I never want to see another post about how "AMD never sandbags!!! Why would a company ever do that???"... -_- ... They do this crap for basically every single major launch. Sandbagging is practically in AMD's fundamental marketing DNA at this point.
You say that like it's a bad thing. It's a great marketing decision; usually marketing benchmarks and the like are heavily cherry picked and to be taken with a grain of salt. With AMD, they're generally not advertising anything they can't at least meet if not beat. We're so used to the usual marketing it seems like you're getting actively offended when a company tries to be honest.
If you believe AMD, recent changes improved IPC by about +2% over the previous figures. Maybe the improvements were made after they were already being accused of sandbagging.
Also, because clock speeds and TDPs play a bigger role in these performance gains, a single number doesn't represent both the 7600X and 7950X very well.
"The combination of which will allow AMD to deliver what they are saying is now a 13% increase in IPC over their Zen 3 architecture – up from an 11% claim as of Computex"
That's straight up a misquote/misrememberence by Ryan. It was listed as "+≈8-10% IPC" on the official "Zen 4 Core" slide from Computex. Stop regurgitating Ryan's mistake as facts without you ACTUALLY fact checking it yourself. Makes you look kinda like a self obsessed, tech illiterate imbecile. 🤷♂️
I'll look forward to the reviews, but given their pricing it looks like my inclination to wait until '23 and MTL/Zen5 is the right one. I'm looking for something in the $400 ish range and this pricing just looks absurd compared to where 12th gen Intel is and launched.
Perhaps they really will maintain enough of an advantage, or Intel will raise their prices enough to make this make sense. At the lower end this just doesn't seem even remotely possible given that there isn't a DDR4 option to keep RAM prices lower.
How so? Everyone knows Raptor lake is coming. It will be a 5% increase at best. AMD will win most benchmarks against Raptor and at a cheaper price (but with higher margins), and consume less power. Then in about 3 months along comes Zen4 X3D to completely take the win in all benchmarks. AMD firmly has Intel in its back pocket. But go ahead if you want to buy Raptor it’s your choice. AMD will completely sell every Zen4 die it makes.
Not 5%. More like +10-15% single-threaded performance for a 13900K vs. 12900K, and 12900K is the chip AMD wanted to compare to. And the period of time between Zen 4 and Raptor Lake launches will be short. You have no idea about pricing.
Intel have already publicly stated they will need to increase prices by 15%-20% due to inflation & supply chain issues etc. Meanwhile the 7950X is already $100 cheaper than the 5950X when it was released. It is also no secret that due to the die size and usage of chiplets AMD have much more pricing freedom than Intel. Also the period of time between Raptor Lake & X3D will be short. You clearly have no idea about anything.
Intel has absolutely not said anything about what products will have higher prices nor what that increase will be just that there will be increases.
The 7950X is still around $100 higher than the 12900K launched at which gives Intel plenty of room to increase the price for the 13900K and still be cheaper than AMD.
AMD has said X3D for datacenter is coming 1H23, but anything regarding desktop timeline are rumors. There's no doubt it will arrive as it'll be fantastic for gaming, but since Zen 5 is 2024 they may hold onto it longer as a MTL competitor rather than RPL if Zen 5 is later in the year.
Raptor Lake has more E cores, by not moving to a smaller node the die is going to be bigger. Current leaks have it at 257mm^2 a good 50mm^2 more than Alder Lake. That means less chip per wafer, less yields and harder to bin for faster parts. Intel will have to increase prices for sure just on the fact that Raptor Lake uses more silicon. If Raptor Lake is to be made in the US, it will cost even more. TSMC founder have stated this year, that chips made in its Oregon fab cost 50% more than chips made in Taiwan.
There is definitely more silicon being used, and likely through the entire product stack if the leaks are accurate. This does not actually mean proportional cost increases as there have undoubtedly been yield improvements on the Intel 7 node over its lifetime.
As for clockspeed I'd point you to the Chips and Cheese analysis of the ES RPL part they did cache testing on. They explain a bit of the architectural changes that should allow it to hit higher clocks.
Lastly never believe anything out of the mouth of someone with a vested interest against what they're talking about.
Yield improvements will never be as good as a chiplet design. And yes I would agree with you that the TSMC does have a vested interest to keep production in Taiwan. But the US does have higher labor cost and it is further from the supply chains. Intel also has less revenue stream to recoup its development cost on Intel 7 node (10nm). The battle this gen will be won on price and availability, real world performance will be too close to matter between Zen4 and Raptor Lake.
Intel fabrication cost at iSF10/7 is in parity with TSMC 7 nm. Intel 7 at run down will be in parity with TSMC 5 through its production volume peak after which TSMC 5 will have the run down advantage. mb
Where TSMC gains on Intel cost wise is of course legacy fabrication and for products priced less than Intel variable costs range $39/$79/$109 varies on product lines. Intel contracting out its own products priced less than variable cost works. This is also why Intel never produced StrongARM because its production cost structure subject variable cost would have made SA a loser. Bay Trail was similar on so said contra revenue producing at a loss and hoping to make up in the future never happen. mb
Silicon area in Raphael subject [including] the 6 nm i/o + iGPU block appears the same as Vermeer; measure the die in the photo. Raphael costs the same as Vermeer on a per unit basis at + 20% cost + 30% good dice. Vermeer price from TSMC to AMD $165 and Raphael is the same, maybe < $10 but there are R7x premium price voids missing and those SKUs as a price support will increase cost on average across the full line. mb
The AMD Raphael advantage is + 30% dice = less total wafers that can be redirected to other products. mb
"5nm costs twice as much per wafer than 7nm". At risk production.
Moore law doubles transistors every next node. Rocks law doubles cost every next node and when reaching peak production volume, theoretically doubling volume, cost equals parity with the prior node, then through run down volumes the cost continuously drops to marginal cost at run end; the very last unit produced. Profit maximization produces until the very last unit's revenue value = cost and that's run end.
Vermeer in relation to Raphael. V5x normalized area across the grade SKUs over the full run with its i/o die = 243 mm2. R7x normalized on V5x full run percent grade SKU split = 239 mm2 near same. Which is interesting NEW data because I originally estimated + 20% cost and + 30% good dice on 5 nm normalized ccd and the i/o dropping to 171 mm2 of area IT DID NOT and the reason is the 6 nm i/o + iGPU is no less area than the 12 nm i/o.
There seems to be disagreement about the area of the 5 nm ccd. Some report 77 mm2 and others report 71 mm2 a less 6.8% difference. At R7x 5 nm 77 mm2 ccd is 8.4% smaller than V5x 7 nm 84 mm2.
Whether V5x or R7x an 80% yield on edge and area defects delivers roughly the same number of normalized dice sufficient for 230 to 255 produced components per wafer.
On V5x full run by grade SKU data the average weighed $1K price for R7x is $466 / 3 way stakeholder (TSMC/AMD/OEM) split = TSMC price to AMD at $155 and V5x was $165 so within $10.
AMD markup dependent OEM volume is * 1.5 to * x2 so a full line procurement mirroring V5x full run at 16C = 19.22%, 12C = 23.01%, 8C = 32.67%, 6C = 25.09% = range $233 to $310 per unit across 10 1st tier customers across 60 M unit production run means ten 6 M unit procurements to get the $233 to $310 price per component on the grade split noted.
OEMs are brokers and they will resell whatever overage they don't need or SKUs they don't want lowers their procurement cost. This is how AMD and Intel distribute to the secondary market beyond AMD direct master distributors.
The cost : price / margin of Raphael is the same as Vermeer. That's your price increase from 7 nm to 5 nm = no cost savings.
If an efficient TSMC production regime they are not always that way on TSMC effective method that flattens change in quantities produced the efficient first 5% of production volume will cost $155. At 20% of the run marginal cost to produce efficiently drops to $45, at 65% pops to $55 on peak volume, at 88% complete this is all about sort now marginal cost drops to $9 and the last 12% cost drops to $3. TSMC full run marginal cost to produce the complete component on a 5 phase run = $41.27
If TSMC increases efficiency by producing more volume between phase 2 and phase 3 peak production the cost can drop measurably.
Unless AMD negotiates for a percent of this phased volume cost decrease, AMD won't see a penny of it all going to TSMC fab and the package house.
You're 5 nm price increase is no cost savings in relation 7 nm, not double just flat.
As I collect R7x supply volume data I'll be able to perform a more precise cost : price / margin assessment
of course is calculations arent right , but because he posts NO sources to ANYTHING he posts, how do does he prove it ? is the way this guy works. he just posts a bunch of BS, and fluff. the guy is a flake and a fraud. just try to ignore his spam posts.
The comparison to the 12900k is due to the fact that the 13900k IS NOT AVAILABLE. Tell me how could a rival company compare their new product to a rival's that hasn't been released?
I didn't say that they should. All I said is that Raptor Lake is the elephant in the room and y'all sperged. If Raptor Lake launches on October 20, that's not even a full month after Zen 4 CPUs.
While you said it is the elephant in the room, you also said "12900K is the chip AMD wanted to compare to." I was pointing out that in no way is AMD able to compare their Zen 4 CPU to Raptor Lake when Raptor isn't even out yet. Therefore they did the best they could and compared it to the current best from Intel.
for games zen4 with 3d cache will probably beat raptor lake.
and for content creator like myself. i have too look at power draw in these times. i have 8 systems. mostly threadrippers and 3x 5950x. the energy bill is pretty high.
if AMD can deliver 90% of the performacne at 50-60% the energy draw... i know what i am buying.
Whenever it comes out, sure the v-cache 7000 will absolutely take the gaming crown, but the poster I was responding to was acting like the 7600X was faster than the 12900K period and it won't be.
Absolutely if you run your hardware at the edge all the time things like power draw certainly become infinitely more important. I'd make the same decision in your shoes as it's definitely the right one, but for me where I do mostly lighter threaded with periodic all core max performance for both is what I'm looking for.
Assuming availability and street prices are reasonable, my next system will be a 7950X. That price cut to $699 means that if you can use the power there's no reason not to go all the way.
3d cache seems to increase price about $150 for single chiplet, so about $300 for dual chipset CPUs... It will always be flagship priced product! Otherwise they would put it on all their CPUs, but it does not make sense, because it is expensive and reduce most productivity workloads, unless AMD has made magic and cache does not heat the cpu below at all...
And extra $150 to $300 CPU sound worse than extra $150 to $500 CPU... It just does not make sense in low cost products!
It cost TSMC $45 to add the 32 mm2 64 MB SRAM to 5800X, But by the time 3D gets to the end buyer price reflects the added margin AMD to OEM, OEM to End Sales. mb
I third this. I'm waiting for the X3D as a *gamer*. By the time the X3D CPUs come out, the B series mobos might be out, what NVIDA & AMD got for new gen GPUs will be revealed or also released, DDR5 pricing might drop a bit, Intel will have released their 13th gen CPUs for competition (GPUs too?), and more ATX 3.0 power supply options will have dropped for "FuTuRe PrOoFiNg."
Their iGPU competition from Intel has around 0.5TFLOPS (eg UHD 770 as in 12500 and up which is plenty for desktop) - I doubt they'll try to beat it (much), and not come close to the 2TFLOPs of their best APUS.
The RDNA2 iGPUs in the G series are better than anything Intel has. What we will see in Zen 4 though are iGPUs designed around basic desktop functionality. Basically you will be able to put a non G CPU into a business desktop and not need a dGPU.
The bottom line issue is that there is a $320 price premium to go with a Ryzen 5 7600 with 32GB of DDR5 6400 CL20 over a Ryzen 5 5600 with 32GB of DDR4 3200 CL16. If you are on a budget that $320 can buy a whole lot more GPU.
Your comparison makes no sense at all. You are using the prices for top of the line DDR5 but comparing it against every day DDR4. When you use every day vs every day level RAM, the difference is about $80 more expensive for the DDR5. Those prices will be coming down as more DDR5 hits the market. Plus DDR5 has the advantage that companies will use higher density chips to make it so prices will drop even further.
> So while all of the Ryzen 7000 chips come with an iGPU
Does this mean they won't support ECC UDIMMs anymore? I hope that's not the case, because the AM4 Ryzen provided a great way to get ECC support without going with an expensive Xeon or ThreadRipper.
First - Die Size comparison is kinda not perfect, because Intel uses 30% of the ADL P core Golden Cove for AVX512 transistors which they disabled on the 12900KS and some of the newer batches probably because they are run by a bunch of stupid bean counters.
Two - 7600X vs 12900K is silly, because a damn 12900K is not just used for stupid gaming, because a 12700K can deliver 95%+ performance of that processor. And the MT of 12900K will be higher than 7600X despite the crappy biglittle junk.
Now that is out the major improvement I'm seeing is superior Clock speeds. 5.7GHz on an AMD chip is insane. That's super high mark. I bet Intel Raptor Lake will have 5.8GHz to win the Clock speed battle but the 8P cores of RPL might be super fast in gaming for 13900K paired with L2 improvements and new IMC. Although for MT I think 7950X will dominate that new RPL chip.
EXPO is looking interesting, instead of the absolute nonsense IMC on Ryzen until 5000 this time they have a solid low latency on DDR5 plus a real standard of Memory profile like XMP. That's really means the IMC on Ryzen is really a big improvement.
The major aspect is AM5 until 2025+ that's a win, AMD is going to saturate the market again like AM4 but the stability and the quality of the BIOS and CPU is up in the air, AMD failed to get me as a customer because their CPUs cannot be tuned properly due to lack of datasheets and lack of fixed clockspeed, they won't hold Boost as well. Next the IMC was a mess. IOD was a disaster, as even today we have this WHEA / USB issues on the platform it won't be fixed. So AMD has to prove it. However I think their OC is still a dud with Zen 4 too, the Clockspeed is way too much I bet they are not going to leave anything esp the fact that Zen 4 is a clear cut derivative of Zen 3 just like Intel's Comet Lake 10th Gen being the apex position for Skylake design maxing on Cache / Clocks / Core count. Same for AMD's Zen 4 vs their rest except for the core counts.
I think X3D is going to come next year maybe for 2-3SKUs ? 7950X3D / 7800X3D ?
All in all somewhat very interesting timeline seeing this high competition. Shame how Intel is succumbed to their biglittle junk. Good thing is AMD Zen 4 will destroy those BGA ARM trash which is still MIA and also Apple's nonsense as well in one shot. M series lost vs ADL Zen 4 will destroy it even more.
I forgot to mention ILM, why did AMD not mention any of it ? It would have been a solid move esp the fact how LGA1700 is plagued by that Intel's factory design failure. AM5 socket has extra reinforcement as per Igor's Lab, it's a no brainer over Alder Lake unless you want more tuning and fun as AMD processors won't have those features, I doubt they will add more since the Voltage will be sky high on this N5 refined for High performance more like High K Metal gate of older nodes from Intel and TSMC. Zen 3 was 1.4v on all X3D got binned due to low heat dissipation of Cache Stack needed a good silicon.
All in all Intel's move is next esp on the ILM side I will be curious if they care or gonna dump the LGA1700 along (It's going EOL, RPL will be last) with it's failed ILM quality.
> Good thing is AMD Zen 4 will destroy those BGA ARM trash which is still MIA and also Apple's nonsense as well in one shot. M series lost vs ADL Zen 4 will destroy it even more.
On dear. 170w boosted to 230w. Amazing. I thought 5nm meant that Apple can match the M1 in efficiency. Not even close.
Nobody cares about a BGA use and throw dumpster class machine vs a Desktop Socket with PCIe slots and massive PCH I/O. Apple can release MX on 1nm or whatever they want, it will always be inferior to an LGA / PGA socket machine. The OS itself is crippled to death, constant updates and death of all old Software due to 32bit abandonment. Who even cares about this Apple nonsense ? The market is 12% tops for that Mac OS. If you even compare BGA junk from AMD and Intel it will also render the Apple useless, esp the fact that they have affordable pricing with user optional PCIe SSDs and SODIMMs and 144Hz displays.
This is a bottom feeding load of shite. Seriously, 32 bit abandonment? Linux and Windows have abandoned that as well. Toss a VM and you get 32 bit back. UNIX on OS X isn't crippled one bit. You want extra features you've got options to get them back.
I would have personally hoped Apple abandoned Intel for Mac Pro with Zen 3.x but that never happened.
Zen 6 at the latest will introduce a unified memory back plane ala Apple, multiple FPGA based Engines and more ala Apple, a multi-chiplet integrated GPU and more.
By then RDNA 5 will most certainly not remotely resemble today's RDNA 3.0 and most board manufacturers will cut back on PCIe slots as unnecessary. PCIe 7.0 will be coming out and the bandwidth will be massive. Apple still uses PCI if you forgot, how else do you think TB3 interacts?
I love the OS updates. Never a broken system. Linux on the other hand breaks constantly. Windows is irrelevant in this conversation.
You are smoking some high tech crack. 32Bit means not 32Bit OS smart pants, means 32bit applications. Mac OS abandoned it, first understand before you go off hard. Windows is as relevant as it gets, the OS has highest backwards compatibility on the planet. Windows 7 still gets ESU updates and it's going upto 2025 when Windows 10 gets it's 10 year policy of EOL done, that said LTSC 2019 will have support upto 2030 since it has 10 year update since release. Meanwhile Jokebook Pros all are trashed out with failing KBs, Heaitng issues, Lack of I/O ports forget PCIe expansions. Then the crap 60Hz garbage decade old panels selling for super premium.
The actual bullshit is yours from the Zen 6 part. OS X is not equal to Unix, nope. Not even a bit. Zen 6 and all that nonsense from where did you get that information ? More Apple More Apple. Just refrain from this nonsense here, just throwing up in the air and see what sticks to asset your fallacy and weak arguments. Everything is ala Apple crap. What a damn joke lol. And more boards will remove PCIe slots rofl yeah did ASUS tell you that or AMD leaked that information. And now the cope is Thunderbolt 3, yeah sure like I love to haggle a stupid BGA pile of junk with a colostomy Thunderbolt 3 outdated junk dock, USB 4 beats even Thunderbolt 4 because TB4 doesn't have a lot of bandwidth it's just essentially 3 but with more stringent limits to stop OEMs to skimp on bandwidth, even Z590 boards have TB4 Maple Ridge controllers so this Mac Junk is severely limited on all fronts - SW, HW, DIY end of story.
Silver5urfer: "OS X is not equal to Unix, nope. Not even a bit."
UNIXtutorial.org: "macOS is the second most widely used desktop operating system after Microsoft Windows. It is also a desktop OS that has UNIX 03 certification, meaning it’s a truly Unix OS rather than Unix-like OS" https://www.unixtutorial.org/software/macos
Silver5urfer "OS X is not equal to Unix, nope. Not even a bit. "
The Open Group: UNIX® Certified Products
• Apple Inc.: macOS version 12.0 Monterey on Apple silicon-based Mac computers • Apple Inc.: macOS version 12.0 Monterey on Intel-based Mac computers • IBM Corporation: AIX version 7, at 7.2 TL5 (or later) on systems using CHRP system architecture with POWER™ processors • IBM Corporation: AIX version 7, at either 7.1 TL5 (or later) or 7.2 TL2 (or later) on systems using CHRP system architecture with POWER™ processors • Hewlett Packard Enterprise: HP-UX 11i V3 Release B.11.31 or later on HP Integrity Servers • Huawei Technology Co., Ltd: Huawei EulerOS 2.0 on Huawei KunLun Mission Critical Server • IBM Corporation: AIX 6 Operating System V6.1.2 with SP1 or later on Systems using CHRP system architecture with POWER™ processors and 2, 8 or 128 port async cards • IBM Corporation: AIX 5L for POWER V5.3 dated 7-2006 or later on Systems using CHRP system architecture with POWER™processors • IBM Corporation: z/OS V2R1 or later with: z/OS V2R1 or later Security Server and z/OS V2R1 or later C/C++ Compiler on IBM zSeries Processors that support z/OS Version 2 Release 1 or later • The SCO Group, Inc.: UnixWare ® 7.1.3 and later for single and multiprocessor systems based on IA-32 and compatible processors and conforming to PC architecture • The SCO Group, Inc.: SCO OpenServer Release 5 and OpenServer Release 6 on Single and Multi-processor Industry Standard Intel architecture platforms
Better luck next time, that is a strawman as it doesn't have literally anything except the UNIX omg it's the certified OS !! If anyone wants to use UNIX OS they will use Linux flavors not that closed source garbage OS called Mac OS which runs on a disposable Hardware vs Linux running on Pi or even iPod Classic.
Just last week my arch startup broke and I had to chroot from another image to fix it. And that was updating a few days after they discovered the bug, that AFAIK is still not corrected needing manual intervention: https://old.reddit.com/r/archlinux/comments/x1lcoi...
I think what they were trying to show is that even their "small" CPU beats, or matches, a top one from Intel. "Look, Ma, even our little brother beats the competition!"
Intel simply can’t match AMD specialized 3DV tech. This time intel has to contend against both new arch and the extra cache, suffice to say Intel can’t contend with their traditional tech which is simply not good enough in this case (gaming). And for multi cpu, I don’t think 24 cores are that relevant, anyone who really needs more than 16 can go and buy Threadripper, which is also easily faster than Intels stuff.
I was looking at that R7 7700X and it's MSRP is just slightly higher than I paid for my R7 5800X. I suspect that by Black Friday, we'll be seeing some good deals on this CPU. As much as I'd love to upgrade, it's only been about a year since my system was upgraded, so sadly I'll be sitting out this upgrade.
No 5800X msrp was 450$ and 7700X is 399$ which resembles the msrp of the 3800X. But I should say it’s not the direct successor, so this is a bit misleading.
I'am still OK with my [email protected] . Skipped DDR4 entirely and will go with Zen3 or Raptor DDR5. But the prices are rising steeply, 12700 got from 340Eur to 405Eur in 2 months.
I don't know how in the world you get by on such an old architecture for today's applications, hardware, and security needs. I had an i5 2500K 4.7GHz overclock gaming build up to 2017 when it started bottlenecking my new RTX 1080 Ti GPU (still have it as a Win7 offline only retro gaming PC with the GTX 970 it was last tied to). Hard to believe that LGA 1155 chipset is over 10 years old now.
It all depends what you use it for. I have an i5-2600 (I think, might be 2500) laptop that I still use for work. Has 16 GB of RAM, an SSD, and a new wifi module. Ran Windy for years, and upgraded to Win10 easier this year without issues.
Used mainly for network troubleshooting, serial console for switches and servers, remote access to servers via IPMI and SSH, and some word processing.
Obviously it's not my primary workstation, but it serves it's purpose.
"Driving AMD’s gains in this newest generation of desktop CPUs is a combination of architectural improvements underpinning the Zen 4 architecture, as well as moving production of the CPU core chiplets to TSMC’s leading-edge 5nm process."
No. Driving AMD's gains are absurd power draw increases for this generation.
Well 13% average IPC increase means depending on application you will get that much more performance AT THE SAME clock speed, and lower power due to 5nm, as Zen 3. AMD also showed that Zen 4 was more efficient than Zen 3 at all TDP levels. Take that with a grain of salt as it was 1st party information but once we have the benchmarks we will know for sure.
i will sure buy a 7000x at some point. but this time im not jumping into the first generation of mainboards. not that i had bad experiences. my first crosshair x370 was pretty great. but with a 5950x i can wait a bit before i upgrade that system. i guess by april next year there will be new mainboards with even better specs for a lower price. even if raptor lake beats zen4 by 10% ..... with the rumored power draw raptor lake needs i stick to AMD.
I highly doubt you can compare 7th gen to first gen Ryzen. Just because it’s a new platform doesn’t make all their experience and advances go away. So no it will probably function smoothly.
A lot of them look like they're built off of the X570 designs, with maybe a little more bling (that's a good thing). Features seem to be basically "more of the same" e.g. Wifi 6E, 2.5G LAN, etc. Some do look affordable even if most are TBA on pricing.
I'll be curious to see some reviews that include actual power efficiency, i.e. how much power is required to finish the same unit of work, versus Zen 3. And if it's easy to turn the maximum TDP in the motherboard, both 65W and 105W figures for Zen 4 octo-cores (and/or 105/170 for dodeca-cores) would be great. Just knowing how easy it is to configure that TDP would also be great.
Plan A is still to upgrade the old Sandy Bridge to Zen4 + AM5, but the decision to chase clockspeeds and increase TDP has me tempted to go with a potentially second-hand Zen 3 option instead, especially as even brand new the prices on those are rather tempting.
Given the configurable TDP on Zen, you don't really need to do that. I was eyeing an upgrade from my 3700x to a 5950x with expected price drops on zen4 launch, and then run it at lower PPT/... - google "5950x eco mode" for some results what this gives.
However, given that AMD claims the 7950x would have 70+% more performance at 65w TDP than the 5950x, I'm tempted to actually replace the mobo/ram and go for that instead.
@AnandTech: not sure if it has been mentioned in the comments yet, but AMD has disclosed the on-package graphics for all of the Ryzen 7000 series. It looks like the RDNA-2 GPU is labeled as just "AMD Radeon Graphics" for now, all four SKUs have a base clockspeed of 400MHz with a top speed of 2200MHz. All have the same graphics config: just 2 CUs, for 128 Stream Processors. This means it should also have 2 Ray Accelerators.
How does this compare with the last decade's desktop graphics cards? e.g., 10-year-old mid-range desktop Radeon 7750? or current low-end Radeon 6400? Or something in between?
The media should sanction companies for publishing useless "Up to" performance gains and not repeat any of those numbers in their reporting. A CPU can be slower than another one on average but still be "up to" 50% faster in a single specifically designed benchmark.
Other than that, looking forward to the reviews. Please run benchmarks in ISO TDP. Given the extremely high electricity costs at the moment I am not interested in running my CPU at too high frequencies with low efficiency. Interested to see average performance of 3600x vs 5600x vs 7600x at 65W TDP.
Examining PC Tech media coverage of Zen 4 intro, whether web or You Tube, its primarily mimicking repeater of AMD positioning which suggests AMD is utilizing an access system rewarded for parroting AMD and punished for not parroting AMD. Don't tow the line and get the boot essentially. What I'm amazed at is no media outlet covered that between the ccx and 6 nm i/o the silicon area in Raphael has not changed from Vermeer. Not one PC tech media outlet touched this one that keeps AMD PR pleased? Ot at bay? mb
‘utilizing an access system rewarded for parroting AMD and punished for not parroting AMD.’
Why is that important even if true, unless you’re a stocks jockey?
AMD will sell what it sells. So will Intel. Benchmarks will continue to be done over the lifespans of products. Today’s hot tech will soon be yesterday’s old hat.
Perhaps it’s my age but until I see something actually exciting, rather than bland ‘give consumers the minimum’ iterations, I’ll refrain from caring much about how independent the press is.
Even the kneecapping of Zen 1 and 2 with JEDEC is a self-inflicted wound on AMD’s part. Yawn.
Let’s see a consumer-priced wafer-scale GPU. Something actually interesting.
Raphael price from TSMC to AMD is the same as Vermeer, maybe $10 less at + 20% cost + 30% good die, however, AMD price to high volume OEMs will be the same as Vermeer. $100 MSRP cut on Raphael 16C and 8C from Vermeer is purely an AMD PR ploy. The channel will price Ryzen 7K accordingly on performance out the gate. The channel will rely on R7x Extremes price premium above MSRP to pay for channel's AMD Rembrandt, Cezanne primary production overage stuck on the shelf subject financial loss offset. After the channel protects against its AMD mobile inventory financial losses it can start covering its Intel Tiger mobile overage on Raphael and Raptor price premiums > MSRP. Its the way the world works and this is AMD and Intel's fault not the channel! mb
Provided this reassessment up comment stream on NEW die area data;
Vermeer in relation to Raphael. V5x normalized area across the grade SKUs over the full run with its i/o die = 243 mm2. R7x normalized on V5x full run percent grade SKU split = 239 mm2 near same. Which is interesting NEW data because I originally estimated + 20% cost and + 30% good dice on 5 nm normalized ccd and the i/o dropping to 171 mm2 of area IT DID NOT and the reason is the 6 nm i/o + iGPU is no less area than the 12 nm i/o.
There seems to be disagreement about the area of the 5 nm ccd. Some report 77 mm2 and others report 71 mm2 a less 6.8% difference. At R7x 5 nm 77 mm2 ccd is 8.4% smaller than V5x 7 nm 84 mm2.
Whether V5x or R7x an 80% yield on edge and area defects delivers roughly the same number of normalized dice sufficient for 230 to 255 produced components per wafer.
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
105 Comments
Back to Article
Cooe - Monday, August 29, 2022 - link
I never want to see another post about how "AMD never sandbags!!! Why would a company ever do that???"... -_- ... They do this crap for basically every single major launch. Sandbagging is practically in AMD's fundamental marketing DNA at this point.Dustin Sklavos - Monday, August 29, 2022 - link
You say that like it's a bad thing. It's a great marketing decision; usually marketing benchmarks and the like are heavily cherry picked and to be taken with a grain of salt. With AMD, they're generally not advertising anything they can't at least meet if not beat. We're so used to the usual marketing it seems like you're getting actively offended when a company tries to be honest.nandnandnand - Monday, August 29, 2022 - link
If you believe AMD, recent changes improved IPC by about +2% over the previous figures. Maybe the improvements were made after they were already being accused of sandbagging.Also, because clock speeds and TDPs play a bigger role in these performance gains, a single number doesn't represent both the 7600X and 7950X very well.
del42sa - Tuesday, August 30, 2022 - link
over 2% ? they say ZEN4 desktop IPC is 8 % at AMD’s 2022 Financial Analyst Dayhttps://cdn.mos.cms.futurecdn.net/euNqMmugdfyVHe9q...
nandnandnand - Tuesday, August 30, 2022 - link
They said 11% at Computex according to Anandtech, and the screencap of this is floating around.del42sa - Tuesday, August 30, 2022 - link
where they said that ? Show me !nandnandnand - Tuesday, August 30, 2022 - link
"The combination of which will allow AMD to deliver what they are saying is now a 13% increase in IPC over their Zen 3 architecture – up from an 11% claim as of Computex"Cooe - Wednesday, August 31, 2022 - link
That's straight up a misquote/misrememberence by Ryan. It was listed as "+≈8-10% IPC" on the official "Zen 4 Core" slide from Computex. Stop regurgitating Ryan's mistake as facts without you ACTUALLY fact checking it yourself. Makes you look kinda like a self obsessed, tech illiterate imbecile. 🤷♂️del42sa - Wednesday, August 31, 2022 - link
don´t post the sentence , whichmakes no sense anyway... post the slide from AMDnandnandnand - Wednesday, August 31, 2022 - link
https://c0.lestechnophiles.com/images.frandroid.co...del42sa - Saturday, September 3, 2022 - link
ha ha , you´re imbecil. I´t not IPC figure :-Dthestryker - Monday, August 29, 2022 - link
I'll look forward to the reviews, but given their pricing it looks like my inclination to wait until '23 and MTL/Zen5 is the right one. I'm looking for something in the $400 ish range and this pricing just looks absurd compared to where 12th gen Intel is and launched.Perhaps they really will maintain enough of an advantage, or Intel will raise their prices enough to make this make sense. At the lower end this just doesn't seem even remotely possible given that there isn't a DDR4 option to keep RAM prices lower.
jrocket - Monday, August 29, 2022 - link
Compare the price of 12900k vs 7600x. have similar performance. AMD is half the price and overall the better option here.nandnandnand - Monday, August 29, 2022 - link
Raptor Lake is the elephant in the room.DannyH246 - Tuesday, August 30, 2022 - link
How so? Everyone knows Raptor lake is coming. It will be a 5% increase at best. AMD will win most benchmarks against Raptor and at a cheaper price (but with higher margins), and consume less power. Then in about 3 months along comes Zen4 X3D to completely take the win in all benchmarks. AMD firmly has Intel in its back pocket. But go ahead if you want to buy Raptor it’s your choice. AMD will completely sell every Zen4 die it makes.nandnandnand - Tuesday, August 30, 2022 - link
Not 5%. More like +10-15% single-threaded performance for a 13900K vs. 12900K, and 12900K is the chip AMD wanted to compare to. And the period of time between Zen 4 and Raptor Lake launches will be short. You have no idea about pricing.DannyH246 - Tuesday, August 30, 2022 - link
Intel have already publicly stated they will need to increase prices by 15%-20% due to inflation & supply chain issues etc. Meanwhile the 7950X is already $100 cheaper than the 5950X when it was released. It is also no secret that due to the die size and usage of chiplets AMD have much more pricing freedom than Intel. Also the period of time between Raptor Lake & X3D will be short. You clearly have no idea about anything.thestryker - Tuesday, August 30, 2022 - link
Intel has absolutely not said anything about what products will have higher prices nor what that increase will be just that there will be increases.The 7950X is still around $100 higher than the 12900K launched at which gives Intel plenty of room to increase the price for the 13900K and still be cheaper than AMD.
AMD has said X3D for datacenter is coming 1H23, but anything regarding desktop timeline are rumors. There's no doubt it will arrive as it'll be fantastic for gaming, but since Zen 5 is 2024 they may hold onto it longer as a MTL competitor rather than RPL if Zen 5 is later in the year.
Strawman1234 - Tuesday, August 30, 2022 - link
Raptor Lake has more E cores, by not moving to a smaller node the die is going to be bigger. Current leaks have it at 257mm^2 a good 50mm^2 more than Alder Lake. That means less chip per wafer, less yields and harder to bin for faster parts. Intel will have to increase prices for sure just on the fact that Raptor Lake uses more silicon. If Raptor Lake is to be made in the US, it will cost even more. TSMC founder have stated this year, that chips made in its Oregon fab cost 50% more than chips made in Taiwan.thestryker - Tuesday, August 30, 2022 - link
There is definitely more silicon being used, and likely through the entire product stack if the leaks are accurate. This does not actually mean proportional cost increases as there have undoubtedly been yield improvements on the Intel 7 node over its lifetime.As for clockspeed I'd point you to the Chips and Cheese analysis of the ES RPL part they did cache testing on. They explain a bit of the architectural changes that should allow it to hit higher clocks.
Lastly never believe anything out of the mouth of someone with a vested interest against what they're talking about.
Strawman1234 - Tuesday, August 30, 2022 - link
Yield improvements will never be as good as a chiplet design. And yes I would agree with you that the TSMC does have a vested interest to keep production in Taiwan. But the US does have higher labor cost and it is further from the supply chains. Intel also has less revenue stream to recoup its development cost on Intel 7 node (10nm). The battle this gen will be won on price and availability, real world performance will be too close to matter between Zen4 and Raptor Lake.Bruzzone - Wednesday, August 31, 2022 - link
Intel fabrication cost at iSF10/7 is in parity with TSMC 7 nm. Intel 7 at run down will be in parity with TSMC 5 through its production volume peak after which TSMC 5 will have the run down advantage. mbBruzzone - Wednesday, August 31, 2022 - link
Where TSMC gains on Intel cost wise is of course legacy fabrication and for products priced less than Intel variable costs range $39/$79/$109 varies on product lines. Intel contracting out its own products priced less than variable cost works. This is also why Intel never produced StrongARM because its production cost structure subject variable cost would have made SA a loser. Bay Trail was similar on so said contra revenue producing at a loss and hoping to make up in the future never happen. mbBruzzone - Wednesday, August 31, 2022 - link
Silicon area in Raphael subject [including] the 6 nm i/o + iGPU block appears the same as Vermeer; measure the die in the photo. Raphael costs the same as Vermeer on a per unit basis at + 20% cost + 30% good dice. Vermeer price from TSMC to AMD $165 and Raphael is the same, maybe < $10 but there are R7x premium price voids missing and those SKUs as a price support will increase cost on average across the full line. mbThe AMD Raphael advantage is + 30% dice = less total wafers that can be redirected to other products. mb
Dante Verizon - Wednesday, August 31, 2022 - link
5nm costs twice as much per waffer than 7nm. .. there is no way your calculation is rightBruzzone - Thursday, September 1, 2022 - link
"5nm costs twice as much per wafer than 7nm". At risk production.Moore law doubles transistors every next node. Rocks law doubles cost every next node and when reaching peak production volume, theoretically doubling volume, cost equals parity with the prior node, then through run down volumes the cost continuously drops to marginal cost at run end; the very last unit produced. Profit maximization produces until the very last unit's revenue value = cost and that's run end.
Vermeer in relation to Raphael. V5x normalized area across the grade SKUs over the full run with its i/o die = 243 mm2. R7x normalized on V5x full run percent grade SKU split = 239 mm2 near same. Which is interesting NEW data because I originally estimated + 20% cost and + 30% good dice on 5 nm normalized ccd and the i/o dropping to 171 mm2 of area IT DID NOT and the reason is the 6 nm i/o + iGPU is no less area than the 12 nm i/o.
There seems to be disagreement about the area of the 5 nm ccd. Some report 77 mm2 and others report 71 mm2 a less 6.8% difference. At R7x 5 nm 77 mm2 ccd is 8.4% smaller than V5x 7 nm 84 mm2.
Whether V5x or R7x an 80% yield on edge and area defects delivers roughly the same number of normalized dice sufficient for 230 to 255 produced components per wafer.
On V5x full run by grade SKU data the average weighed $1K price for R7x is $466 / 3 way stakeholder (TSMC/AMD/OEM) split = TSMC price to AMD at $155 and V5x was $165 so within $10.
AMD markup dependent OEM volume is * 1.5 to * x2 so a full line procurement mirroring V5x full run at 16C = 19.22%, 12C = 23.01%, 8C = 32.67%, 6C = 25.09% = range $233 to $310 per unit across 10 1st tier customers across 60 M unit production run means ten 6 M unit procurements to get the $233 to $310 price per component on the grade split noted.
OEMs are brokers and they will resell whatever overage they don't need or SKUs they don't want lowers their procurement cost. This is how AMD and Intel distribute to the secondary market beyond AMD direct master distributors.
The cost : price / margin of Raphael is the same as Vermeer. That's your price increase from 7 nm to 5 nm = no cost savings.
If an efficient TSMC production regime they are not always that way on TSMC effective method that flattens change in quantities produced the efficient first 5% of production volume will cost $155. At 20% of the run marginal cost to produce efficiently drops to $45, at 65% pops to $55 on peak volume, at 88% complete this is all about sort now marginal cost drops to $9 and the last 12% cost drops to $3. TSMC full run marginal cost to produce the complete component on a 5 phase run = $41.27
If TSMC increases efficiency by producing more volume between phase 2 and phase 3 peak production the cost can drop measurably.
Unless AMD negotiates for a percent of this phased volume cost decrease, AMD won't see a penny of it all going to TSMC fab and the package house.
You're 5 nm price increase is no cost savings in relation 7 nm, not double just flat.
As I collect R7x supply volume data I'll be able to perform a more precise cost : price / margin assessment
mb
Qasar - Tuesday, September 6, 2022 - link
wow thats a lot of BS to post, that no one can prove, OR verify.Qasar - Tuesday, September 6, 2022 - link
of course is calculations arent right , but because he posts NO sources to ANYTHING he posts, how do does he prove it ? is the way this guy works. he just posts a bunch of BS, and fluff. the guy is a flake and a fraud.just try to ignore his spam posts.
schujj07 - Tuesday, August 30, 2022 - link
The comparison to the 12900k is due to the fact that the 13900k IS NOT AVAILABLE. Tell me how could a rival company compare their new product to a rival's that hasn't been released?nandnandnand - Tuesday, August 30, 2022 - link
I didn't say that they should. All I said is that Raptor Lake is the elephant in the room and y'all sperged. If Raptor Lake launches on October 20, that's not even a full month after Zen 4 CPUs.schujj07 - Tuesday, August 30, 2022 - link
While you said it is the elephant in the room, you also said "12900K is the chip AMD wanted to compare to." I was pointing out that in no way is AMD able to compare their Zen 4 CPU to Raptor Lake when Raptor isn't even out yet. Therefore they did the best they could and compared it to the current best from Intel.thestryker - Monday, August 29, 2022 - link
They might have similar performance in gaming/lightly threaded, but they most certainly won't in multithreaded workloads.Also as nandnandnand said you absolutely cannot ignore RPL coming by the end of the year.
Gothmoth - Tuesday, August 30, 2022 - link
for games zen4 with 3d cache will probably beat raptor lake.and for content creator like myself. i have too look at power draw in these times.
i have 8 systems. mostly threadrippers and 3x 5950x. the energy bill is pretty high.
if AMD can deliver 90% of the performacne at 50-60% the energy draw... i know what i am buying.
thestryker - Tuesday, August 30, 2022 - link
Whenever it comes out, sure the v-cache 7000 will absolutely take the gaming crown, but the poster I was responding to was acting like the 7600X was faster than the 12900K period and it won't be.Absolutely if you run your hardware at the edge all the time things like power draw certainly become infinitely more important. I'd make the same decision in your shoes as it's definitely the right one, but for me where I do mostly lighter threaded with periodic all core max performance for both is what I'm looking for.
PixyMisa - Monday, August 29, 2022 - link
Assuming availability and street prices are reasonable, my next system will be a 7950X. That price cut to $699 means that if you can use the power there's no reason not to go all the way.meacupla - Monday, August 29, 2022 - link
I look forward to the 7xxxX3D version.(But mainly because I don't want to be an early adopter for DDR5)
brucethemoose - Monday, August 29, 2022 - link
This ^And I hope they have some X3D SKU at a non-flagship price.
haukionkannel - Tuesday, August 30, 2022 - link
3d cache seems to increase price about $150 for single chiplet, so about $300 for dual chipset CPUs...It will always be flagship priced product! Otherwise they would put it on all their CPUs, but it does not make sense, because it is expensive and reduce most productivity workloads, unless AMD has made magic and cache does not heat the cpu below at all...
And extra $150 to $300 CPU sound worse than extra $150 to $500 CPU... It just does not make sense in low cost products!
Bruzzone - Wednesday, August 31, 2022 - link
It cost TSMC $45 to add the 32 mm2 64 MB SRAM to 5800X, But by the time 3D gets to the end buyer price reflects the added margin AMD to OEM, OEM to End Sales. mbRomanPixel - Friday, September 2, 2022 - link
I third this. I'm waiting for the X3D as a *gamer*. By the time the X3D CPUs come out, the B series mobos might be out, what NVIDA & AMD got for new gen GPUs will be revealed or also released, DDR5 pricing might drop a bit, Intel will have released their 13th gen CPUs for competition (GPUs too?), and more ATX 3.0 power supply options will have dropped for "FuTuRe PrOoFiNg."Dante Verizon - Monday, August 29, 2022 - link
OMG.. 32% faster on dolphin at same clockrate. lolKendog52404 - Monday, August 29, 2022 - link
I would be interested in how the two lowest options (7600X & 7700X will compare to the Ryzen 7 5700G, in performance and graphics.xol - Monday, August 29, 2022 - link
No optimistic/holding my breathTheir iGPU competition from Intel has around 0.5TFLOPS (eg UHD 770 as in 12500 and up which is plenty for desktop) - I doubt they'll try to beat it (much), and not come close to the 2TFLOPs of their best APUS.
thestryker - Monday, August 29, 2022 - link
It's 2 RDNA2 cores, so it liable to be even worse than Intel's desktop processor IGP.schujj07 - Tuesday, August 30, 2022 - link
The RDNA2 iGPUs in the G series are better than anything Intel has. What we will see in Zen 4 though are iGPUs designed around basic desktop functionality. Basically you will be able to put a non G CPU into a business desktop and not need a dGPU.boozed - Monday, August 29, 2022 - link
They're going to easily beat the 5700G. Even the 5700X easily beats the 5700G because of its smaller L3 cache.phoenix_rizzen - Tuesday, August 30, 2022 - link
OP is talking about the iGPU side of things, whether the RDNA2 cores on the 7600X/7700X would be competitive we're the Vega cores on the 5700G.boozed - Tuesday, August 30, 2022 - link
Quite right, I missed that bit. It sounds like it's going to be pretty basic.TeXWiller - Monday, August 29, 2022 - link
The Zen4 based APUs were previously suggested coming later. There will probably be 65W options as well.haukionkannel - Tuesday, August 30, 2022 - link
Most likely yes!mcnabney - Monday, August 29, 2022 - link
The bottom line issue is that there is a $320 price premium to go with a Ryzen 5 7600 with 32GB of DDR5 6400 CL20 over a Ryzen 5 5600 with 32GB of DDR4 3200 CL16.If you are on a budget that $320 can buy a whole lot more GPU.
Hossein - Monday, August 29, 2022 - link
What about DDR5 6000 CL40?schujj07 - Tuesday, August 30, 2022 - link
Your comparison makes no sense at all. You are using the prices for top of the line DDR5 but comparing it against every day DDR4. When you use every day vs every day level RAM, the difference is about $80 more expensive for the DDR5. Those prices will be coming down as more DDR5 hits the market. Plus DDR5 has the advantage that companies will use higher density chips to make it so prices will drop even further.Jp7188 - Saturday, September 3, 2022 - link
6400 CL20 !? What in the world?Koumkouat - Monday, August 29, 2022 - link
Ddr5 kits with xmp will work on am5 motherboards?phoenix_rizzen - Tuesday, August 30, 2022 - link
According to the article on Ars Technica XMP is supported as well as EXPO.JayNor - Monday, August 29, 2022 - link
It is amazing that TSM N5 has made such a big jump in the clock rate with this generation.The Gracemont cores are 1/4 the size of Golden Cove cores, which makes them 1/2 the size of the zen4 cores.
The Golden Cove cores have dual avx512 units on them, at least they do on the SPR server chips.
What's happening with the SPR HEDT chips?
Mikewind Dale - Monday, August 29, 2022 - link
> So while all of the Ryzen 7000 chips come with an iGPUDoes this mean they won't support ECC UDIMMs anymore? I hope that's not the case, because the AM4 Ryzen provided a great way to get ECC support without going with an expensive Xeon or ThreadRipper.
Silver5urfer - Monday, August 29, 2022 - link
First - Die Size comparison is kinda not perfect, because Intel uses 30% of the ADL P core Golden Cove for AVX512 transistors which they disabled on the 12900KS and some of the newer batches probably because they are run by a bunch of stupid bean counters.Two - 7600X vs 12900K is silly, because a damn 12900K is not just used for stupid gaming, because a 12700K can deliver 95%+ performance of that processor. And the MT of 12900K will be higher than 7600X despite the crappy biglittle junk.
Now that is out the major improvement I'm seeing is superior Clock speeds. 5.7GHz on an AMD chip is insane. That's super high mark. I bet Intel Raptor Lake will have 5.8GHz to win the Clock speed battle but the 8P cores of RPL might be super fast in gaming for 13900K paired with L2 improvements and new IMC. Although for MT I think 7950X will dominate that new RPL chip.
EXPO is looking interesting, instead of the absolute nonsense IMC on Ryzen until 5000 this time they have a solid low latency on DDR5 plus a real standard of Memory profile like XMP. That's really means the IMC on Ryzen is really a big improvement.
The major aspect is AM5 until 2025+ that's a win, AMD is going to saturate the market again like AM4 but the stability and the quality of the BIOS and CPU is up in the air, AMD failed to get me as a customer because their CPUs cannot be tuned properly due to lack of datasheets and lack of fixed clockspeed, they won't hold Boost as well. Next the IMC was a mess. IOD was a disaster, as even today we have this WHEA / USB issues on the platform it won't be fixed. So AMD has to prove it. However I think their OC is still a dud with Zen 4 too, the Clockspeed is way too much I bet they are not going to leave anything esp the fact that Zen 4 is a clear cut derivative of Zen 3 just like Intel's Comet Lake 10th Gen being the apex position for Skylake design maxing on Cache / Clocks / Core count. Same for AMD's Zen 4 vs their rest except for the core counts.
I think X3D is going to come next year maybe for 2-3SKUs ? 7950X3D / 7800X3D ?
All in all somewhat very interesting timeline seeing this high competition. Shame how Intel is succumbed to their biglittle junk. Good thing is AMD Zen 4 will destroy those BGA ARM trash which is still MIA and also Apple's nonsense as well in one shot. M series lost vs ADL Zen 4 will destroy it even more.
Silver5urfer - Monday, August 29, 2022 - link
I forgot to mention ILM, why did AMD not mention any of it ? It would have been a solid move esp the fact how LGA1700 is plagued by that Intel's factory design failure. AM5 socket has extra reinforcement as per Igor's Lab, it's a no brainer over Alder Lake unless you want more tuning and fun as AMD processors won't have those features, I doubt they will add more since the Voltage will be sky high on this N5 refined for High performance more like High K Metal gate of older nodes from Intel and TSMC. Zen 3 was 1.4v on all X3D got binned due to low heat dissipation of Cache Stack needed a good silicon.All in all Intel's move is next esp on the ILM side I will be curious if they care or gonna dump the LGA1700 along (It's going EOL, RPL will be last) with it's failed ILM quality.
lemurbutton - Monday, August 29, 2022 - link
> Good thing is AMD Zen 4 will destroy those BGA ARM trash which is still MIA and also Apple's nonsense as well in one shot. M series lost vs ADL Zen 4 will destroy it even more.On dear. 170w boosted to 230w. Amazing. I thought 5nm meant that Apple can match the M1 in efficiency. Not even close.
Silver5urfer - Monday, August 29, 2022 - link
Nobody cares about a BGA use and throw dumpster class machine vs a Desktop Socket with PCIe slots and massive PCH I/O. Apple can release MX on 1nm or whatever they want, it will always be inferior to an LGA / PGA socket machine. The OS itself is crippled to death, constant updates and death of all old Software due to 32bit abandonment. Who even cares about this Apple nonsense ? The market is 12% tops for that Mac OS. If you even compare BGA junk from AMD and Intel it will also render the Apple useless, esp the fact that they have affordable pricing with user optional PCIe SSDs and SODIMMs and 144Hz displays.mdriftmeyer - Tuesday, August 30, 2022 - link
This is a bottom feeding load of shite. Seriously, 32 bit abandonment? Linux and Windows have abandoned that as well. Toss a VM and you get 32 bit back. UNIX on OS X isn't crippled one bit. You want extra features you've got options to get them back.I would have personally hoped Apple abandoned Intel for Mac Pro with Zen 3.x but that never happened.
Zen 6 at the latest will introduce a unified memory back plane ala Apple, multiple FPGA based Engines and more ala Apple, a multi-chiplet integrated GPU and more.
By then RDNA 5 will most certainly not remotely resemble today's RDNA 3.0 and most board manufacturers will cut back on PCIe slots as unnecessary. PCIe 7.0 will be coming out and the bandwidth will be massive. Apple still uses PCI if you forgot, how else do you think TB3 interacts?
I love the OS updates. Never a broken system. Linux on the other hand breaks constantly. Windows is irrelevant in this conversation.
Silver5urfer - Tuesday, August 30, 2022 - link
You are smoking some high tech crack. 32Bit means not 32Bit OS smart pants, means 32bit applications. Mac OS abandoned it, first understand before you go off hard. Windows is as relevant as it gets, the OS has highest backwards compatibility on the planet. Windows 7 still gets ESU updates and it's going upto 2025 when Windows 10 gets it's 10 year policy of EOL done, that said LTSC 2019 will have support upto 2030 since it has 10 year update since release. Meanwhile Jokebook Pros all are trashed out with failing KBs, Heaitng issues, Lack of I/O ports forget PCIe expansions. Then the crap 60Hz garbage decade old panels selling for super premium.The actual bullshit is yours from the Zen 6 part.
OS X is not equal to Unix, nope. Not even a bit. Zen 6 and all that nonsense from where did you get that information ? More Apple More Apple. Just refrain from this nonsense here, just throwing up in the air and see what sticks to asset your fallacy and weak arguments. Everything is ala Apple crap. What a damn joke lol. And more boards will remove PCIe slots rofl yeah did ASUS tell you that or AMD leaked that information. And now the cope is Thunderbolt 3, yeah sure like I love to haggle a stupid BGA pile of junk with a colostomy Thunderbolt 3 outdated junk dock, USB 4 beats even Thunderbolt 4 because TB4 doesn't have a lot of bandwidth it's just essentially 3 but with more stringent limits to stop OEMs to skimp on bandwidth, even Z590 boards have TB4 Maple Ridge controllers so this Mac Junk is severely limited on all fronts - SW, HW, DIY end of story.
dendem - Wednesday, August 31, 2022 - link
Silver5urfer: "OS X is not equal to Unix, nope. Not even a bit."UNIXtutorial.org: "macOS is the second most widely used desktop operating system after Microsoft Windows. It is also a desktop OS that has UNIX 03 certification, meaning it’s a truly Unix OS rather than Unix-like OS" https://www.unixtutorial.org/software/macos
dendem - Wednesday, August 31, 2022 - link
Silver5urfer "OS X is not equal to Unix, nope. Not even a bit. "The Open Group: UNIX® Certified Products
• Apple Inc.: macOS version 12.0 Monterey on Apple silicon-based Mac computers
• Apple Inc.: macOS version 12.0 Monterey on Intel-based Mac computers
• IBM Corporation: AIX version 7, at 7.2 TL5 (or later) on systems using CHRP system architecture with POWER™ processors
• IBM Corporation: AIX version 7, at either 7.1 TL5 (or later) or 7.2 TL2 (or later) on systems using CHRP system architecture with POWER™ processors
• Hewlett Packard Enterprise: HP-UX 11i V3 Release B.11.31 or later on HP Integrity Servers
• Huawei Technology Co., Ltd: Huawei EulerOS 2.0 on Huawei KunLun Mission Critical Server
• IBM Corporation: AIX 6 Operating System V6.1.2 with SP1 or later on Systems using CHRP system architecture with POWER™ processors and 2, 8 or 128 port async cards
• IBM Corporation: AIX 5L for POWER V5.3 dated 7-2006 or later on Systems using CHRP system architecture with POWER™processors
• IBM Corporation: z/OS V2R1 or later with: z/OS V2R1 or later Security Server and z/OS V2R1 or later C/C++ Compiler on IBM zSeries Processors that support z/OS Version 2 Release 1 or later
• The SCO Group, Inc.: UnixWare ® 7.1.3 and later for single and multiprocessor systems based on IA-32 and compatible processors and conforming to PC architecture
• The SCO Group, Inc.: SCO OpenServer Release 5 and OpenServer Release 6 on Single and Multi-processor Industry Standard Intel architecture platforms
https://www.opengroup.org/openbrand/register/
'Not even a little bit', Silver5urfer?
Silver5urfer - Wednesday, August 31, 2022 - link
Better luck next time, that is a strawman as it doesn't have literally anything except the UNIX omg it's the certified OS !! If anyone wants to use UNIX OS they will use Linux flavors not that closed source garbage OS called Mac OS which runs on a disposable Hardware vs Linux running on Pi or even iPod Classic.max - Saturday, September 3, 2022 - link
Don't feed the troll. Just let him live in his own Apple hater world. Thank U :)bji - Tuesday, August 30, 2022 - link
You don't need to claim falsehoods to try to enhance your point. Linux does not "break constantly".Manabu - Saturday, September 3, 2022 - link
Just last week my arch startup broke and I had to chroot from another image to fix it. And that was updating a few days after they discovered the bug, that AFAIK is still not corrected needing manual intervention: https://old.reddit.com/r/archlinux/comments/x1lcoi...Dante Verizon - Wednesday, August 31, 2022 - link
I agree, and most who use apple products are not able to understand this..GeoffreyA - Tuesday, August 30, 2022 - link
"7600X vs 12900K is silly"I think what they were trying to show is that even their "small" CPU beats, or matches, a top one from Intel. "Look, Ma, even our little brother beats the competition!"
Jimbo123 - Monday, August 29, 2022 - link
AMD can claim to have the fastest gaming machine, before Raptor Lake starts, after that, it will be the second Fidler again.Khanan - Tuesday, August 30, 2022 - link
Unlikely and even if the X3D parts will destroy intel easily. Bad times for intel fanboys.haukionkannel - Tuesday, August 30, 2022 - link
Until 7800x3d is released... That is how it goes. And after that Intel will release something faster. That is how the merry carrousel goes :)Khanan - Wednesday, August 31, 2022 - link
Yes 1-2 years later. The X3D chips will be unbeatable for a good while.meacupla - Tuesday, August 30, 2022 - link
I am hoping Intel 13 series remains competitive.Pricing wars please.
Khanan - Saturday, September 10, 2022 - link
Intel simply can’t match AMD specialized 3DV tech. This time intel has to contend against both new arch and the extra cache, suffice to say Intel can’t contend with their traditional tech which is simply not good enough in this case (gaming). And for multi cpu, I don’t think 24 cores are that relevant, anyone who really needs more than 16 can go and buy Threadripper, which is also easily faster than Intels stuff.[email protected] - Monday, August 29, 2022 - link
I was looking at that R7 7700X and it's MSRP is just slightly higher than I paid for my R7 5800X. I suspect that by Black Friday, we'll be seeing some good deals on this CPU. As much as I'd love to upgrade, it's only been about a year since my system was upgraded, so sadly I'll be sitting out this upgrade.Khanan - Tuesday, August 30, 2022 - link
No 5800X msrp was 450$ and 7700X is 399$ which resembles the msrp of the 3800X. But I should say it’s not the direct successor, so this is a bit misleading.haukionkannel - Tuesday, August 30, 2022 - link
It is in many way. 5700x was 65w cpu, 5800x was 105w. So 7700x is definitely direct successor. They most likely are saving 7800 for 7800x3d...Khanan - Wednesday, August 31, 2022 - link
No, TDP was generally increase for Zen4, so this has a different reason. 7800X could have a even higher TDP.dizzynosed - Tuesday, August 30, 2022 - link
I'am still OK with my [email protected] . Skipped DDR4 entirely and will go with Zen3 or Raptor DDR5. But the prices are rising steeply, 12700 got from 340Eur to 405Eur in 2 months.nandnandnand - Tuesday, August 30, 2022 - link
Zen3 meaning what, desktop Rembrandt? You won't be skipping DDR4 if you pick Zen3 Vermeer.dizzynosed - Wednesday, August 31, 2022 - link
I meant Zen4 of course. Typo.Nfarce - Tuesday, August 30, 2022 - link
I don't know how in the world you get by on such an old architecture for today's applications, hardware, and security needs. I had an i5 2500K 4.7GHz overclock gaming build up to 2017 when it started bottlenecking my new RTX 1080 Ti GPU (still have it as a Win7 offline only retro gaming PC with the GTX 970 it was last tied to). Hard to believe that LGA 1155 chipset is over 10 years old now.phoenix_rizzen - Friday, September 16, 2022 - link
It all depends what you use it for. I have an i5-2600 (I think, might be 2500) laptop that I still use for work. Has 16 GB of RAM, an SSD, and a new wifi module. Ran Windy for years, and upgraded to Win10 easier this year without issues.Used mainly for network troubleshooting, serial console for switches and servers, remote access to servers via IPMI and SSH, and some word processing.
Obviously it's not my primary workstation, but it serves it's purpose.
James5mith - Tuesday, August 30, 2022 - link
"Driving AMD’s gains in this newest generation of desktop CPUs is a combination of architectural improvements underpinning the Zen 4 architecture, as well as moving production of the CPU core chiplets to TSMC’s leading-edge 5nm process."No. Driving AMD's gains are absurd power draw increases for this generation.
schujj07 - Tuesday, August 30, 2022 - link
Well 13% average IPC increase means depending on application you will get that much more performance AT THE SAME clock speed, and lower power due to 5nm, as Zen 3. AMD also showed that Zen 4 was more efficient than Zen 3 at all TDP levels. Take that with a grain of salt as it was 1st party information but once we have the benchmarks we will know for sure.Khanan - Tuesday, August 30, 2022 - link
Somebody didn’t understand the article or presentation. Try to readGothmoth - Tuesday, August 30, 2022 - link
i will sure buy a 7000x at some point. but this time im not jumping into the first generation of mainboards. not that i had bad experiences. my first crosshair x370 was pretty great. but with a 5950x i can wait a bit before i upgrade that system. i guess by april next year there will be new mainboards with even better specs for a lower price. even if raptor lake beats zen4 by 10% ..... with the rumored power draw raptor lake needs i stick to AMD.Khanan - Tuesday, August 30, 2022 - link
I highly doubt you can compare 7th gen to first gen Ryzen. Just because it’s a new platform doesn’t make all their experience and advances go away. So no it will probably function smoothly.Threska - Tuesday, August 30, 2022 - link
https://wccftech.com/roundup/amd-x670e-x670-mother...A lot of them look like they're built off of the X570 designs, with maybe a little more bling (that's a good thing). Features seem to be basically "more of the same" e.g. Wifi 6E, 2.5G LAN, etc. Some do look affordable even if most are TBA on pricing.
Khanan - Saturday, September 10, 2022 - link
The boards are vastly more complex due to PCIE 5.0 traces. You can’t see it but it’s there.IBM760XL - Tuesday, August 30, 2022 - link
I'll be curious to see some reviews that include actual power efficiency, i.e. how much power is required to finish the same unit of work, versus Zen 3. And if it's easy to turn the maximum TDP in the motherboard, both 65W and 105W figures for Zen 4 octo-cores (and/or 105/170 for dodeca-cores) would be great. Just knowing how easy it is to configure that TDP would also be great.Plan A is still to upgrade the old Sandy Bridge to Zen4 + AM5, but the decision to chase clockspeeds and increase TDP has me tempted to go with a potentially second-hand Zen 3 option instead, especially as even brand new the prices on those are rather tempting.
Spoelie - Thursday, September 1, 2022 - link
Given the configurable TDP on Zen, you don't really need to do that. I was eyeing an upgrade from my 3700x to a 5950x with expected price drops on zen4 launch, and then run it at lower PPT/... - google "5950x eco mode" for some results what this gives.However, given that AMD claims the 7950x would have 70+% more performance at 65w TDP than the 5950x, I'm tempted to actually replace the mobo/ram and go for that instead.
NextGen_Gamer - Tuesday, August 30, 2022 - link
@AnandTech: not sure if it has been mentioned in the comments yet, but AMD has disclosed the on-package graphics for all of the Ryzen 7000 series. It looks like the RDNA-2 GPU is labeled as just "AMD Radeon Graphics" for now, all four SKUs have a base clockspeed of 400MHz with a top speed of 2200MHz. All have the same graphics config: just 2 CUs, for 128 Stream Processors. This means it should also have 2 Ray Accelerators.cjcoats - Wednesday, August 31, 2022 - link
How does this compare with the last decade's desktop graphics cards?e.g., 10-year-old mid-range desktop Radeon 7750? or current low-end Radeon 6400? Or something in between?
thestryker - Wednesday, August 31, 2022 - link
6400 is 12 CUs this is 2. It will be about the same as Intel's 32 EU desktop IGP.Bruzzone - Tuesday, August 30, 2022 - link
Who's measured the 6 nm i/o + graphics area, seems around 130 mm2? . . .in relation 77mm2 ccx looking at the picture. Please advise. mbdonquixote42 - Wednesday, August 31, 2022 - link
The media should sanction companies for publishing useless "Up to" performance gains and not repeat any of those numbers in their reporting. A CPU can be slower than another one on average but still be "up to" 50% faster in a single specifically designed benchmark.Other than that, looking forward to the reviews. Please run benchmarks in ISO TDP. Given the extremely high electricity costs at the moment I am not interested in running my CPU at too high frequencies with low efficiency. Interested to see average performance of 3600x vs 5600x vs 7600x at 65W TDP.
Bruzzone - Wednesday, August 31, 2022 - link
Examining PC Tech media coverage of Zen 4 intro, whether web or You Tube, its primarily mimicking repeater of AMD positioning which suggests AMD is utilizing an access system rewarded for parroting AMD and punished for not parroting AMD. Don't tow the line and get the boot essentially. What I'm amazed at is no media outlet covered that between the ccx and 6 nm i/o the silicon area in Raphael has not changed from Vermeer. Not one PC tech media outlet touched this one that keeps AMD PR pleased? Ot at bay? mbOxford Guy - Monday, September 5, 2022 - link
‘utilizing an access system rewarded for parroting AMD and punished for not parroting AMD.’Why is that important even if true, unless you’re a stocks jockey?
AMD will sell what it sells. So will Intel. Benchmarks will continue to be done over the lifespans of products. Today’s hot tech will soon be yesterday’s old hat.
Perhaps it’s my age but until I see something actually exciting, rather than bland ‘give consumers the minimum’ iterations, I’ll refrain from caring much about how independent the press is.
Even the kneecapping of Zen 1 and 2 with JEDEC is a self-inflicted wound on AMD’s part. Yawn.
Let’s see a consumer-priced wafer-scale GPU. Something actually interesting.
Bruzzone - Wednesday, August 31, 2022 - link
Raphael price from TSMC to AMD is the same as Vermeer, maybe $10 less at + 20% cost + 30% good die, however, AMD price to high volume OEMs will be the same as Vermeer. $100 MSRP cut on Raphael 16C and 8C from Vermeer is purely an AMD PR ploy. The channel will price Ryzen 7K accordingly on performance out the gate. The channel will rely on R7x Extremes price premium above MSRP to pay for channel's AMD Rembrandt, Cezanne primary production overage stuck on the shelf subject financial loss offset. After the channel protects against its AMD mobile inventory financial losses it can start covering its Intel Tiger mobile overage on Raphael and Raptor price premiums > MSRP. Its the way the world works and this is AMD and Intel's fault not the channel! mbBruzzone - Thursday, September 1, 2022 - link
Provided this reassessment up comment stream on NEW die area data;Vermeer in relation to Raphael. V5x normalized area across the grade SKUs over the full run with its i/o die = 243 mm2. R7x normalized on V5x full run percent grade SKU split = 239 mm2 near same. Which is interesting NEW data because I originally estimated + 20% cost and + 30% good dice on 5 nm normalized ccd and the i/o dropping to 171 mm2 of area IT DID NOT and the reason is the 6 nm i/o + iGPU is no less area than the 12 nm i/o.
There seems to be disagreement about the area of the 5 nm ccd. Some report 77 mm2 and others report 71 mm2 a less 6.8% difference. At R7x 5 nm 77 mm2 ccd is 8.4% smaller than V5x 7 nm 84 mm2.
Whether V5x or R7x an 80% yield on edge and area defects delivers roughly the same number of normalized dice sufficient for 230 to 255 produced components per wafer.
mb