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!

 

[Exist50]

Just FYI as to why we have to ask, there is a 125W processor group (w/170W PPT) as well. So uh... it's really not very clear ATM.

Is there? AMD's statement only mentions 65W and 105W.

This new TDP group will enable considerably more compute performance for high core count CPUs in heavy compute workloads, which will sit alongside the 65W and 105W TDP groups that Ryzen is known for today.

 

[PJVol]

There is a LOT more headroom available in MT loads than ST loads.

I'm more interested in what the EDC limit for the 16-core part is. Or is there is a new metric introduced with Zen4 (peak PPT)

 

[uzzi38]

Is there? AMD's statement only mentions 65W and 105W.

Yeah, definitely. It was brought up a few days ago by Skyjuice, who's articles you should probably be reading (seriously, follow him on Twitter).

Anyway, the Gigabyte documentation is old, and so the values aren't really up to date, but you can still see a 120W group here too.

https://media.discordapp.net/attachments/807442045068640296/979383271769112586/AMD-AM5-Socket-TPD.png

 

[Timorous]

Downside being that efficiency goes down the drain.

That is what I would have thought but then Sony released an RDNA2 games console with a 2.23Ghz GPU clock in a console TDP so no idea what magic AMD are doing but they seem to be finding ways to increase clock speed without needing tons more power.

 

[igor_kavinski]

https://www.tomshardware.com/news/amd-corrects-socket-am5-for-ryzen-7000-power-specs-230w-peak-power-170w-tdp

AMD issued the following to Tom's Hardware:

"AMD would like to issue a correction to the socket power and TDP limits of the upcoming AMD Socket AM5. AMD Socket AM5 supports up to a 170W TDP with a PPT up to 230W. TDP*1.35 is the standard calculation for TDP v. PPT for AMD sockets in the “Zen” era, and the new 170W TDP group is no exception (170*1.35=229.5).

230W peak! Looks like those 5.5+ GHz frequencies don't come for free.

 

[RTX]

There's 3-4 versions of the retention clips for AM5? Crosshair Extreme is slightly different than the others in the first image.

https://images.anandtech.com/doci/17399/AMD%20Computex%202022%20Slide%20Deck%2012.JPG

https://videocardz.com/ezoimgfmt/cdn.videocardz.com/1/2022/05/MSI-Project-ZERO-3.jpg?ezimgfmt=ng%3Awebp%2Fngcb1%2Frs%3Adevice%2Frscb1-1

https://videocardz.com/ezoimgfmt/cdn.videocardz.com/1/2022/05/MSI-MEG-UNIFY.jpg?ezimgfmt=ng%3Awebp%2Fngcb1%2Frs%3Adevice%2Frscb1-1

 

[uzzi38]

That is what I would have thought but then Sony released an RDNA2 games console with a 2.23Ghz GPU clock in a console TDP so no idea what magic AMD are doing but they seem to be finding ways to increase clock speed without needing tons more power.

2.23GHz is gonna look weak this year lol, but we're getting a bit off topic here lol

 

[gdansk]

So it's actually 230W? That won't work well in my machine. That's Threadripper territory. I hate this trend but it remains to be seen if the performance increase is worth ~90W more.

 

[Timorous]

2.23GHz is gonna look weak this year lol, but we're getting a bit off topic here lol

Yea but the point is prior to launch I scoffed at that clock speed as not possible and a console TDP. I was wrong then and I won't make the mistake to assume that AMD are cranking clock speeds at the cost of excesive power consumption.

 

[Abwx]

https://www.tomshardware.com/news/amd-corrects-socket-am5-for-ryzen-7000-power-specs-230w-peak-power-170w-tdp

230W peak! Looks like those 5.5+ GHz frequencies don't come for free.

Some quote about the demo :

So what we want to clarify is that it's a 170 Watt socket power which with AMD, that spec is PPT (Package Power) for us. That doesn't mean that every CPU is going to go up to 170 Watts but it's 30 (Watt) higher than the socket AM4 power cap which was a 142 (watts).

AMD's Robert Hallock & Frank Azor Talk AMD Ryzen 7000 CPUs & AM5 Platform Features: 170W Socket Power Limits, 5.5 GHz Stock Clock Speeds, Smart Access Storage & More

AMD Ryzen 7000 CPUs & AM5 platform features have been detailed by Robert Hallock and Frank Azor during the PCWorld 'The Full Nerd' show.

wccftech.com

 

[igor_kavinski]

AMD's Robert Hallock & Frank Azor Talk AMD Ryzen 7000 CPUs & AM5 Platform Features: 170W Socket Power Limits, 5.5 GHz Stock Clock Speeds, Smart Access Storage & More

AMD Ryzen 7000 CPUs & AM5 platform features have been detailed by Robert Hallock and Frank Azor during the PCWorld 'The Full Nerd' show.

wccftech.com

Not all PCIe 4.0/5.0 NVMe drives will be able to take advantage of Smart Access Storage. Bad for those who already have something good but that just doesn't make the cut for AMD's SAS.

 

[Hitman928]

https://www.tomshardware.com/news/amd-corrects-socket-am5-for-ryzen-7000-power-specs-230w-peak-power-170w-tdp

230W peak! Looks like those 5.5+ GHz frequencies don't come for free.

That's specifically the spec for the socket, not for a particular SKU, i.e. it's unknown if the 16 core CPU AMD demoed is using 230W or 170W peak. I lean towards the demo CPU being 170W peak for two reasons:

1. Robert Hallock specifically said it was 170W PPT.
2. If it was 230W PPT, then that means the CPU shown is significantly less efficient than a 5950x. This would be a big surprise given that the demo CPU is on 5 nm versus a 5950x being on 7 nm. Maybe AMD blew out the F/V curve to hit highest performance possible like Intel, but given their track record with Zen based cores thus far and their emphasis on improved efficiency over Zen3, I don't think this is the case. Time will tell.

 

[moinmoin]

Finally having caught up with all posts amassed over the holidays only now I have suggestions for a couple new threads so we can focus on Zen 4 tech in this thread:

• AMD's sandbagging, does it still work or did they actually drop the ball for real now?
• Maths, how does it work?
• The competitive landscape, what do AMD and Intel have to do give each other a fatal blow?
• Excessive fandom, who exaggerated what when and where?
• The return of tech leak youtubers, real track records and their mysterious disappearance.
• Anything else?

 

[Thunder 57]

It's not a theory. It's a fact. Disabling hyperthreading on coffee lake processors prevented spectre from being a problem. When hyperthreading was enabled, spectre was a problem. The 9900K wasn't affected by Spectre. I can't remember why but as another poster here said. The 9700k doesn't need hyperthreading against the 2700x because it can compete just fine without it. That was an accurate quote but the reason why hyperthreading was disabled for the 9700K had to do with the spectre security issue.

Nonsense, so the 9900k magically didn't have any issue with HT? And I suppose the reason the 2500k, 3570k, etc all lacked HT because of Spectre, which was unknown at that time? And the i7 variants weren't affected? Intel segmentation and marketing seems to be doing their job.

I still firmly believe that AMD still owes us long term customers. More than 10 years of darkness and misery with worthless AMD CPU's. I personally have a desk drawer full of my very old AMD CPU's from the golden era of AMD. Starting with the Athlon 1800+ to the dual Core AMD 64's.

AMD doesn't owe you anything, unless by chance you work there.

I should point out that I currently have AMD Ryzen 5600. That processor was released in April 2022 and I am getting grief from some AMD fanboys in this thread. It should have been released in early 2021.

In this case you could make that argument, but AMD went for higher ASP because they were selling everything they could. If anything they could've priced the launch chips higher. When you say the 5800X3D or Zen 4 should've launched earlier though, that get's annoying. Where is Zen 5 already? Oh, that's right, it's not ready yet.

The reason AMD would increase the TDP on the AM5 socket. To better handle 24core 48thread and 32core 64thread CPU's that AMD will inevitably release during the Zen 4 era at some point.

Where are they going to fit those extra cores?

 

[inf64]

Here is my estimate for top Ryzen 7000 series 16C/32T part, similar to the ones I made before for Zen1/Zen3.

Top 16C/32T SKU 170W: ~14% higher ST clock vs 5950X (4.95GHz->5.65GHz); ~22% higher all-core boost clock( 4.2GHz*->5.15GHz); ~13%** higher IPC vs Zen3 (geo-mean; ~12% integer, 15% in FP).

Gaming: 1.28x or 28% better performance vs 5950X (~12% ipc x ~14% higher effective clock)


ST workloads: 1.29x or 29% better ST performance vs 5950X (~13% ipc x ~14% higher effective clock)


MT workloads: 1.38x or 38% better MT performance vs 5950X (~13% ipc x ~22% higher effective clock)


*4.2GHz is commonly reported all-core boost for 5950X but there are workloads where it will not boost beyond 3.9GHz.
**Initially I wanted to say 15% higher avg IPC, but lowered it a bit in light of very high clock speed targets. 15% in FP gen-on-gen is a trend so far as shown in AT deep dive article here. I think that integer-specific IPC jump will be somewhat lower with Zen4 and this will drag the average down a bit

EDIT: If there is going to be a huge miss when it comes to my estimate, it's likely going to be gaming section. It's possible there is some sort of bottleneck(GPU?) that limits the performance so the scaling will be much lower, similar to how 12900KS has 21% higher ST performance vs 5950X but in games it's only 8% faster.

 

[nicalandia]

Why are people skeptical of the 38,000 MT points in Cine Bench R23 for the 7950X? The stock 5950X get about 26,000 points but it never reaches 4 Ghz all core clocks while doing the MT scores,

Here is a 5950X with an all core boost of 5.4 Ghz that gets 36,000 points.




38,000 Points on a 5 Ghz Zen4 would be easy work.

 

[nicalandia]

Where are they going to fit those extra cores?

I keep asking myself the same question for Bergamo, Where in the heck are they going to place 16C/32T Chiplets on the same substrate?



So The Dragon Range 24C and any Higher than 16 Core Ryzen CPU will use the Zen4c cores that allow 16C/32T per chiplet.

 

[Anhiel]

I'm more interested in what the EDC limit for the 16-core part is. Or is there is a new metric introduced with Zen4 (peak PPT)

According to this: https://www.tomshardware.com/news/a...zen-7000-power-specs-230w-peak-power-170w-tdp
No.
Anyhow, EDC should be dependent on the node design which is still FinFet for N5. Only N2 makes a change. Anyhow, for N5 it seems the voltage range is the same given their posted example on N5/N5P (in 2019) of 0.75 to 1.2V.

TSMC Details 5 nm

TSMC details its 5-nanometer node for mobile and HPC applications. The process features the industry's highest density transistors with a high-mobility channel and highest-density SRAM cells.

fuse.wikichip.org

 

[Hans Gruber]

Nonsense, so the 9900k magically didn't have any issue with HT? And I suppose the reason the 2500k, 3570k, etc all lacked HT because of Spectre, which was unknown at that time? And the i7 variants weren't affected? Intel segmentation and marketing seems to be doing their job.



AMD doesn't owe you anything, unless by chance you work there.



In this case you could make that argument, but AMD went for higher ASP because they were selling everything they could. If anything they could've priced the launch chips higher. When you say the 5800X3D or Zen 4 should've launched earlier though, that get's annoying. Where is Zen 5 already? Oh, that's right, it's not ready yet.



Where are they going to fit those extra cores?

Sam Kinison

There is no hyperthreading on i5 processors back in the 2500k and 3570k era. Only AMD offers hyperthreading aka. SMT on all their CPU's.
Here is anandtech's own article on the Spectre issue on coffee lake.

Analyzing Core i9-9900K Performance with Spectre and Meltdown Hardware Mitigations

www.anandtech.com

 

[Anhiel]

Here is my estimate for top Ryzen 7000 series 16C/32T part, similar to the ones I made before for Zen1/Zen3.

Top 16C/32T SKU 170W: ~14% higher ST clock vs 5950X (4.95GHz->5.65GHz); ~22% higher all-core boost clock( 4.2GHz*->5.15GHz); ~13%** higher IPC vs Zen3 (geo-mean; ~12% integer, 15% in FP).

Gaming: 1.28x or 28% better performance vs 5950X (~12% ipc x ~14% higher effective clock)
ST workloads: 1.29x or 29% better ST performance vs 5950X (~13% ipc x ~14% higher effective clock)
MT workloads: 1.38x or 38% better MT performance vs 5950X (~13% ipc x ~22% higher effective clock)

ST calculation is simple but MT calculation is not like that.
ST is without SMT-2 but MT includes SMT-2.
So the result for MT, given average Zen arch SMT gain should be ~1.72.

 

[inf64]

ST calculation is simple but MT calculation is not like that.
ST is without SMT-2 but MT includes SMT-2.
So the result for MT, given average Zen arch SMT gain should be ~1.72.

I don't quite follow what you're saying there. The MT projection is versus 5950X's results that already have SMT counted in, so you can't count it in twice...

To be clear, the above projection is my estimate based on computerbase results you can find here: https://www.computerbase.de/thema/prozessor/rangliste/

 

[Anhiel]

I don't quite follow what you're saying there. The MT projection is versus 5950X's results that already have SMT counted in, so you can't count it in twice...

To be clear, the above projection is my estimate based on computerbase results you can find here: https://www.computerbase.de/thema/prozessor/rangliste/

Oops. Right.
The simple calculation got me talking unnecessary. Sorry

 

[Thunder 57]

There is no hyperthreading on i5 processors back in the 2500k and 3570k era. Only AMD offers hyperthreading aka. SMT on all their CPU's.
Here is anandtech's own article on the Spectre issue on coffee lake.

Analyzing Core i9-9900K Performance with Spectre and Meltdown Hardware Mitigations

www.anandtech.com

So if they fixed Spectre on the 9000 series, why did only the 9900k offer HT? As for the old i5's not having HT while i7 did, that was my point. It was all about segmentation, not Spectre. Sounds like you are the one that needs a little help.

AMD has also released at least a few models without SMT, see the 2200G or 1200. Pretty sure a few in mobile too, but thankfully they seemed to drop that practice.

 

[yuri69]

Yea but the point is prior to launch I scoffed at that clock speed as not possible and a console TDP. I was wrong then and I won't make the mistake to assume that AMD are cranking clock speeds at the cost of excesive power consumption.

No doubt RDNA2 has very nice perf/W but the consoles keep getting more power hungry.

PS3 (90nm) - ~190W thanks to Cell magic
PS3 Slim (45nm) - ~90W
PS4 - ~140W
PS5 - ~200W

This trend is awful since 20 years ago we got <40W Radeon R9700 paired with 70W Athlon XP. Now you run a 300W GPU with 150+W CPU. We are heading towards purposedly air-conditioned "computer rooms" again. It's the 70s again.

 

[nicalandia]

So if they fixed Spectre on the 9000 series, why did only the 9900k offer HT? As for the old i5's not having HT while i7 did, that was my point. It was all about segmentation, not Spectre. Sounds like you are the one that needs a little help.

AMD has also released at least a few models without SMT, see the 2200G or 1200. Pretty sure a few in mobile too, but thankfully they seemed to drop that practice.

Would you stop feeding the Troll and derailing this thread about Intel HT Security issues?