bruzzone - Wednesday, January 5, 2022 - link
There is a conundrum occurring up comment string on V5x refresh and I did add again thanks for the price rung void observation. mbAbRASiON - Wednesday, January 5, 2022 - link
The v-cache is a big disapointment, should've been 5800, 5900, 5950 at least, should be available to buy within weeks.As it stands, they're sitting ducks vs intel until Zen 4 launch. That new $167 12400f sounds like a monster for the price.
BushLin - Wednesday, January 5, 2022 - link
For someone with a R5-3600 or below, this CPU might make sense.For someone building a system from scratch the choice is not so obvious, DDR4 board for 12th gen Intel or wait a bit for better options?
Lothyr - Thursday, January 6, 2022 - link
Gonna be a wait on my end for multiple reasons:- I always keep my build for a very long time and only ever update the GPU and SSD so I really want to have bleeding edge support at MOBO level. It's is not the case with Zen 3 and I find the latest Intel to be too power hungry.
- The chip shortage makes prices ridiculous, I hope that it gets better by next fall
- I have a 4K screen so I'd like to wait for the next GPU gen (please let mining be close to dead by then with most crypto using proof of stake)
- Hopefully 4Tb M.2 SSD at a reasonable price
meacupla - Wednesday, January 5, 2022 - link
5800X3D might not be an interesting product, but it does fill in while we wait for 7000 series, and it adds an option for the final AM4 upgrade.However, at this point, I'm waiting to see how much 5600X and 5800X come down in price, because i5 12400 is an absolute beast for cost/performance in games.
abufrejoval - Wednesday, January 5, 2022 - link
Ex post, their choice seems very sensible. We have yet to see AMD doing stupid or desperate things under Lisa Su, even if it takes a little longer to understand the wisdom of its choices.Very few games really profit from high core counts yet and this is a chip specifically designed to offer the best gaming experience on AM4. I'm pretty sure they've tested extensively and somehow those numbers must have reached Intel, too, because otherwise team blue wouldn't bin to 5.3/5.0 GHz to contest the gaming crown.
In terms of pricing, I'd expect them to stay pretty close to actual extra cost. The stacked cache only exists, because they could meet a reasonable price and that cost could be very similar to an extra CCD--a little less for the silicon, a little extra for the tricky packaging. This isn't like Intel using GP3 iGPUs and e-DRAM to meet Apple's performance demands.
Of course I would have liked to own a 5950X with the extra cache, much less sure I'd have been ready to pay the premium, given that gaming is only a very minor side benefit in my use cases *and* the fact, that the 'ordinary' 5950X is doing quite well enough, thank you.
I am rather sure that in a couple of months review data will point out that AMD again was able to deliver performance with a lot less energy and brute force using a smarter and forward looking approach which will gain them market share in consoles and servers while adding nicely to their bottom line. They caught Chipzilla somewhat left-footed and I find that very entertaining.
Beyond the gaming crown volume production of these chips will build skills and confidence for future products of which we'll now see much more variety than without this stacking technology. It will keep this race much more interesting than cores, clocks and IPC alone.
Spunjji - Wednesday, January 5, 2022 - link
A sensible take@Sasuke - Friday, January 14, 2022 - link
Well said!abufrejoval - Wednesday, January 5, 2022 - link
Looking at the Zen 4 render, I'm afraid we could see tons of issues with excess thermal paste going where it shouldn't. "Liquid metat" doesn't sound like much of an insulator and without a side view there's no telling just how deep these caves for these die carrier SMD parts are...t.s - Wednesday, January 5, 2022 - link
it's a second hand market nightmare.ballsystemlord - Wednesday, January 5, 2022 - link
Yes, it's going to be fun watching this play out.Kevin G - Wednesday, January 5, 2022 - link
The R7 5800X3D appears to be AMD's version of the I7-5775C from back in in the day.The Von Matrices - Wednesday, January 5, 2022 - link
That was an i7. The i5 was the 5675C.croc - Thursday, January 6, 2022 - link
Hmm. No TRX40 or sWRX80 news. No new HEDT processors. Suddenly, HEDT is a term that AMD cannot get its head around. Or, more likely, cannot engineer properly at the 7nm node. I can right now get a Xeon that has enough PCIe 4.0 lanes (64) and is faster than a 'leaked' zen 3 pro, and was launched in '21 Only 16 threads, but at 1299$ I COULD run two for cheaper than a 5xxx pro. or four, to get the max 64 threads....evilpaul666 - Thursday, January 6, 2022 - link
Emulation tests, please! I appreciate Anandtech reviews including the Dolphin test. Emulation is one area where a low number of fast threads is very advantageous and in the more cutting edge emulators can make the difference between a smooth frame rate or a stuttery mess.MDD1963 - Friday, January 7, 2022 - link
coming in '2nd half of 2022', eh?That means 31 Dec, I fear...
imaskar - Saturday, January 8, 2022 - link
Strange, that no one is talking about PCI-E 5.0. What's the point in that? Barely anything moved to 4.0 and the ceiling of throughput wasn't hit. So, why?dicobalt - Monday, January 17, 2022 - link
I wonder what AMD has planned for 2023? Since DDR5 vendors won't have manufacturing capacity till Q4 2024+ there's going to be a DDR4 requirement on everything that isn't the high end market. Maybe they'll backport DDR4 support to Zen 4 and put Zen 4 on AM4??? AMD already said AM4 will be sticking around for quite a while yet.distributor boiler - Saturday, February 19, 2022 - link
Hi! this is nice article you shared with great information. <a href=https://tokomesinku.com/2019/09/30/jual-thermal-oi... thermal oil CPO kapal</a> Thanks for giving such a wonderful informative information. I hope you will publish again such type of post. Also, please check out about