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!

 

[maddie]

Slightly worse, because Raptor Lake is going to be better than Alder Lake and AMD has yet to show any of their CPUs beating ADL in single threaded performance. Zen 4 may be forced to consume just a bit more power to keep Raptor Lake away at a safe distance in MT workloads. However, I think Raptor Lake may still turn out to be the ST king. Intel will defend their territory ferociously, power be damned!

"However, I think Raptor Lake may still turn out to be the ST king. Intel will defend their territory ferociously, power be damned!"

Sure you know how this works?

 

[igor_kavinski]

Sure you know how this works?

I think Raptor Lake will have the highest GB5 ST score.

 

[Hans Gruber]

I thought I read the 5nm (TSMC) process yields a 30% power reduction. So a 7950x would in theory run 30% lower power clock for clock vs. a 5950x. If you consider higher clock speeds at least a 10-15% reduction in real world power usage vs. Zen 3. A lot of people think because the power delivery to the AM5 boards means that somehow Zen 4 CPU's will produce more power. Most do not see the increase in power handling a precursor to more cores for Zen 4 down the road. I imagine a Zen4+ type of release in 2023 with 24 or 32 core CPU's.

 

[moinmoin]

I thought I read the 5nm (TSMC) process yields a 30% power reduction. So a 7950x would in theory run 30% lower power clock for clock vs. a 5950x. If you consider higher clock speeds at least a 10-15% reduction in real world power usage vs. Zen 3. A lot of people think because the power delivery to the AM5 boards means that somehow Zen 4 CPU's will produce more power. Most do not see the increase in power handling a precursor to more cores for Zen 4 down the road. I imagine a Zen4+ type of release in 2023 with 24 or 32 core CPU's.

AMD customizes the nodes and has targets of its own.

This slide has been confirmed to refer only to the node change. Things like IPC improvements in Zen 4 are said to be on top of that.

 

[maddie]

I think Raptor Lake will have the highest GB5 ST score.

Possibly, but what does power be damned have to do with this?

 

[igor_kavinski]

Possibly, but what does power be damned have to do with this?

Meaning Intel can go ahead and raise the TDP even more. AMD likely won't (probably can't). So it's up to Intel to push the ST boundary further.

 

[Thibsie]

Thanks. I think I did see this a while ago. 8 channels is a little odd considering Genoa at 12, but it would be easy to do by just connecting 2 channels per quadrant. There is no guarantee we get this as a regular Threadripper part though.

6 channels I think.

 

[eek2121]

I think Robert Hallock confirmed that Ryzen 7000 would, yes.

I think it's a given that the 16 core will be 170. The 8 and 12 won't be.

i think i read he confirmed 16 cores max for the 7000 generation, so even if he did not specify that PPT for CPU with up to 16 cores, since there is not going to be more cores than that, what else could it be than 16c product?

except future products like zen 5, ryzen 8000 potentially with more cores than 16, but thats anyones guess at this point.

He confirmed no such thing. He did say that the demo used less power than the limit.

Amd have to worry because the e-cores are getting for l2 plus clock speed increase. That 12-15 ipc on the e-core

The e-cores aren’t receiving much of a bump in frequency. Note that currently, the e-cores contribute between 20 and 30% of most multicore workloads. I would be extremely surprised if the overall multicore uplift is more than 35%. If you ignored power limits and simply doubled the e-core count of alder lake, the uplift would be around 15-25%. Intel can’t work magic, and they have a max of 233w PL2 to work with. I doubt AMD has anything to worry about.

 

[lobz]

I doubt that AMD will be pushing it too far up the frequency / power curve. If it is significantly more power efficient at any clock speed that Zen 3 can run at but it can clock significantly higher, and therefore pull more power, would you still classify that as less efficient than Zen 3?
I don’t think I would call that less efficient. Yeah, your performance per watt is likely lower at the extreme end, but your absolute performance would still be significantly higher.

e x a c t l y

 

[lobz]

Slightly worse, because Raptor Lake is going to be better than Alder Lake and AMD has yet to show any of their CPUs beating ADL in single threaded performance. Zen 4 may be forced to consume just a bit more power to keep Raptor Lake away at a safe distance in MT workloads. However, I think Raptor Lake may still turn out to be the ST king. Intel will defend their territory ferociously, power be damned!

I really hope you'll still be around in these threads to see how much more efficient Zen4 is compared to Zen3.

 

[lobz]

Amd have to worry because the e-cores are getting for l2 plus clock speed increase. That 12-15 ipc on the e-core

Yet everyone seems to think that this will come for free for Intel regarding Wattage? Sure.

 

[lobz]

I thought I read the 5nm (TSMC) process yields a 30% power reduction. So a 7950x would in theory run 30% lower power clock for clock vs. a 5950x. If you consider higher clock speeds at least a 10-15% reduction in real world power usage vs. Zen 3. A lot of people think because the power delivery to the AM5 boards means that somehow Zen 4 CPU's will produce more power. Most do not see the increase in power handling a precursor to more cores for Zen 4 down the road. I imagine a Zen4+ type of release in 2023 with 24 or 32 core CPU's.

Let's put it this way. Using the same power, Zen4 will be much more performant than Zen3.

 

[lobz]

Meaning Intel can go ahead and raise the TDP even more (...)

for being 'ST King'? How on Earth would that help??

 

[igor_kavinski]

for being 'ST King'? How on Earth would that help??

It would help in games and poorly threaded tasks. Thing about MT is, you can get more and more performance by adding cores and improving the inter-core communication latency but improving ST takes a whole lot more effort. The average person has no use for more than 8 cores / 16 threads anyway. It's the ST that should be focused on and is the most difficult to improve upon.

 

[biostud]

Meaning Intel can go ahead and raise the TDP even more. AMD likely won't (probably can't). So it's up to Intel to push the ST boundary further.

The joker being vcache that does nothing for geekbench, but quite a lot for gaming.

 

[igor_kavinski]

The joker being vcache that does nothing for geekbench, but quite a lot for gaming.

Intel can put their eDRAM on an ADL die too. It may not be as effective or as fast as V-cache but it would help a lot in games that really crave the extra cache. Unfortunately, Intel is either too shy to play fire with fire or eDRAM is just too expensive for them.

 

[krumme]

Amd have said Zen 5 will give the same performance increase as from bd to zen1. Sounds insane to me, but if zen 4 is viewed as the excavator of the zen line, i think its pretty darn performant.

 

[maddie]

Meaning Intel can go ahead and raise the TDP even more. AMD likely won't (probably can't). So it's up to Intel to push the ST boundary further.

Look, single thread is not constrained by any modern desktop CPU of ~65W or more. Do you think a 200W+ power limit would mean the 1 core can use that? The reason you get higher 1 thread performance on higher count CPUs is due to better silicon, thus clocks.

Higher power is for multi-thread work.

 

[lobz]

It would help in games and poorly threaded tasks. Thing about MT is, you can get more and more performance by adding cores and improving the inter-core communication latency but improving ST takes a whole lot more effort. The average person has no use for more than 8 cores / 16 threads anyway. It's the ST that should be focused on and is the most difficult to improve upon.

I guess you didn't fully understand my question. How would more power help Intel in one of the only areas where they don't have any power consumption issues whatsoever?

 

[igor_kavinski]

For Raptor Lake, higher ST perf would be aided by higher clocks and higher turbo boosting which would increase power draw considerably. If Zen 4 has a more efficient architecture, it should offer even more ST uplift at comparatively lower clocks but so far, AMD has not provided any evidence of that. If they were confident about their ST performance, they would have no problem trumpeting it. Also, if 5nm can't let them claim the ST performance crown, they left something to be desired in their design.

 

[Abwx]

for being 'ST King'? How on Earth would that help??

Just before Zen 1 was launched i stated that once AMD was to be competitive ST wise we could expect that an eventual intel advantage of 10% in this metric would be hyped as if it was 100%.

There s really nothing left other than non significant ST differences to try promoting a failed concept, i just looked at Computerbase 12900K review and it take 240W for ADL to barely beat a 5950X@125W in Cinebench, overall the latter is still faster in MT.

 

[biostud]

Intel can put their eDRAM on an ADL die too. It may not be as effective or as fast as V-cache but it would help a lot in games that really crave the extra cache. Unfortunately, Intel is either too shy to play fire with fire or eDRAM is just too expensive for them.

As long as they don't it is quite the hypothetical statement. But sure...

 

[gruffi]

Amd have to worry because the e-cores are getting for l2 plus clock speed increase. That 12-15 ipc on the e-core

Doubt that. You don't get 8 more E-cores for free. THG measured ~50W for 8 ADL E-cores. More cache also needs more energy. Which might negatively affect clock speed. But is often quite useless in highly multi threaded apps. Look at 5800X3D which lost like 3-5% MT performance compared to 5800X. So, we might see even some lower clocks on RPL at 240W. Of course there is still the possibility to increase the power budget. But that likely won't make RPL more efficient.

Yeah, I'm beginning to think there's a good reason why Patty said AMD's in the rear view mirror. They probably got their hands on a ZEN 4 engineering sample, extrapolated its performance and reached the conclusion that they can easily beat it, at least in desktops.

Doubt that too. When did Gelsinger say that? Last year after ADL launch? Even AMD don't seem to have enough performance data of Ryzen 7000 right now. Hallock recently said they are collecting Ryzen 7000 data in the upcoming weeks for a detailed breakdown of performance, efficiency, clock speed, IPC, and so on. That's the reason why they wrote ">15% Single-Thread" on that Computex slide. They are sure it will be at least 15%. But it could turn out to be higher after collecting enough performance data. Though I'm not sure if Hallock is messing up some stuff again. He also said in that interview that already 15% would be quite good because Ryzen 5000 increased Cinebench ST performance not more than 10%. Which is definitely not true. Reviews showed more like >20% between 3950X and 5950X.

I thought I read the 5nm (TSMC) process yields a 30% power reduction. So a 7950x would in theory run 30% lower power clock for clock vs. a 5950x.

That's the standard 5nm process, or N5. AMD is using an improved 5nm process. On CES early this year they showed a slide with 50% power reduction at the same clock speed or >25% higher clock speed at the same power. But of course that's quite theoretical. We don't know what power requirements the Zen 4 design will have.

But it's not just the 5nm core chiplets. For me one of the most interesting parts considering power will be the new "low power" 6nm IOD. Especially at light load the current 12nm IOD needs too much power, torpedoing the excellent power efficiency of the Zen 3 cores.

 

[lobz]

For Raptor Lake, higher ST perf would be aided by higher clocks and higher turbo boosting which would increase power draw considerably. If Zen 4 has a more efficient architecture, it should offer even more ST uplift at comparatively lower clocks but so far, AMD has not provided any evidence of that. If they were confident about their ST performance, they would have no problem trumpeting it. Also, if 5nm can't let them claim the ST performance crown, they left something to be desired in their design.

You still don't get it. ST is the abbreviation for single threaded. Increasing the clockspeed on ONE SINGLE THREAD arbitrarily high would run into thermal limitations waaaaaaay-way earlier than any power limit in our modern day 6-8-12-16+ core processors.

 

[igor_kavinski]

Increasing the clockspeed on ONE SINGLE THREAD arbitrarily high would run into thermal limitations waaaaaaay-way earlier than any power limit in our modern day 6-8-12-16+ core processors.

I get it. Increasing ST perf shouldn't increase the TDP much as that relates more to MT. Thanks for the correction.