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!

 

[poke01]

Geekbench isn't that great as it doesn't stress the cpu in all aspects needed.

That's why I said for ST its great. AMD also use Geekbench for ST.

 

[Abwx]

Geekbench subscores have Zen 4 with 4/5% better ST IPC in FP/INT than ADL, that s a little more than an error of margin, and is aknowledged by the gaming perf of the 7600X.

 

[poke01]

Geekbench ST analysis between 12900KS and 7600X, AES excluded.

Text Compression	1635​	2060​	0.79
Image Compression	2049​	2046​	1.00
Navigation	1683​	1679​	1.00
HTML5	1833​	1766​	1.04
SQLite	1812​	1709​	1.06
PDF Rendering	1943​	1653​	1.18
Text Rendering	1960​	1820​	1.08
Clang	1794​	1928​	0.93
Camera	1822​	1638​	1.11
N-Body Physics	1986​	2193​	0.91
Rigid Body Physics	1966​	1960​	1.00
Gaussian Blur	1949​	2323​	0.84
Face Detection	2155​	2121​	1.02
Horizon Detection	1764​	1720​	1.03
Image Inpainting	3582​	3853​	0.93
HDR	3629​	3928​	0.92
Ray Tracing	2625​	2507​	1.05
Structure from Motion	1784​	1580​	1.13
Speech Recognition	1864​	1959​	0.95
Machine Learning	1605​	1787​	0.90
Average			0.99
12900KS ran at slightly higher peak clock (5461 vs 5394; 0.987x factor)
IPC excluding AES: 0.99 x 0.987 = 0.977 or ~ 2% lower IPC than Zen 4

This fits perfectly with AMD's 13% number, as Golden Cove has ~11% higher IPC than Zen 3 according to Ian's deep dive article and ComputerBase average ST chart (normalized for clockspeed)

but but "GeEkBeNCh is a horrible benchmark and its meant for analysis".

I always said that Geekbench is perfect for ST as a synthetic tool . Now, Userbenchmark now that's horse poop.

AMD, Qualcomm, Intel, Mediatek all use it for ST.

 

[moinmoin]

The core efficiency yes, though we can't take the power efficiency as a guess here as mobile Zen 4 uses a different process node (N4P?). 6000 matches up with M2 for performance, but at 15 vs 10 watts. It'll be interesting to see what the Zen 4 mobile and RDNA3 will bring.

Dragon Range should be the mobile counterpart to Raphael, launch first and still be on the same node N5. Phoenix Point should be on N4 together with Bergamo, but launch much later. AMD lists both for the beginning of next year, but with AMD's mobile chips good availability takes many months every year, so I expect AMD to obfuscate the time gap between Dragon Range and Phoenix Point through that.

 

[nicalandia]

Locuza already working on the details

 

[Carfax83]

That's all well and good. I don't really disagree with most of what you've just posted. I wasn't disputing that the 12900k setup is non optimal, I'm questioning if it actually matters. Do you have data that shows a significant performance difference between a 12900k w/ 6400 CL32 vs 6000 CL30 in gaming or any other meaningful benchmark? Is this a small detail that makes a practical, demonstrable difference or is it a nitpick? I'm not trying to pick a fight I'm genuinely interested.

Does it matter? Probably not. Good memory scaling benchmarks are difficult to find because most reviewers are either incompetent or lazy. I can't tell you how many times I've seen memory scaling benches done at GPU limited settings, which defeats the entire point.

At any rate, I already said why I had a problem with AMDs methodology when it comes to how they configured the Intel setup. Alder Lake is not restrained from benefitting from faster memory frequencies while Zen 4 is.

Therefore any review which tries to equalize memory frequency is automatically suspect in my opinion. I can't say I expected any different from AMD since they have the most to gain from doing a stunt like this, but what I am saying applies mostly to the third party reviewers.

 

[poke01]

When Zen 4 mobile launches it will be a sight to behold. AMD will easily beat Raptor in mobile and Meteor Lake is not focused on PPW like AMD Zen4 is.

AMD will usher in a new era for handheld gaming, AMD ultrabooks will be better than M1/M2 in efficency, pref/$ and a MUCH larger software programs and RDNA3 will also help PS, xbox, steam deck and also gaming on ultrabooks will be possible as the 7800U with 8GB LPDRR5/x is more powerful than a PS4.


AMD is no LONGER a budget brand with slow and power hungry products. In the FX era AMD was known to be a 2nd citizen to Intel with slow and power hungry cores. Nope, not anymore. This new era in this decade will belong to AMD as long they continue to deliver like they do now.

AMD is now premium because they now have the hardware stack to ask the price they are asking for.

 

[Vope45]

Do you have some kind of inside info to claim this or its just an assumption/guess on your part?
It would be nice to know for sure, so i wont wait several months for 7950x3D and then they wont even release it.

Silicon valley gossip can be made into oscar nominated hbo limited release

 

[poke01]

To all the Intel pudits in this thread. AMD is matching Intel ST using less power. Intel is no longer the ST king. Raptor won't change a thing cause its will use more power.

Intel sucks at efficieny and perf/w just look at mobile ADL. Horrible battery life.

 

[Carfax83]

On the topic of memory scaling, what are your thoughts on Zen4 DDR5 scaling in gaming?

View attachment 66792

From the leaks, they are saying the max IFC will be 3ghz, which means to run at 1:1 ratio you will need to pair it with DDR5 6000. You will be able to use faster memory to be sure in gear 2 mode (and increase bandwidth) but the scaling won't be there and performance will suffer because latency will increase. It's just like Zen 3.

In fact, that slide you posted shows memory latency at 63ns which is already more than what Alder Lake can achieve. According to Techpowerup, 12900K plus DDR5 6600 gets 59ns.

That to me will put it at a disadvantage to Raptor Lake which will be capable of hitting even higher memory frequencies than Alder Lake due to a more robust memory controller.

DDR5 7000 should be in the mid 50s.

 

[HurleyBird]

AMD ultrabooks will be better than M1/M2 in efficency

Don't count on it.

 

[Markfw]

Don't count on it.

I don't know if the 6800u had M1/M2 included in their reviews, but they are great.

 

[poke01]

I don't know if the 6800u had M1/M2 included in their reviews, but they are great.

Don't count on it.

HUB did include the 6800u in his M2 review and it had got 8hr battery life vs 12hr for M2.

I suspect since zen 4 mobile is going to be on TSMC 4nm and impovements from Zen4 will also help maybe better is too much but close or on par in efficiency is the better word. But hey with AMD I still can access my games and non-64 bit apps. So if AMD still loses by 30 min - 1hr in battery life I don't mind.

 

[PJVol]

Geekbench does not push CPUs to reach their max power consumption. And two articles mean nothing if they recycle the same leak.

Depends what you mean by "max". 5600X peaks at 100W PPT (76W stock) in MT part.
And gb5 result shows nothing about the power limit used.

 

[Saylick]

Greymon55 on Twitter is hinting at which models get V-cache... Looks like it's the 7800X, 7900X, and 7950X.

If true.... *deep inhale*

Spoiler

https://twitter.com/x/status/1564807986497081346

https://twitter.com/x/status/1564808672282558465

 

[Thunder 57]

That surprises me a bit. I figured vcahce would only be on the 7800X and 7900X.

 

[tamz_msc]

Geekbench subscores have Zen 4 with 4/5% better ST IPC in FP/INT than ADL, that s a little more than an error of margin, and is aknowledged by the gaming perf of the 7600X.

Geekbench has ADL ahead in int IPC, but behind in fp IPC against Zen 4. Compare against clean GB runs for example from our forum member AdamK47:

Gigabyte Technology Co., Ltd. Z690 AORUS MASTER - Geekbench

Benchmark results for a Gigabyte Technology Co., Ltd. Z690 AORUS MASTER with an Intel Core i9-12900KS processor.

browser.geekbench.com

vs

7600X

Micro-Star International Co., Ltd. MS-7D69 - Geekbench

Benchmark results for a Micro-Star International Co., Ltd. MS-7D69 with an AMD Ryzen 5 7600X processor.

browser.geekbench.com

 

[Det0x]

From the leaks, they are saying the max IFC will be 3ghz, which means to run at 1:1 ratio you will need to pair it with DDR5 6000. You will be able to use faster memory to be sure in gear 2 mode (and increase bandwidth) but the scaling won't be there and performance will suffer because latency will increase. It's just like Zen 3.

In fact, that slide you posted shows memory latency at 63ns which is already more than what Alder Lake can achieve. According to Techpowerup, 12900K plus DDR5 6600 gets 59ns.

That to me will put it at a disadvantage to Raptor Lake which will be capable of hitting even higher memory frequencies than Alder Lake due to a more robust memory controller.

DDR5 7000 should be in the mid 50s.

More e-cores -> more stops on the ringsbus = higher latency..

These ~50ns and below AL screenshots are all done with e-cores disabled because of this, ontop of the e-cores also limits your maximum uncore speed..

You wont see mid 50ns by just enabling XMP and hoping for the best with Raptor Lake anytime soon

 

[Thunder 57]

Depends what you mean by "max". 5600X peaks at 100W PPT (76W stock) in MT part.
And gb5 result shows nothing about the power limit used.

65W TDP, 88W PPT.

 

[tamz_msc]

More e-cores -> more stops on the ringsbus = higher latency..

Which is vastly improved in Raptor Lake.

https://twitter.com/x/status/1561908308277366785

 

[Markfw]

Geekbench has ADL ahead in int IPC, but behind in fp IPC against Zen 4. Compare against clean GB runs for example from our forum member AdamK47:

Gigabyte Technology Co., Ltd. Z690 AORUS MASTER - Geekbench

Benchmark results for a Gigabyte Technology Co., Ltd. Z690 AORUS MASTER with an Intel Core i9-12900KS processor.

browser.geekbench.com

vs

7600X

Micro-Star International Co., Ltd. MS-7D69 - Geekbench

Benchmark results for a Micro-Star International Co., Ltd. MS-7D69 with an AMD Ryzen 5 7600X processor.

browser.geekbench.com

No offense to adam, but comparing anything to his highly tweaked and cherry picked CPU is crap.

 

[tamz_msc]

No offense to adam, but comparing anything to his highly tweaked and cherry picked CPU is crap.

Uh, both adam's 12900KS and the 7950X (and presumably the 7600X) were using 6000 MT/s DDR5, if I'm not mistaken. So if anything such a comparison is more apples to apples than pulling random scores from the Geekbench database.

 

[Markfw]

Uh, both adam's 12900KS and the 7950X (and presumably the 7600X) were using 6000 MT/s DDR5, if I'm not mistaken. So if anything such a comparison is more apples to apples than pulling random scores from the Geekbench database.

That is one factor in many. cpu speed, memory tweaking are just a couple.

 

[biostud]

That surprises me a bit. I figured vcahce would only be on the 7800X and 7900X.

I would have thought only 7800X and 7950X, but let's see.

 

[Det0x]

Which is vastly improved in Raptor Lake.

https://twitter.com/x/status/1561908308277366785

Yes its better, by how much remains to be seen.. Cant just break the laws of physics even if its a new better design.
There is still a big differanse between 8P+0E and 8P+16E latency-wise @ something like ~5200mhz vs ~ 4600mhz uncore.

Interesting times ahead, looking forward to more leaks on this

But do you actually disagree with my comment below ?

You wont see mid 50ns by just enabling XMP and hoping for the best with Raptor Lake anytime soon