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!

 

[jamescox]

We knew that from day one.

And have repeated repeatedly.

 

[Henry swagger]

AMD cannot afford to have low clocks anywhere, the Skylake days are over and in it's place is brutal GC core backed by efficiency cores.
The reality is that Intel will fight 7600X with 13600K that is rumored 6P + 8E. 7800x will be assaulted by 13700K with 8P + 8E config and so on, increase of E cores trickles down the line.

7950x has 8vs8 (vs 8P) that leaves 8 Z4 cores fighting 16E
7900x has 6+6 (vs 8P) that leaves 4 cores fighting 8E
7800x has 8 (vs 8P) that leaves 8E cores
7600x has 6 (vs 6P)that leaves 8E cores

It is fine to discuss CB scores of 35-40k and high end SKUs, but bulk of sales is not there. It's the lower end SKUs that need clocks the most to not loose throughput tests in disastrous way. AMD is really walking on thin ice here, their ST advantage has evaporated and seems that it is not really coming back and they need to push clocks and TDP to what is most likely stupid efficiency region.
On top of that new platform with DDR5 that makes selling anything with less than 8C an uphill battle. The SKUs and pricing are going to be very interesting this timem we might get 7700x real quick this time. Viva the competition!

Strix point is amd's response to intel's hybrid cores..with zen 5 + zen 4c

 

[eek2121]

They said all game threads were between 5.2-5.5GHz, and as previously established, that game doesn't use 8 cores, much less 16.

Yet, it was > 1 core. Note that Intel can't even carry 5.52 ghz on 1 core. I do bring up the competition because folks are apparently insistent on bashing AMD for setting records.

#1 makes perfect sense, it might be a 16% gain today but they think with another stepping to clean up a few issues it will be higher when released, but they don't want to commit to anything and get egg on their face so they are being conservative until they know for sure.

#2 makes no sense at all. AMD is selling every Zen 3 they make, and will continue selling them even when Zen 4 is on the market, so they have no worries about being Osbourned. If they want to steal share from Intel they should want to advertise the best possible numbers they can, because they would want people considering buying Intel today to decide they need to wait for Zen 4 before making their decision.

There's no world where underselling performance "maintains competitive advantage". It only helps if you want to deliberately hamstring performance in a released product, like Intel has been accused of doing in the past when they effectively had no competition.

You must not be in marketing, or sales for that matter, because you clearly don't have a clue how to sell a product. It isn't about AMD selling every part, it is about AMD selling every part for the highest price possible.

AMD cannot afford to have low clocks anywhere, the Skylake days are over and in it's place is brutal GC core backed by efficiency cores.
The reality is that Intel will fight 7600X with 13600K that is rumored 6P + 8E. 7800x will be assaulted by 13700K with 8P + 8E config and so on, increase of E cores trickles down the line.

7950x has 8vs8 (vs 8P) that leaves 8 Z4 cores fighting 16E
7900x has 6+6 (vs 8P) that leaves 4 cores fighting 8E
7800x has 8 (vs 8P) that leaves 8E cores
7600x has 6 (vs 6P)that leaves 8E cores

It is fine to discuss CB scores of 35-40k and high end SKUs, but bulk of sales is not there. It's the lower end SKUs that need clocks the most to not loose throughput tests in disastrous way. AMD is really walking on thin ice here, their ST advantage has evaporated and seems that it is not really coming back and they need to push clocks and TDP to what is most likely stupid efficiency region.
On top of that new platform with DDR5 that makes selling anything with less than 8C an uphill battle. The SKUs and pricing are going to be very interesting this timem we might get 7700x real quick this time. Viva the competition!

Golden cove is only brutal to my air conditioning and power bill. It most certainly does not beat my 5950x in my workloads.

NOTE: A 64 core Threadripper or EPYC consumes less power than a 12900k/12900ks. Those chips are significantly faster as well.

 

[TESKATLIPOKA]

NOTE: A 64 core Threadripper or EPYC consumes less power than a 12900k/12900ks. Those chips are significantly faster as well.

What's the point of comparing a desktop CPU with barely any headroom in clocks left + no power limits to a server CPU with more cores but very conservative clocks? It's easy to see which one will be more power efficient.

You should know very well that by increasing clockspeed the power consumption increases a lot more than performance.

Increasing P-core frequency by 65% and E-core by 54% results in only 45% more performance, but PL2 increased by 335%.

I expect Zen4 to be very efficient in mobile, but in desktop I think they will go most likely the Intel route, in other words highest performance without regard for efficiency, at least for top 170W TDP model.

 

[biostud]

What's the point of comparing a desktop CPU with barely any headroom in clocks left + no power limits to a server CPU with more cores but very conservative clocks? It's easy to see which one will be more power efficient.

You should know very well that by increasing clockspeed the power consumption increases a lot more than performance.
View attachment 62235
Increasing P-core frequency by 65% and E-core by 54% results in only 45% more performance, but PL2 increased by 335%.

I expect Zen4 to be very efficient in mobile, but in desktop I think they will go most likely the Intel route, in other words highest performance without regard for efficiency, at least for top model.

Doesn't it depends whether you use PBO or not? Those buying the 7950X are likely to prefer efficiency over smaller performance increases, so a bit lower MT boost to keep power consumption reasonable wouldn't surprise me.

 

[TESKATLIPOKA]

Doesn't it depends whether you use PBO or not? Those buying the 7950X are likely to prefer efficiency over smaller performance increases, so a bit lower MT boost to keep power consumption reasonable wouldn't surprise me.

I am not sure if I understand you correctly.
Do you think that 7950X will hit 170W TDP(230W PPT) only when PBO is enabled?

 

[ryanjagtap]

I think the 170W TDP/230W PPT SKUs are just halo products to compete with Intel's halo SKUs like the i9-13900K/i9-13900KS. Just like Intel tried to squeeze out every bit of performance they could by feeding more power just for the benchmark wins forgoing efficiency, these will do the same.

 

[lobz]

That even more egregiously wrong "leaks" exist does not make his any closer to reality. This is well beyond any reasonable margin for error.

yet the crusade seems to elude chipsNcheese and his trusty saucy source

Come on, man. He's really not worth the effort you're putting into defending him. If his minimum IPC number was accurate, then given the clocks shown already, AMD could probably claim +30% performance. And if the clocks actually end up higher, then it'll be above the top of his range! He was wildly off base, and it's not the first time. So let's just accept that and move on.


Lol, then I better be quick, or he'll delete it before I do.

Sure. Then let's accept the facts in this strange way you perceive them: AMD had reached >15% higher ST performance with Zen4 at a 14% higher clockspeed.

Looking forward to dissect exactly how much energy you'll deem worthy of defending your own nonsense once the product launches.

 

[exquisitechar]

yet the crusade seems to elude chipsNcheese and his trusty saucy source

Come on, you can’t equate Cheese and MLID. Cheese made one low profile leak that I think (might be wrong here) was basically one line in an entire article. Otherwise, they host high quality articles written by some very knowledgeable people that contain things like microarchitectural deep dives that you can seldom find anywhere else. MLID peddles his BS constantly and offers nothing of worth, at least in the domain of AMD and Nvidia. His track record with Intel leaks is much better and I do admit he has sources, but the rest of his content being garbage is still true.

 

[Atari2600]

The higher TDP on the socket + what looks almost static IPClk + node shrink would tend to suggest not great improvements in core throughput.

I can't help but feel AMD has had a little bit of a miss with this one. Nothing as bad as a Bulldozer, or even a Barcelona. Maybe a Barton (ironically, it had a problem reaching higher clocks).

Hopefully they've undersold performance at Computex to keep existing products moving off shelves, but I doubt it.

 

[lobz]

Come on, you can’t equate Cheese and MLID. Cheese made one low profile leak that I think (might be wrong here) was basically one line in an entire article. Otherwise, they host high quality articles written by some very knowledgeable people that contain things like microarchitectural deep dives that you can seldom find anywhere else. MLID peddles his BS constantly and offers nothing of worth, at least in the domain of AMD and Nvidia. His track record with Intel leaks is much better and I do admit he has sources, but the rest of his content being garbage is still true.

Actually I can. With MLID under a magnifier (despite taking any backlash far more mature than an actual supposedly 'mature' person like adored), why should I lapse over someone saying something as deliberate as 'a trusted source'? Apart from all the positives I really like in his website, my only comment on that: low profile my trusted backside 😊

And still I have zero problems with the guy. I have (big) problems with bias and acting like 20/20 hindsight malevolent vultures like some here do.

And you also managed to somehow miss the actually important point of my comment, that made up like 90% of the characters used. Why is that, I wonder.

 

[Asterox]

The higher TDP on the socket + what looks almost static IPClk + node shrink would tend to suggest not great improvements in core throughput.

I can't help but feel AMD has had a little bit of a miss with this one. Nothing as bad as a Bulldozer, or even a Barcelona. Maybe a Barton (ironically, it had a problem reaching higher clocks).

Hopefully they've undersold performance at Computex to keep existing products moving off shelves, but I doubt it.

Lol, how can you even mention red failure of epic proportions with any bits around Ryzen CPU-s.The red plague destroyed a very good AMD FX CPU brand.

FX will not return in the future, hm or maybe he will around year 2050.

 

[Timmah!]

What's the point of comparing a desktop CPU with barely any headroom in clocks left + no power limits to a server CPU with more cores but very conservative clocks? It's easy to see which one will be more power efficient.

You should know very well that by increasing clockspeed the power consumption increases a lot more than performance.
View attachment 62235
Increasing P-core frequency by 65% and E-core by 54% results in only 45% more performance, but PL2 increased by 335%.

I expect Zen4 to be very efficient in mobile, but in desktop I think they will go most likely the Intel route, in other words highest performance without regard for efficiency, at least for top 170W TDP model.

so the golden cove efficiency goes heywire somewhere inbetween 4,2 GHz and 4,8GHz…. The question is where exactly? Is it linear increase? Or does it happen suddenly at 4,3 or 4,5 or 4,7 GHz? Too bad neither of those clocks are in that table…

 

[mikk]

Here is Robert Hallock stating that they are getting 40+% higher performance right now ( with the current state of silicon) : LINK
I guess he is referring to Zen4 vs Zen3 in MT workloads but that's really encouraging as it implies that multi core boost is much higher (north of 20% vs 4.2Ghz on 5950X) and IPC should be at minimum >10%

40% compared to what? 12900K? 5950x? What power limits? Sorry this number is meaningless without any context. This is damage control after his big TDP fail what he is doing.

 

[Hougy]

Tom is not a reliable leaker, and especially not for AMD.Tom is a tech Youtuber(YouTube earnings + Patreons), or similar can be said for RGT.

Real or trusted leakers, a couple of them mostly only post on Twitter.

Up to 37% higher IPC, it is pure shooting at geese in a thick fog=clickbait.

Who are the trusted leakers? Please link

 

[inf64]

40% compared to what? 12900K? 5950x? What power limits? Sorry this number is meaningless without any context. This is damage control after his big TDP fail what he is doing.

He stated the pre-QS part they showed was not running at the maximum TDP (170W). As the 40+% figure, does that even matter? According to computer base latest article, 5950X is ~6.3% faster than 12900KS and ~10% faster than 12900K in MT. So 40+% versus any 3 of them is still massive uplift that will likely put the top (170W) part above anything intel will come up with in MT.

 

[Atari2600]

Lol, how can you even mention red failure of epic proportions with any bits around Ryzen CPU-s.The red plague destroyed a very good AMD FX CPU brand.

If I don't, someone else would only end up interpreting it that way.

Unfortunately, thats what happens on the interwebz.

 

[biostud]

I am not sure if I understand you correctly.
Do you think that 7950X will hit 170W TDP(230W PPT) only when PBO is enabled?

Yes, I think I 7950X will be less than 200W, but obviously it depends on how aggressively AMD think they need to clock it.

 

[eek2121]

What's the point of comparing a desktop CPU with barely any headroom in clocks left + no power limits to a server CPU with more cores but very conservative clocks? It's easy to see which one will be more power efficient.

You should know very well that by increasing clockspeed the power consumption increases a lot more than performance.
View attachment 62235
Increasing P-core frequency by 65% and E-core by 54% results in only 45% more performance, but PL2 increased by 335%.

I expect Zen4 to be very efficient in mobile, but in desktop I think they will go most likely the Intel route, in other words highest performance without regard for efficiency, at least for top 170W TDP model.

The Threadripper parts run at decent clocks, unsure what you are blabbing on about. All of the Zen 3 Threadripper parts top out at 4.5ghz. AMD could push higher, if needed, but at the expense of power. I brought it up to show exactly how far behind Intel is. AMD can squeeze 64 cores into a 280W TDP. Intel can’t.

Writing off Zen 4 because they are possibly releasing a halo SKU that matches Intel’s power limits is just silly. Most, if not all Zen 4 chips will be at much lower TDPs and power limits. You do you, however.

40% compared to what? 12900K? 5950x? What power limits? Sorry this number is meaningless without any context. This is damage control after his big TDP fail what he is doing.

Damage control? TDP fail? Where? Why so you insist on making a big deal out of this without also not making a big deal out of the 12900ks consuming nearly 300w loaded?

I saw no issues with anything AMD has announced personally. They had to announce a clarification, sure, but companies, including Intel, do that all the time. Still worse, Intel flat out lies about release dates.

Look my dude, many of us are pretty balanced regarding Intel, AMD, or chipmaker xyz, so when you post comments like this, to us, it looks like you and the poster above are simply trolling an AMD thread. If you have nothing insightful to add then I encourage you and your partner in crime to resist the urge to click that “Post Reply” button.

 

[mikk]

He stated the pre-QS part they showed was not running at the maximum TDP (170W). As the 40+% figure, does that even matter? According to computer base latest article, 5950X is ~6.3% faster than 12900KS and ~10% faster than 12900K in MT. So 40+% versus any 3 of them is still massive uplift that will likely put the top (170W) part above anything intel will come up with in MT.

First he claimed it's 170W PPT and now hey says less than 170W TDP which is a completely meaningless claim as well! 169W TDP is less than 170W with an unknown PPT. This guy is unreliable, it's John Fruehe 2.0. He failed so hard it isn't funny. Stop repeating his marketing nonsense.

 

[inf64]

First he claimed it's 170W PPT and now hey says less than 170W TDP which is a completely meaningless claim as well! 169W TDP is less than 170W with an unknown PPT. This guy is unreliable, it's John Fruehe 2.0. He failed so hard it isn't funny. Stop repeating his marketing nonsense.

Well, we are all free to believe what we think is correct. I think you will be unpleasantly surprised when Zen4 launches, that's just my gut feeling

 

[eek2121]

First he claimed it's 170W PPT and now hey says less than 170W TDP which is a completely meaningless claim as well! 169W TDP is less than 170W with an unknown PPT. This guy is unreliable, it's John Fruehe 2.0. He failed so hard it isn't funny. Stop repeating his marketing nonsense.

He didn't fail at all. Socket TDP != processor TDP.

For the CPU, AMD did not have set parameters for the chip they demoed.

For the socket, AMD attempted to clarify information regarding supported TDPs (and PPTs) of the new socket.

AMD did state the chip was NOT running at a 170W TDP (230w PPT).

It remains to be seen where everything will land. That chip could be a chip with a 125W TDP (168.75W PPT). If so, that means they have the option to scale up in the future.

It is also AMD basically saying to Intel: "We can use just as much power as you, so please feel free to continue ignoring perf/watt."

Hopefully Zen 4 will bring a reckoning to Intel. Intel needs another Core 2 moment. AMD continues to fire on all cylinders. Zen 4 looks to be a monster.

40% compared to what? 12900K? 5950x? What power limits? Sorry this number is meaningless without any context. This is damage control after his big TDP fail what he is doing.

AMD is going to demonstrate massive multicore gains because the 5950x was power limited. By giving the Zen 4 chip more power, more cores can operate at higher clocks, so IF they roll out a 230W PPT (an increase of over 61%) chip you will see absolutely massive gains. Combined with the node shrink and IPC gains it is possible some workloads could be > 100%. I'm not saying that will be the case, but it is within the realm of possibility. By tweaking my own 5950x (raising PPT + undervolting) I've been able to squeeze quite a bit more performance out of my own 5950x (though I operate at stock speeds due to this machine being used for production work).

 

[Kedas]

Moore's Law Is Dead said 7-9% IPC

That's not bad, if it would be only a shrink it would be 0% IPC
N5P gives probably +20% clock (vs 7nm), add some extra power budget to 125W and you get your +40% multi thread 16 core over Zen3.
And Vcache version still has to come on top of that.

Do you really care that much how they got to this +40%?
intel had a tik-tok strategy, nobody subtracted benchmarks points because it was mainly an advanced intel die node that gave them the fastest CPU. The fastest CPU gets the price anyway. (assuming you don't hide a 1000W chiller under the table)

Be happy that AMD can hit high clocks and even increase IPC in the same generation and add an instruction set and make a new platform all at the same time. For Zen5 they can put more resources on the core again.
Doesn't it mean that Zen5 is being designed by the Zen 3 team?

Also we don't have any idea about how good their AVX512 implementation is. Since intel doesn't have it any-more on the desktop market, some benchmarks will be 'weird' in favour of AMD.
intel dropped it due to a bad implementation.
I call it 'AVX512' but I'm starting to think AMD will give it a new name like AMD512 or something to stay away from the bad reputation linked to AVX512. (like intel renamed AMD64 to intel64 for their CPUs, also to remove the obvious association)

The platform power limit had to be increased, would you let intel win just because they max out on power....
AMD can still release the most power efficient CPU and have an SKU taking the top spot also now.

 

[JoeRambo]

Golden cove is only brutal to my air conditioning and power bill. It most certainly does not beat my 5950x in my workloads.

NOTE: A 64 core Threadripper or EPYC consumes less power than a 12900k/12900ks. Those chips are significantly faster as well.

64 core Threadrippers are very relevant to discussion about Zen4 6 and 8 core desktop chips and their Intel's counterparts . Probably a lot of potential buyers are cross shopping them, no doubt about that. /s
Bringing 12900K "efficiency" to every ADL discussion is getting old as well. In general, testing other SKUS like 12700K @ Igor and other places found Alder Lake to be more efficient during gaming than most efficient of original release Zen3 SKUs -> 5600x. But people are free to point out 5800X3D on $50 A320 and $50 OEM DDR4 ram has outstanding perf and efficiency as well.
Sadly none of above aplies to Zen4 vs Raptor lake discussion and esp the part of it where AMD 6 and 8 CPUs will have same priced counterparts with 4-8 extra efficiency cores and will need as much clocks as possible to not fall badly behind in throughput tests.

 

[maddie]

64 core Threadrippers are very relevant to discussion about Zen4 6 and 8 core desktop chips and their Intel's counterparts . Probably a lot of potential buyers are cross shopping them, no doubt about that. /s
Bringing 12900K "efficiency" to every ADL discussion is getting old as well. In general, testing other SKUS like 12700K @ Igor and other places found Alder Lake to be more efficient during gaming than most efficient of original release Zen3 SKUs -> 5600x. But people are free to point out 5800X3D on $50 A320 and $50 OEM DDR4 ram has outstanding perf and efficiency as well.
Sadly none of above aplies to Zen4 vs Raptor lake discussion and esp the part of it where AMD 6 and 8 CPUs will have same priced counterparts with 4-8 extra efficiency cores and will need as much clocks as possible to not fall badly behind in throughput tests.

The truth does not get old. Only for those who need to mislead.