Discussion Speculation: Zen 4 (EPYC 4 "Genoa", Ryzen 7000, etc.)

[Vattila]

Except for the details about the improvements in the microarchitecture, we now know pretty well what to expect with Zen 3.

The leaked presentation by AMD Senior Manager Martin Hilgeman shows that EPYC 3 "Milan" will, as promised and expected, reuse the current platform (SP3), and the system architecture and packaging looks to be the same, with the same 9-die chiplet design and the same maximum core and thread-count (no SMT-4, contrary to rumour). The biggest change revealed so far is the enlargement of the compute complex from 4 cores to 8 cores, all sharing a larger L3 cache ("32+ MB", likely to double to 64 MB, I think).

Hilgeman's slides did also show that EPYC 4 "Genoa" is in the definition phase (or was at the time of the presentation in September, at least), and will come with a new platform (SP5), with new memory support (likely DDR5).

What else do you think we will see with Zen 4? PCI-Express 5 support? Increased core-count? 4-way SMT? New packaging (interposer, 2.5D, 3D)? Integrated memory on package (HBM)?

Vote in the poll and share your thoughts!

 

[Kaluan]

He's wrong about 7600X competing against Raptor Lake i7, but the 7600X will definitely be competing against Alder Lake i7 + DDR4.

i7 12700K = $370 + Decent Z690 DDR4 mobo = $200 + OK 3200 CL16 32 GB DDR4 = $120. Total = $690.
R5 7600X = $300 + Decent B650 motherboard = $200 + OK 5200 CL40 32 GB DDR5 = $170. Total = $670.

...5900X is a thing you know 😂

But yeah I guess if we all act like AMD has no DDR4 options for whatever reason, like 12th/13th gen are just as good with DDR4 as with DDR5, like AM5 is a dead end platform (and somehow DDR4 1700 isn't, at the same time) and like B650 mobos will cost as much as Z690 mobos... all that adds up nicely for Intel.

And I said "something does not add up here", I was not denying it. I am asking for reasons for the disparity I see here.

Well, random idea guys but I think treating PPT as power consumption may be misleading, or even wrong.

The so-called 350W "Extreme OC" mode is a spin given by some rumor-mongers to a leak by Raichu, which shows the 13900K consuming 345 W when you set PL2 manually to 4096 W, which is the limit allowed in current motherboards.

Core i9-13900K breaks 40K points in Cinebench R23 with unlimited power and 5.8 GHz clock - VideoCardz.com

Intel Core i9-13900K compared with and without power limiting settings Leaker OneRaichu shares new benchmark results featuring the flagship Raptor Lake CPU. The 24-core and 32-thread Raptor Lake processor scores 2290 points in single-core Cinebench R23 test and 35693 points in multi-core. It...

videocardz.com

There is no basis to claim that such an extreme OC mode exists.

Moot point how it's called or what it is, that doesn't change the fact that we've been flooded with some Raptor Lake performance leaks that don't represent stock behavior. It's nothing new tho, there were quite a few 12th gen leaks that showed substantially better performance than what we got with the retail versions as well. Same with Rocket Lake...

How come people talk about game scaling as if the cpu alone is responsible for generating the fps? I believe the newer generation of gpus will reveal the real difference between a chip like the 5600x and 12900k.

Or 12900K and 7600X 😉

DDR4 with AL or Raptor Lake will be noticeably inferior in gaming to 7600X. Also, where did you get 400$ price tag for 13700K ?

Well, you know... trust him bro! The ability to wishful think is strong with that one.
I wonder if we can have a 500$ 13900K next? 😅

Always amazes me how Intel squad think they can have their DDR4/DDR5 cake and eat it too when talking about pure value builds but also pure performance builds.
Seriously, I can't wait for the next Intel socket, so at least this nonsense will stop.
Hell it might be over by mid-next year, when DDR5 is estimated to become cheaper than DDR4.

 

[coercitiv]

In fact, that slide you posted shows memory latency at 63ns which is already more than what Alder Lake can achieve. According to Techpowerup, 12900K plus DDR5 6600 gets 59ns.

That slide I posted also shows a figure, "up to 11% enhanced gaming performance at 1080p". Here's where that claim came from:

When optimized for high-performance gaming, consumers can expect to see up to 11% faster gaming performance with AMD EXPO technology in F1® 202212.

Spoiler: Endnote

12. RPL-011: Testing as of 15 August, 2022, by AMD Performance Labs using the following hardware: AMD Socket AM5 Reference Motherboard with Ryzen™ 5 7600X and G.Skill DDR5-6000C30 (F5-6000J3038F16GX2-TZ5N) with AMD EXPO™ versus ROG Maximus Z690 Hero with Core i9-12900K and G.Skill DDR5-6000C30 (F5-6000J3038F16GX2-TZ5N) with AMD EXPO™ loaded. ALL SYSTEMS configured with NXZT Kraken X63, open air test bench, Radeon™ RX 6950XT (driver 22.7.1 Optional), Windows® 11 22000.856, AMD Smart Access Memory/PCIe® Resizable Base Address Register (“ReBAR”) ON, Virtualization-Based Security (VBS) OFF. F1 2022 tested at 1920x1080 with HIGH in-game preset and the chronologically newest graphics industry API available within the game’s rendering engine (e.g. Vulkan® over OpenGL™, DirectX® 12 over DirectX® 11). Results may vary.

A few weeks ago you contradicted someone claiming Zen 4 could also make use of DDR5 in gaming. You were very adamant on your position, the large cache and chiplet design just wouldn't play nice with faster memory. Now we have the 7600X winning in game performance against the 12900K, both using DDR5-6000C30. 32MB L3 versus 30MB L3. Chiplet vs. monolithic with better latency. Early firmware versus matured firmware.

Sure we can argue Alder & Raptor may be able to scale further with even faster memory, but that was never the major point of contention. The simple fact that Zen 4 beat Golden Cove in (some) games while using the same DDR5 memory should raise a huge red flag for you, maybe a more reserved position would be in order until we're able to understand the sources of improved performance and properly evaluate DDR5 scaling in the context of Zen vs. Cove matchup.

 

[Carfax83]

You wont see mid 55ns by just enabling XMP with Raptor Lake anytime soon is my take

@tamz_msc already showed that Intel has addressed that flaw in Raptor Lake.

A lot of people are thinking that Raptor Lake will just be Alder Lake with higher clocks and more cache but that's not true.

Zen 4 looks like a powerful CPU, but don't be surprised if it takes a backseat to Raptor Lake when it comes to gaming due to the enhanced cache structure and performance in addition to the ability to utilize higher memory frequencies.

 

[tamz_msc]

...5900X is a thing you know 😂

But yeah I guess if we all act like AMD has no DDR4 options for whatever reason, like 12th/13th gen are just as good with DDR4 as with DDR5, like AM5 is a dead end platform (and somehow DDR4 1700 isn't, at the same time) and like B650 mobos will cost as much as Z690 mobos... all that adds up nicely for Intel.

A similar setup with the 5900X would cost $50 less, but it will be slower than both the 7600X and 12700K in gaming.

For gaming the order is 7600X>12700K>5900X

For applications, 5900X>12700K>7600X

So the 12700K emerges as a good all-rounder, with the potential to upgrade to Raptor Lake if you want to.

AM4 should not even be under consideration right now for enthusiasts - it is a dead-end platform.

 

[TESKATLIPOKA]

I'm sure Phoenix will be amazing, but people are reading way too much into that perf@65w slide. First, Rembrandt is already very efficient on 6nm. Second, a large, chunk of Raphael's efficiency over Vermeer at lower TDPs is certainly down to a more efficient 6nm IOD, leaving more budget for the cores by lowering static power draw.

A ~30% more efficient Phoenix (vs. Rembrandt) may be doable, maybe even a bit more, but I wouldn't expect anything approaching the 75% we see in the slide deck.

As you said, that slide can't be directly applied to Dragon Ridge or Phoenix Point.
What we know based on the AMD slide and Det0x's data is this:

	5950x	7950x (AMD slide)	7700x (calculated)
TDP(PPT)	65W (88W)	65W (88W)	~35W (44W)
Frequency	2377-2402 MHz	3803-3843 MHz	3803-3843 MHz
CB R23	16,210	28,205	~14,103

R9 6900HS tested at ComputerBase (TDP gemessen -> TCP measured).

So my expectation is that 8C16T Zen4 will manage comparable score at 1/2 power, but absolute performance at the same TDP will increase by ~20-30%.
If you want true power, then you have to buy a mobile 16C32T Dragon Ridge.

 

[Carfax83]

A few weeks ago you contradicted someone claiming Zen 4 could also make use of DDR5 in gaming. You were very adamant on your position, the large cache and chiplet design just wouldn't play nice with faster memory. Now we have the 7600X winning in game performance against the 12900K, both using DDR5-6000C30. 32MB L3 versus 30MB L3. Chiplet vs. monolithic with better latency. Early firmware versus matured firmware.

I already posted about the issues I had with the configuration that AMD used in that test.

First they used what is perhaps the most optimal memory configuration possible for Zen 4. DDR5 6000 in a 1:1 ratio with 3ghz IFC. How easy or difficult it will be to achieve that config remains to be seen. Presumably, not every Zen 4 CPU will hit 3ghz IFC.

But it also shows to me that Zen 4 is in many ways like Zen 3 when it comes to memory performance in the sense that 1:1 ratio will be optimal. It will likely not perform as well in gear 2 mode, which will limit the usefulness of higher speed DDR5.

This is the opposite of Intel, which can still benefit from faster speeds in gear 2. And despite this, AMD still used the same speed memory on the 12900K rig, knowing damn well that Intel can benefit from much faster speeds.

Sure we can argue Alder & Raptor may be able to scale further with even faster memory, but that was never the major point of contention. The simple fact that Zen 4 beat Golden Cove in (some) games while using the same DDR5 memory should raise a huge red flag for you, maybe a more reserved position would be in order until we're able to understand the sources of improved performance and properly evaluate DDR5 scaling in the context of Zen vs. Cove matchup.

I'm all for more benchmarks and leaks, and even more so, for final reviews.

I could be mistaken of course, but so far it appears I may be right. 1:1 ratio is very important to Zen 4 thus limiting the usefulness of DDR5 memory frequencies that exceed 6ghz. Also, that AMD slide implies the lowest latencies in gear 1 mode will be 63ns.

Alder Lake can already hit lower latencies than that in gear 2 mode with faster DDR5 and Raptor Lake will definitely up the ante with support for higher frequency DDR5 out the box in addition to a revamped cache structure, ring bus and more robust IMC.

So basically, Raptor Lake won't be no daisy like what many AMD proponents are hoping.

 

[TESKATLIPOKA]

This is a continuation of my previous post.
6900HS 45->70W: 13% gain
6900HS 35->45W: 9% gain
6900HS 25->35W: 17% gain
6900HS 15->25W: 56% gain

I expect >50% increase in performance at 15W TDP, because Zen4 should have clocks comparable to at least a 25W 6900HS.

edit: a 5800U at 15W has a lot better score(+20%) than this 6900HS(CB R23: 6238) at 15W, because It's a binned part meant for lower TDP.

Now, I am not so sure how good Zen 4 will be in mobile.

 

[uzzi38]

I'm kinda disappointed nobody seems to have really thought about what AMD talked about with regards to Zen 4C honestly.

Same ISA support, same IPC, and now also confirmed to use half of the core area (key word being core, mind you).

That puts Zen 4C in similar size regions to ARM cores such as V1. That's kind of a big deal.

Also entirely unrelated sidenote but:

https://twitter.com/x/status/1564858445546807297

I did warn you guys. For Raphael vs Raptor Lake the issue isn't power consumption, it's cooling. Thermal density is not a fun thing.

 

[Vope45]

I'm kinda disappointed nobody seems to have really thought about what AMD talked about with regards to Zen 4C honestly.

Same ISA support, same IPC, and now also confirmed to use half of the core area (key word being core, mind you).

That puts Zen 4C in similar size regions to ARM cores such as V1. That's kind of a big deal.

Also entirely unrelated sidenote but:

https://twitter.com/x/status/1564858445546807297

I did warn you guys. For Raphael vs Raptor Lake the issue isn't power consumption, it's cooling. Thermal density is not a fun thing.

Wow that is scarily hot

 

[yuri69]

The liquid cooling requirement for the 170W AM5 SKUs is present in the GigaByte leaked docs. So this should be no surprise.

TBH cooling the current 7nm 5800X can be tricky either. The upcoming gen has even higher power density.

 

[Kuiva maa]

I'm kinda disappointed nobody seems to have really thought about what AMD talked about with regards to Zen 4C honestly.

Same ISA support, same IPC, and now also confirmed to use half of the core area (key word being core, mind you).

That puts Zen 4C in similar size regions to ARM cores such as V1. That's kind of a big deal.

Also entirely unrelated sidenote but:

https://twitter.com/x/status/1564858445546807297

I did warn you guys. For Raphael vs Raptor Lake the issue isn't power consumption, it's cooling. Thermal density is not a fun thing.

Obviously consumption still matters a great deal,how much heat are you dumping in your room while running this is what I want to hear, not tmax. If boost algo allows this it might not be that bad, but If the AIO gets saturated and the cpu is throttling then yeah we have a problem.

 

[Zucker2k]

But do you actually disagree with my comment below ?

Since it's based on your earlier assumptions about RPL inheriting adverse ADL latencies, a fat YES!

What AMD has done here with Zen 4 looks desperate at first, but Intel has raised the TDP ceiling so high in order to remain competitive that AMD could easily take advantage of that power gap, Intel being their main and only real competitor in the x86 ecosystem. This is Zen 4. AMD has upped the TDP 62% each on the lower end, from 65w to 105w, and from 105w to 170w on the upper end.
I said a while ago in this thread, that if AMD doesn't raise cores from 16 to 24 they could lose the multithreaded war to the competition. In hindsight, my observation was quite myopic because it failed to take into account the one other way AMD could attain parity or superiority besides raw IPC, power. 62% increase in TDP is the answer, then.
I believe a precedent has been set here, relative efficiency on the desktop may well be sacrificed by both Intel and AMD in their bid to either stay competitive or outdo each other. This is because they'll likely prefer to chase clocks rather than invest in real estate (bigger or more cores). Let's see how things pan out.

 

[PJVol]

From the leaks, they are saying the max IFC will be 3ghz, which means to run at 1:1 ratio you will need to pair it with DDR5 6000.

What leaks? Haven't yet seen any hint at the max achievable fabric clock.

I did warn you guys

This?

Cooling Challenge
Ryzen 7000 Desktop will therefore represent the ultimate cooling challenge seen thus far. The die can approach 2 Watts/mm² if your cooler is up to the task. The smaller heatspreader area on Ryzen 7000 package design only makes things worse. We discussed the cooling difficulty and why the heatspreader is smaller in our Computex piece in May.

Due to the cooling difficulty and higher power limits on Ryzen 7000 Dekstop, the philosophy has changed to follow a more laptop-like approach. Instead of fixing a clockspeed or power limit, the processor is now temperature limited. This means maximizing performance for a given cooling capability. The processor will adjust to stay within the temperature limits. This also means a more significant performance difference that changes with cooling ability.

Ryzen 7000 Desktop Preview

Transistor Count + Microarchitecture

www.angstronomics.com

 

[uzzi38]

What leaks? Haven't yet seen any hint at the max achievable fabric clock.


This?

Ryzen 7000 Desktop Preview

Transistor Count + Microarchitecture

www.angstronomics.com

I'm not Skyjuice lol. No, I said it multiple times in here

Page 308 - Discussion - Speculation: Zen 4 (EPYC 4 "Genoa", Ryzen 7000, etc.)

Page 308 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

Page 300 - Discussion - Speculation: Zen 4 (EPYC 4 "Genoa", Ryzen 7000, etc.)

Page 300 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

 

[PJVol]

I'm not Skyjuice lol

Who knows...)

Btw, do you know what could be the max reasonable FCLK for these?

 

[uzzi38]

Who knows...)

Btw, do you know what could be the max reasonable FCLK for these?

Nope, not a clue lol.

And no, I'm definitely not Skyjuice. I've known him for a pretty long time though through Discord, but that's really just about it.

 

[Zucker2k]

Now we have the 7600X winning in game performance against the 12900K, both using DDR5-6000C30. 32MB L3 versus 30MB L3. Chiplet vs. monolithic with better latency. Early firmware versus matured firmware.

Sure we can argue Alder & Raptor may be able to scale further with even faster memory, but that was never the major point of contention. The simple fact that Zen 4 beat Golden Cove in (some) games while using the same DDR5 memory should raise a huge red flag for you, maybe a more reserved position would be in order until we're able to understand the sources of improved performance and properly evaluate DDR5 scaling in the context of Zen vs. Cove matchup.

Please do you know the settings that Ram run in the ADL system? Do you know if it ran at CL30? What about the other timings? It's not far fetched for any enthusiast worth their salt to understand that putting an AMD EXPO tuned RAM into a non EXPO rated bios, much more a non AMD system, isn't going to run optimally. This is not so hard to understand. The 7600x run optimally. The ADL didn't. Add the cherry-picked games on top and you should understand why your "winning in games" comment is not only laughable but smacks of gullibility since these are not even third-party reviews but AMD slides. If nothing at all, history should teach us to exercise some patience until we get reviews from other sources since this isn't the first time AMD has pulled this sort of stunt with their gaming numbers.

 

[coercitiv]

Since it's based on your earlier assumptions about RPL inheriting adverse ADL latencies, a fat YES!

Even if it does inherit them, they're not what defines memory performance on ADL/RPL any longer. AFAIK with Alder Lake, Intel changed their memory controller "philosophy" towards favoring multi-channel throughput. Previous generations inlcuding Skylake clones favored latency first and throughput second. This worked great for consumer workloads until CPUs evolved into these MT workhorses and memory evolved into DDR5 (with sub-channels, again optimized for memory ops throughput).

So, do we want lower "measured" latency? Sure we do, but other performance metrics may weigh in more nowadays. Some incremental improvements in Raptor Lake memory and inter-core latency would be great, but far outweighed by stronger cache performance.

Comet Lake reached 40-45ns in mem latency and had massivley better inter-core latencies. It still had no chance against Golden Cove, especially in gaming where these latencies were favored the most.

 

[naad]

Fantastic job by AMD, more IPC than expected and around 5.8GHz, Raptor lake will not have a easy job in desktop, and it will get decimated by Phoenix and Dragon Range in laptops if those 65W claims are anywhere remotely true.

Don't think AMD has to worry much with every single market besides desktop DIY ending up a complete domination for them

 

[PJVol]

Hmm... Geekbench AES-XTS is strange
7600X and my 5600X scored
ST: 11.0 and 7.37 gb/s
MT: 15.7 and 15.6 gb/s

----------
figured out

In some MT tests such as Strucuture from Motion, AES-XTS, Text compression, 7600X is most likely thermally throttled and hence clockgated. 5600X is quickly approaches 100W PPT there.

 

[coercitiv]

Please do you know the settings that Ram run in the ADL system? Do you know if it ran at CL30? What about the other timings? It's not far fetched for any enthusiast worth their salt to understand that putting an AMD EXPO tuned RAM into a non EXPO rated bios, much more a non AMD system, isn't going to run optimally. This is not so hard to understand. The 7600x run optimally. The ADL didn't.

Are you saying Zen 1-3 ran with a deficit all these years? Imagine that, Skylake/Cove running optimally with XMP while Zen didn't. /s

At least @Carfax83's complaint made sense.

you should understand why your "winning in games" comment is not only laughable but smacks of gullibility

let's read what I wrote:

The simple fact that Zen 4 beat Golden Cove in (some) games

damn wording, changing nuances and stuff...

I eagerly await the Raptor Lake launch where you will warn everyone on the forum to remember XMP is not optimized for AMD and Intel gaming performance slides are for gullible people only. Imagine that, if Raptor manages to surprise everyone with massively improved gaming performance, you will be the one stoic in the Intel thread, patiently awaiting full coverage with 3rd party results while refusing to even discuss marketing slides.

 

[Abwx]

Geekbench has ADL ahead in int IPC, but behind in fp IPC against Zen 4. Compare against clean GB runs for example from our forum member AdamK47:

Gigabyte Technology Co., Ltd. Z690 AORUS MASTER - Geekbench

Benchmark results for a Gigabyte Technology Co., Ltd. Z690 AORUS MASTER with an Intel Core i9-12900KS processor.

browser.geekbench.com

vs

7600X

Micro-Star International Co., Ltd. MS-7D69 - Geekbench

Benchmark results for a Micro-Star International Co., Ltd. MS-7D69 with an AMD Ryzen 5 7600X processor.

browser.geekbench.com

His 12900KS score as much as a 13900K, certainly not a clean results as such, it can be compared only to scores from future owners of a Zen 4.

We ll see when people from this forum do tests, for the time being Zen 4 has better IPC in Geekbench, FTR the 12900K score 2040, so you had to rely on a highly tweaked 12900KS.

 

[DrMrLordX]

Their entire presentation is polluted by the fact that they effectively ran overclocked systems, arbitrarily stopping at DDR5-6000 using AMD specific performance enhancing timings.

Would you rather they ran everything at DDR4-4800 with crap timings? Or maybe just the Alder Lake system? Seems like you're going to complain no matter what.

I won't argue with you there. I think my b-die kit had trfc at 700 something with xmp. I got it down to 280. It didn't exactly equate to any massive performance gains though.

Take what I say with a grain of salt, but from my limited understanding of DDR4 subtimings, lower doesn't necessarily mean better. You want to find the right subtiming ranges where you can push more clocks without making compromises on your primaries, requiring more voltage, or crippling performance somehow. Some of those subtimings can destroy your performance if you play with them too much. So lower trfc isn't fundamentally going to make you faster, but it might help you push another few MHz RAM clocks assuming you have IF headroom (on an AMD system) without loosening something else or pushing too much extra voltage.

That's why XMP 2.0 on AMD can be so cringeworthy. It'll give you those insane subtimings, and then ramp up vDIMM plus some "hidden" voltages you might not notice like VDDP and VDDG to get everything stable. If the subs were right then you could probably do with less voltage.

Also XMP messes with vSoC and that can be bad.

Thermal density is not a fun thing.

Thermal density has been un-fun since Zen2. I'm hoping for less hotspot headaches but I won't hold my breath. There will probably be a hard wall where even the best ambient cooling (water included) just won't improve anything.

 

[HurleyBird]

I'm kinda disappointed nobody seems to have really thought about what AMD talked about with regards to Zen 4C honestly.

Same ISA support, same IPC, and now also confirmed to use half of the core area (key word being core, mind you).

Did they say same IPC? I might have missed that. Citation?

 

[nicalandia]

Some dude on Twitter posted this