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!

 

[Racan]

Nice little chip

https://www.amd.com/en/products/cpu/amd-ryzen-5-7500f

 

[Hitman928]

Nice little chip

https://www.amd.com/en/products/cpu/amd-ryzen-5-7500f

Interesting, the budget AMD chip has a more significant lead in gaming over Intel's budget chip than I had expected. It's actually cheaper as well which is nice to see.

 

[eek2121]

That looks like a solid budget chip IMO.

 

[Abwx]

Some of the slides from AMD at HChips.

Hot Chips 2023: AMD verrät wenig mehr über „Siena“ und verwirrt

Nach Genoa, Bergamo und Genoa-X fehlt noch Siena in AMDs Familie der Zen-4-CPUs für Server. Zur Hot Chips gibt es (verwirrende) Details.

www.computerbase.de

 

[Joe NYC]

I don't know if this is a Starfield effect, but 7800x3d is sold out or with low stock numbers in MicroCenters in my area. They always had sufficient numbers in stock in prior months.

 

[Markfw]

I don't know if this is a Starfield effect, but 7800x3d is sold out or with low stock numbers in MicroCenters in my area. They always had sufficient numbers in stock in prior months.

newegg has it for $399. Wa MC a lot cheaper ?

 

[Joe NYC]

newegg has it for $399. Wa MC a lot cheaper ?

$369

Only $20 discount on mobo.

No crazy giveaways, like 64 GB of free DDR5 that you get with 7900X or 7950X

 

[yuri69]

Genoa is finally available at AWS. Its pricing is... high. TCO, availability, and power advantage don't seem to apply. AMD used to be the cheaper option but now, it's 15% more expensive than Intel. Compared to the ARM solution, AMD is 1.42 of the ARM solution!

4 vCPUs EC2 yearly Ohio:

vendor	CPU name	type	cost $	% to Cascade Lake
Intel	Cascade Lake	m5	102,32​	100,00%​
AMD	Rome	m5a	91,66​	89,58%​
AMD	Milan	m6a	92,09​	90,00%​
AWS	Graviton 2	m6g	82,07​	80,21%​
Intel	ICL	m6i	102,32​	100,00%​
AMD	Genoa	m7a	123,55​	120,75%​
Intel	SPR	m7i	107,43​	104,99%​
AWS	Graviton 3	m7g	86,97​	85,00%​

 

[SteinFG]

I think AWS prices them based on demand and not what you're mentioning. AWS will gladly take bigger profit if they can.

 

[Markfw]

Genoa is finally available at AWS. Its pricing is... high. TCO, availability, and power advantage don't seem to apply. AMD used to be the cheaper option but now, it's 15% more expensive than Intel. Compared to the ARM solution, AMD is 1.42 of the ARM solution!

4 vCPUs EC2 yearly Ohio:

vendor	CPU name	type	cost $	% to Cascade Lake
Intel	Cascade Lake	m5	102,32​	100,00%​
AMD	Rome	m5a	91,66​	89,58%​
AMD	Milan	m6a	92,09​	90,00%​
AWS	Graviton 2	m6g	82,07​	80,21%​
Intel	ICL	m6i	102,32​	100,00%​
AMD	Genoa	m7a	123,55​	120,75%​
Intel	SPR	m7i	107,43​	104,99%​
AWS	Graviton 3	m7g	86,97​	85,00%​

You also are not considering performance. I know personally that Genoa is quite a bit faster than Rome or Milan, and pretty sure its a fair amount faster than SPR or cascade lake based on the benchmarks I have seen at Phoronix and the like. Performance DOES count. And while power is higher than Rome, performance/watt Genoa still is better.

You are only looking at $$ and thats not right.

 

[NostaSeronx]

So far high powered PCs are of no interest for 80-90% of the users, most would be content with a 8C Zen 4, like the one below using a U version at 25W :

Minisforum presents Mercury EM780 palm-size mini PC with AMD Ryzen Phoenix 7840U and dual-USB4 ports

The EM780 is among the smallest mini PCs made by Minisforum, but it can support up to a Ryzen 7 7840U with eight Zen 4 cores and an RDNA 3 Radeon 780M. Despite the tiny case, Minisforum even managed to add active coolers for the processor and the M.2 storage.

www.notebookcheck.net

That isn't the EM780 anymore. It is this, with an aluminum case and all:


EM680 => Plastic Shell
EM780 => Aluminum Shell

I am shocked that reddit/x/or whatever hasn't leaked this for the 7840u variant of EM. We had to wait for the minisforum showcase event thing and someone to take pictures. As well as me posting it here.

 

[TESKATLIPOKA]

That isn't the EM780 anymore. It is this, with an aluminum case and all:
View attachment 85141

EM680 => Plastic Shell
EM780 => Aluminum Shell

I am shocked that reddit/x/or whatever hasn't leaked this for the 7840u variant of EM. We had to wait for the minisforum showcase event thing and someone to take pictures. As well as me posting it here.

I think I will cry. Finally, not a half page long post from you.
BTW thanks for posting this info.

 

[DrMrLordX]

You also are not considering performance. I know personally that Genoa is quite a bit faster than Rome or Milan, and pretty sure its a fair amount faster than SPR or cascade lake based on the benchmarks I have seen at Phoronix and the like. Performance DOES count. And while power is higher than Rome, performance/watt Genoa still is better.

You are only looking at $$ and thats not right.

Well to be fair, he does have the relative performance of the AWS instance versus the default (100%) which in this case is Cascade Lake. Actual performance difference will vary based on workload.

It's bizarre that Milan performs worse than Cascade Lake in this comparison, which is not reflective of any real-world bare metal benchmark. There's got to be some configuration issue with the specific instances shown.

 

[Markfw]

Well to be fair, he does have the relative performance of the AWS instance versus the default (100%) which in this case is Cascade Lake. Actual performance difference will vary based on workload.

It's bizarre that Milan performs worse than Cascade Lake in this comparison, which is not reflective of any real-world bare metal benchmark. There's got to be some configuration issue with the specific instances shown.

I believe that was the price vs cacade lake, not performance.. Hence my comment.

 

[DrMrLordX]

I believe that was the price vs cacade lake, not performance.. Hence my comment.

Ah okay. I see your point.

 

[yuri69]

I think AWS prices them based on demand and not what you're mentioning. AWS will gladly take bigger profit if they can.

Well, if it was based on demand then the Cascade Lake is still requested more than Milan.

You also are not considering performance. I know personally that Genoa is quite a bit faster than Rome or Milan, and pretty sure its a fair amount faster than SPR or cascade lake based on the benchmarks I have seen at Phoronix and the like. Performance DOES count. And while power is higher than Rome, performance/watt Genoa still is better.

You are only looking at $$ and thats not right.

I'm not looking purely on $$. I'm loking at perf/$$.

Both Rome and Milan instances are 10% cheaper than Cascade Lake or Ice Lake. This fits since the performance per-core is similar but total perf/W is notably better in case of EPYC.

For some reason this doesn't apply to Genoa. Genoa is even 15% more expensive than Sapphire Rapids - yet, it still offers similar per-core perf and better total perf/W.

 

[Hitman928]

Well, if it was based on demand then the Cascade Lake is still requested more than Milan.

I'm not looking purely on $$. I'm loking at perf/$$.

Both Rome and Milan instances are 10% cheaper than Cascade Lake or Ice Lake. This fits since the performance per-core is similar but total perf/W is notably better in case of EPYC.

For some reason this doesn't apply to Genoa. Genoa is even 15% more expensive than Sapphire Rapids - yet, it still offers similar per-core perf and better total perf/W.

SPR has similar per core performance as Milan. Genoa is faster per core.

 

[Markfw]

For some reason this doesn't apply to Genoa. Genoa is even 15% more expensive than Sapphire Rapids - yet, it still offers similar per-core perf and better total perf/W.

From what I have seen in benchmarks and reviews, Genoa is quote a bit faster than SPR and still takes quite a bit less power.

Edit: 41% faster. And I will look up the power in that linked article if I can, but I am sure it uses less power.

Page 506 - Discussion - Speculation: Zen 4 (EPYC 4 "Genoa", Ryzen 7000, etc.)

Page 506 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

Different article but 568 vs 464 watts (22% less) power avg. and 31% faster. Actually the 9654 used 3 watts less than 9554 !

AMD EPYC 9754 Benchmarks For The 128-Core Bergamo Review - Phoronix

www.phoronix.com

In both I am comparing dual 8490H vs 9554 Genoa. If you use the dual 9654, the performance diff is 54% and the power is virtually the same !

 

[yuri69]

From what I have seen in benchmarks and reviews, Genoa is quote a bit faster than SPR and still takes quite a bit less power.

Edit: 41% faster. And I will look up the power in that linked article if I can, but I am sure it uses less power.

Is that performance figure on per-core basis? As the pricing is basically per-core.

It is 96 vs 60 cores equals to 160%, right?

 

[Markfw]

Is that performance figure on per-core basis? As the pricing is basically per-core.

It is 96 vs 60 cores equals to 160%, right?

Per core ?? well, then it would most likely be 60 vs 64 if you look at the 9554. You would have to see what chip they are basing it on. The 9554 is better per core for performance.

But the point stands. What they are charging does not appear to be based on performance, but many factors. The percentages you point out don't make much sense, as if Genoa is 41% more performance (a little less per core) then 23% more charges are a bargain.

 

[Markfw]

So, I know you can't really do this, but if you take the dual 9554 and divide by 128 vs the 8490H and divide by 120 you get this, and Genoa still wins:

10.52164063​

8.560083333​

 

[A///]

dunno if it makes a difference to you @Markfw but Samsung has announced they'll be manufacturing 32Gb DDR5 dimms by end of this year. I assume epyc and ryzen platforms will get another update for more RAM capacity.

 

[Markfw]

So, I know you can't really do this, but if you take the dual 9554 and divide by 128 vs the 8490H and divide by 120 you get this, and Genoa still wins:

10.52164063​

8.560083333​

Oh, and power usage... The 9554 is 3.625 watts, and the 8490H is 4.73333 per core, so the Genoa is like ~20% faster and uses 23% less power. (not exactly sure of those numbers, but you have all the details)

And remember, Phoronix provided all the benchmarks, I just did the math on their numbers. But faster is worth more money/hour (better for the user), and less power helps the vendor (AWS) make more money. win/win !

 

[eek2121]

dunno if it makes a difference to you @Markfw but Samsung has announced they'll be manufacturing 32Gb DDR5 dimms by end of this year. I assume epyc and ryzen platforms will get another update for more RAM capacity.

Likely to be enterprise/server only for the immediate future.

 

[A///]

Likely to be enterprise/server only for the immediate future.

He runs servers which is why I brought it up. I think we'll see 128 GB kits on consumer setups being available near the end of DDR5's life which is in 3-4 years going by historical timelines and following development news of ddr6. I think 96 gb on two sticks is doable now but I suspect the market for that is slim or I don't know power user demographics as well as I think I do.