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!

 

[eek2121]

Hoping AMD gives us more flexible options when it comes to PBO. I would love to be able to set power limits on a per-CCD basis, for example.

Raptor Lake is DOA to me due to high power consumption.

 

[nicalandia]

Hoping AMD gives us more flexible options when it comes to PBO. I would love to be able to set power limits on a per-CCD basis, for example.

Raptor Lake is DOA to me due to high power consumption.

You could purchase the non-K version 13900, but it's quite handicapped due to power limit

Gigabyte Technology Co., Ltd. Z690 GAMING X - Geekbench

Benchmark results for a Gigabyte Technology Co., Ltd. Z690 GAMING X with an Intel Core i9-13900 processor.

browser.geekbench.com

 

[Joe NYC]

I always read this claim. As someone who owned & ran a B<>B production business, be very wary of anyone who says I'll buy all you can produce.

Does anyone know the approximate size of the server market in say, cores/yr, NOT revenue?

IIRC, the server market is about 3 million servers per quarter, majority of them 1 to 2 CPUs

 

[coercitiv]

You could purchase the non-K version 13900, but it's quite handicapped due to power limit

Nope, only buy K CPUs for the unlocked SA voltage. Proper TDP config is much better attained through manual settings.

 

[Joe NYC]

when and where did you pick-up such a noble skill?

By being born in the next door Slovakia.

 

[eek2121]

Unsure why people seem to think Raptor Lake will be significantly faster. The 7900x and the 13900k will have the same number of threads. Zen 4 looks to have pretty much equal IPC with the 13900k's P-Cores, while having much greater performance than the E-Cores. On a per-thread basis alone, AMD should be able to edge the Intel chip out.

We will see how it plays out, but even if the Zen 4 cores were slightly slower than the P-Cores, the number of big cores would more than make up for it.

 

[MarkPost]

I'm not saying that it is unfair to measure 7-zip decompress. I'm saying that after a point when you have sufficient performance it doesn't matter because multi-core CPUs are already fast enough to execute it with reasonable speed. People's workflow doesn't involve compressing and decompressing all day.

On the other hand, the only reason we are discussing graphs from Computerbase is because half of their suite consists of render applications, which has been a strong point of AMD since Zen. You can cherry pick all you want, but I don't consider Computerbase suite of benchmarks a representative one because it skews heavily toward one type workload and it doesn't have any benchmark of the Adobe suite at all. People doing work with their computers do not run Cinebench, POV-ray and Corona renderer.

Anyway, to remove bias you should be considering the average of averages across all websites that do benchmarking. Fortunately, 3dcenter has compiled the data for such a comparison, and the result is that the 12600K is slightly faster than the 5800X.

Launch-Analyse Intel Alder Lake (Seite 2) | 3DCenter.org

Sonntag, 7. November 2021 / von Leonidas Im Vorfeld der Alder-Lake-Benchmarks gab es einige Bedenken, ob man so einfach vergleichbare Benchmark-Zahlen würde erreichen können. Schließlich gab es hierzu die Störfaktoren der je nach

www.3dcenter.org

Actually Cinebench doesnt favour Ryzen precisely. And PovRay directly hurts Ryzen perfomance due to the fact that when a Ryzen is detected, PovRay skips AVX2 use (only uses AVX) which harms Ryzen perf significantly.

 

[Timmah!]

Unsure why people seem to think Raptor Lake will be significantly faster. The 7900x and the 13900k will have the same number of threads. Zen 4 looks to have pretty much equal IPC with the 13900k's P-Cores, while having much greater performance than the E-Cores. On a per-thread basis alone, AMD should be able to edge the Intel chip out.

We will see how it plays out, but even if the Zen 4 cores were slightly slower than the P-Cores, the number of big cores would more than make up for it.

I would hazard a guess its all down to different people having different views on what "significantly" faster means. If CB23 ST score for RPL is 2300 and for Zen4 its 2100, is it to be considered "pretty much equal"?

 

[Joe NYC]

I always read this claim. As someone who owned & ran a B<>B production business, be very wary of anyone who says I'll buy all you can produce.

Does anyone know the approximate size of the server market in say, cores/yr, NOT revenue?

BTW, with 60,000 N5 wafers (which is rumored to be AMD allocation for Q4), AMD could make 43 million good Zen 4 dies if the entire capacity were to be allocated to CPU. Just a theoretical number.

Desktop CPU market? The whole PC market (desktop + notebook) is about 300 million units per year, 75m per quarter. I don't know the percentage of desktop, but perhaps 25% or less. So perhaps 15 million desktop CPUs per quarter?

As I said, about, about 3 million servers with 1 to 2 CPUs.

Of course, some percentage of the N5 capacity will be devoted to GPU. And a tiny percentage to some niche products.

But by the time all this production makes it to the market, in Q1 2023, I don't think AMD will be capacity constrained, to the point where capacity of 7800x3d would have to be sacrificed for something else.

It may in fact work the other way around. Other, desktop CPUs may be sacrificed to make more 7800x3d, if demand is there. 7800x3d at $450 would be more profitable than $7700x at $300.

 

[Kaluan]

I'm not saying that it is unfair to measure 7-zip decompress. I'm saying that after a point when you have sufficient performance it doesn't matter because multi-core CPUs are already fast enough to execute it with reasonable speed. People's workflow doesn't involve compressing and decompressing all day.

On the other hand, the only reason we are discussing graphs from Computerbase is because half of their suite consists of render applications, which has been a strong point of AMD since Zen. You can cherry pick all you want, but I don't consider Computerbase suite of benchmarks a representative one because it skews heavily toward one type workload and it doesn't have any benchmark of the Adobe suite at all. People doing work with their computers do not run Cinebench, POV-ray and Corona renderer.

Anyway, to remove bias you should be considering the average of averages across all websites that do benchmarking. Fortunately, 3dcenter has compiled the data for such a comparison, and the result is that the 12600K is slightly faster than the 5800X.

Launch-Analyse Intel Alder Lake (Seite 2) | 3DCenter.org

Sonntag, 7. November 2021 / von Leonidas Im Vorfeld der Alder-Lake-Benchmarks gab es einige Bedenken, ob man so einfach vergleichbare Benchmark-Zahlen würde erreichen können. Schließlich gab es hierzu die Störfaktoren der je nach

www.3dcenter.org

Not a huge fan of the meta review round-up idea. Someone of the outlets or specific reviews there are terrible, but anyway.
As you may have noticed, almost all of the 30 reviews listed tested 12600K w/ DDR5... and as I've said, it's likely a DDR5 setup comes out slightly on top. Lo and behold, it did, by an irrelevant (margin of error for most people) 2,4%. So if anything, this only proves my point. GB is not a good metric to rely on for predicting overall performance and 12600K/ADL didn't smash anything, contrary to the false hype leading up to ADL launch (12900K doing 2050/30,5K R23, 810/11800 in R20 or 12600K doing 7400 in R20 anyone?)... and apparently to the belief of some, to this day.

And I'm getting a sense a few of the Raptor Lake leaks are very misleading as well. I would say "can't wait for the lifting of NDAs/reviews so we can finally get a solid grasp of things", but I already know when next-next gen will be around the corner, we'll get the same story all over.

 

[eek2121]

I would hazard a guess its all down to different people having different views on what "significantly" faster means. If CB23 ST score for RPL is 2300 and for Zen4 its 2100, is it to be considered "pretty much equal"?

Considering CBr23 is just one benchmark (out of dozens), does it matter? If Ryzen wins at say: the majority of gaming benchmarks, but loses in CBr23, does that make it faster or slower than RPL-S?

I honestly suspect the Intel chip will win some, lose more, and ultimately lose the perf/watt battle.

People should honestly care about power consumption more than they do. Especially if you are in Europe. Between energy shortages and climate change, the less power your PC consumes, the better.

 

[Det0x]

Actually Cinebench doesnt favour Ryzen precisely. And PovRay directly hurts Ryzen perfomance due to the fact that when a Ryzen is detected, PovRay skips AVX2 use (only uses AVX) which harms Ryzen perf significantly.

I dont think Cinebench is exactly bad for Zen 3 and 4 per se, its rather that it dont care about memory(subsystem) performance and hence dont show Zen4 new strong points other than the increased clockspeed.

While at the same time its a bestcase example for Intel's P+E cores architecture.. Since the benchmark dont care about memory bandwidth/latency and have minimum thread-scheduling this is one of the (few) places where the E-cores are actually useful and the P-cores are not handicapped by a quasi DDR4/DDR5 memory controller combo. (many people i know with Alder Lake simply disable E-cores because they gimp performance in games with its thread-scheduling problems together with forced lower core/uncore speed)

I would hazard a guess its all down to different people having different views on what "significantly" faster means. If CB23 ST score for RPL is 2300 and for Zen4 its 2100, is it to be considered "pretty much equal"?

I have no doubt Raptor Lake will win in Cinebench R23 ST benchmarks, and your numbers above seems plausible to me. (~2300 vs ~2100).
But at the same time, i think Zen4 will win the majority of other benchmarks suites

Who have the the fastest CPU ? Well it depends on point of view.. Raptor Lake if all you do is play Cinebench ST all day long

Unsure why people seem to think Raptor Lake will be significantly faster. The 7900x and the 13900k will have the same number of threads. Zen 4 looks to have pretty much equal IPC with the 13900k's P-Cores, while having much greater performance than the E-Cores. On a per-thread basis alone, AMD should be able to edge the Intel chip out.

We will see how it plays out, but even if the Zen 4 cores were slightly slower than the P-Cores, the number of big cores would more than make up for it.

They do not have the same amounts of threads ^^

The comparisons should be between the following cpus eventho some of the guys from IDL is scared of the 7950x and only want to compare 13900vs7900 and whatnot since "name'ing":

• 13900k(8P+16E 32t) vs 7950x(16c 32t)
• 13700k(8P+8E 24t) vs 7900x(12c 24t)
• 13600k(6P+8E 20t) vs 7700x(8c16t)
• 13500(6P+8E 20t)
• 13400(6P+4E 16t) vs 7600x (6c12t)

This could possibly mean that its on the low-end intel will do the best vs Zen4 this generation, if you only play Cinebench 24/7
Pricing will decide the winner below the 13700k i guess

Do note 13600, 13500 and 13400 is based on (refreshed) Alder Lake silicon, not the new Raptor lake

 

[nicalandia]

On Geekbench Raptor Lake is in for a Beating. The e cores latencies and lack of AVX512 will be against them.

 

[maddie]

BTW, with 60,000 N5 wafers (which is rumored to be AMD allocation for Q4), AMD could make 43 million good Zen 4 dies if the entire capacity were to be allocated to CPU. Just a theoretical number.

Desktop CPU market? The whole PC market (desktop + notebook) is about 300 million units per year, 75m per quarter. I don't know the percentage of desktop, but perhaps 25% or less. So perhaps 15 million desktop CPUs per quarter?

As I said, about, about 3 million servers with 1 to 2 CPUs.

Of course, some percentage of the N5 capacity will be devoted to GPU. And a tiny percentage to some niche products.

But by the time all this production makes it to the market, in Q1 2023, I don't think AMD will be capacity constrained, to the point where capacity of 7800x3d would have to be sacrificed for something else.

It may in fact work the other way around. Other, desktop CPUs may be sacrificed to make more 7800x3d, if demand is there. 7800x3d at $450 would be more profitable than $7700x at $300.

Agree that AMD might not be capacity constrained as previous years, if at all. Marketshare time gains are here it seems and pricing will have to play a part as max performance in both main lines of product, CPUs & GPUs, will be close to competitors offerings.

Pricing will be less than popularly assumed.

 

[Mopetar]

People should honestly care about power consumption more than they do. Especially if you are in Europe. Between energy shortages and climate change, the less power your PC consumes, the better.

I don't know. With the heating fuel shortage, it might be better to get an Intel system or two, which could keep a small apartment warm enough to stop a person from freezing to death.

Maybe Larry can lend a few of his mining rigs. Even though the electricity is expensive, the mining would offset the most or all of the cost.

 

[Rigg]

I don't think I'm the first to mention it, but Intel is raising prices. AMD knows this, too, and can price accordingly.

Intel is increasing prices up to 20%. That's kind of vague and surely will vary depending on SKU. We'll find out what AMD's MSRP's will be on Monday. If the leak is true, and AMD ask $800 for 5950x/$600 for 5900x, then Intel has more than 20% room to grow in those top tiers. At the same time Intel can't really price the 13600k above the 7700x unless it's just strait up beating it. That means it almost surely will have an MSRP somewhere between $230 - $300.

 

[MarkPost]

I dont think Cinebench is exactly bad for Zen 3 and 4 per se, its rather that it dont care about memory(subsystem) performance and hence dont show Zen4 new strong points other than the increased clockspeed.

While at the same time its a bestcase example for Intel's P+E cores architecture.. Since the benchmark dont care about memory bandwidth/latency and have minimum thread-scheduling this is one of the (few) places where the E-cores are actually useful and the P-cores are not handicapped by a quasi DDR4/DDR5 memory controller combo. (many people i know with Alder Lake simply disable E-cores because they gimp performance in games with its thread-scheduling problems together with forced lower core/uncore speed)


I have no doubt Raptor Lake will win in Cinebench R23 ST benchmarks, and your numbers above seems plausible to me. (~2300 vs ~2100).
But at the same time, i think Zen4 will win the majority of other benchmarks suites

Who have the the fastest CPU ? Well it depends on point of view.. Raptor Lake if all you do is play Cinebench ST all day long

They do not have the same amounts of threads ^^

The comparisons should be between the following cpus eventho some of the guys from IDL is scared of the 7950x and only want to compare 13900vs7900 and whatnot since "name'ing":

• 13900k(8P+16E 32t) vs 7950x(16c 32t)
• 13700k(8P+8E 24t) vs 7900x(12c 24t)
• 13600k(6P+8E 20t) vs 7700x(8c16t)
• 13500(6P+8E 20t)
• 13400(6P+4E 16t) vs 7600x (6c12t)

This could possibly mean that its on the low-end intel will do the best vs Zen4 this generation, if you only play Cinebench 24/7
Pricing will decide the winner below the 13700k i guess

Do note 13600, 13500 and 13400 is based on (refreshed) Alder Lake silicon, not the new Raptor lake
View attachment 66646

Yes, thats the idea about Cinebench. It doesnt hurt Ryzen perf per se (PovRay does). Its just that ADL/RKL features and Cinebench test fits nicely one with the each other. I only wanted to pointing out that Tams disliked a review just because there were several renders, just to fit his agenda, when ironically CB fits better with actual Intel CPUs, and PovRay directly hurts Ryzen perf badly not using AVX2 instruction set, just AVX (on the other hand, PovRay uses AVX2 when an Intel CPU with AVX2 support is detected).

 

[Timmah!]

By being born in the next door Slovakia.

Then you are my fellow countryman. Based on your nick, should i assume you live in NYC now?

 

[Timmah!]

Considering CBr23 is just one benchmark (out of dozens), does it matter? If Ryzen wins at say: the majority of gaming benchmarks, but loses in CBr23, does that make it faster or slower than RPL-S?

I honestly suspect the Intel chip will win some, lose more, and ultimately lose the perf/watt battle.

People should honestly care about power consumption more than they do. Especially if you are in Europe. Between energy shortages and climate change, the less power your PC consumes, the better.

People just like to simplify things and as a result of this CB23 measurements somehow became representative of overall CPU perfomance, even if there is plenty cases, which they dont represent, like games. It is, what it is, its natural human tendency i guess.

That said, i did not mean to imply with my post that i am the one of people who would consider hypothetical 5 percent better score of RPL over Zen4 in CB23 to be significant. Actually, its the other way around, i would rather take 5 percent slower CPU that draws 100W less, so i intend to go with AMD this generation.

 

[Markfw]

People just like to simplify things and as a result of this CB23 measurements somehow became representative of overall CPU perfomance, even if there is plenty cases, which they dont represent, like games. It is, what it is, its natural human tendency i guess.

That said, i did not mean to imply with my post that i am the one of people who would consider hypothetical 5 percent better score of RPL over Zen4 in CB23 to be significant. Actually, its the other way around, i would rather take 5 percent slower CPU that draws 100W less, so i intend to go with AMD this generation.

In total benchmarks (which we may not see until mid to late Sept) I actually expect to see the 7950x win most of them, but yes, the power usage is a big deal for me.

 

[APU_Fusion]

Is there a suspected release date for the 7950x3d other than “early 2023”? I haven’t built a new PC since 2013 (rocking water cooled 4790k lol). I game and Photoshop primarily. Looks at wallet and groans in terror 😁

 

[ondma]

Unsure why people seem to think Raptor Lake will be significantly faster. The 7900x and the 13900k will have the same number of threads. Zen 4 looks to have pretty much equal IPC with the 13900k's P-Cores, while having much greater performance than the E-Cores. On a per-thread basis alone, AMD should be able to edge the Intel chip out.

We will see how it plays out, but even if the Zen 4 cores were slightly slower than the P-Cores, the number of big cores would more than make up for it.

It is not that simple. First you have 8 "P" cores with hyperthreading for both. I suspect RL big core will be a bit faster than Zen 4, but for the sake of argument, lets call them equal.
So 8 big Zen cores vs 8 big RL cores, call it a wash.
So you have left:
8 big Zen cores vs 8 E cores -- obvious Zen wins
But you still have 8 real E cores vs 8 Zen threads from hyperthreading. Advantage should be to Intel.
So the final performance will depend on how 16 real E cores will perform against 8 hyperthreaded Zen cores.

 

[Markfw]

It is not that simple. First you have 8 "P" cores with hyperthreading for both. I suspect RL big core will be a bit faster than Zen 4, but for the sake of argument, lets call them equal.
So 8 big Zen cores vs 8 big RL cores, call it a wash.
So you have left:
8 big Zen cores vs 8 E cores -- obvious Zen wins
But you still have 8 real E cores vs 8 Zen threads from hyperthreading. Advantage should be to Intel.
So the final performance will depend on how 16 real E cores will perform against 8 hyperthreaded Zen cores.

While your logic does have some merit, There are 1,000 variables out there, like core to core and core to small core latency with the interconnects, memory controllers, etc...

And of course how fast will Zen 4 really go ? and how long will Raptor lake be able to sustain full speed due to power and heat ???

Just for power alone (from what I read) I will take a Zen 4, unless things do not turn out anything like we have been reading.

The next month or so is going to be epic (or should I say EPYC)

 

[nicalandia]

It is not that simple. First you have 8 "P" cores with hyperthreading for both. I suspect RL big core will be a bit faster than Zen 4, but for the sake of argument, lets call them equal.
So 8 big Zen cores vs 8 big RL cores, call it a wash.
So you have left:
8 big Zen cores vs 8 E cores -- obvious Zen wins
But you still have 8 real E cores vs 8 Zen threads from hyperthreading. Advantage should be to Intel.
So the final performance will depend on how 16 real E cores will perform against 8 hyperthreaded Zen cores.

16 e cores are more or less equal to 8c/16T P cores but that will depend on apps, but most of the time they are somewhat less

 

[Saylick]

16 e cores are more or less equal to 8c/16T P cores but that will depend on apps, but most of the time they are somewhat less

That's about right. Someone on these forums did a head to head comparison between the 12900K and the 5950X as you ramp up the thread count, and it showed that at certain thread counts, having half as many big cores was better than more E cores. Note, it was just based on approximations and not based on a benchmark.