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!

 

[nicalandia]

Not exactly unexpected score. A bit on a low actually for DDR5 6000 +

In MT interesting is that 32T backed by X3D boosted + 6000 mem cannot beat 16C without HT backed by handtuned 6200 memory. I guess the days of CB23 like scaling in GeekBench are over for good now, lots of fuming inbound.

Geekbench does not benefit from 3D Cache, Gaming will be a different story.

 

[JoeRambo]

Geekbench does not benefit from 3D Cache,

And that "opinion" is backed by what? Comparison of Zen3 vs Zen3D at same clock speeds? Wasn't the claim that 3D cache makes memory tuning irrelevant, and here we have it, down to "gaming story" ?

 

[nicalandia]

And that "opinion" is backed by what? Comparison of Zen3 vs Zen3D at same clock speeds?

Exactly. Do you see any ST/MT boost on Milan-X parts compared to Milan?

 

[JoeRambo]

Exactly. Do you see any ST/MT boost on Milan-X parts compared to Milan?

Pulling bs out of thin air:

 

[igor_kavinski]

Gaming will be a different story.

Not really expecting the 7950X3D to be the absolute gaming champ (I'll be happy if proven wrong). But the 7800X3D should have a great price/performance ratio compared to Intel's top offerings.

 

[nicalandia]

Pulling bs out of thin air:

Hey look Everyone, this dude believes Geekbench benefits from 3D V Cache and since the few entries of 7950X3D have lower scores than the 7950X then it must be a fail..! What at Troll..

 

[Hail The Brain Slug]

Not exactly unexpected score. A bit on a low actually for DDR5 6000 + 3D Cache effort.

Score comparison with my 5Ghz locked no/HT 7950x:


In MT interesting is that 32T backed by X3D boosted + 6000 mem cannot beat 16C without HT backed by handtuned 6200 memory. I guess the days of CB23 like scaling in GeekBench are over for good now, lots of fuming inbound.

I am 100% convinced the MT score is low because of the lower power limit. The ST score being the same as your 5GHz locked 7950X is likely because the ST workload was scheduled on the V$ CCD, probably incorrectly.

 

[nicalandia]

Not really expecting the 7950X3D to be the absolute gaming champ (I'll be happy if proven wrong). But the 7800X3D should have a great price/performance ratio compared to Intel's top offerings.

When I said a different story, I meant that the 7950X3D will beat the 7950X In Gaming. Yeah the 7950X is not even the best Zen4 gaming CPU, a 7700X is due to being a single CCD So the 7800X3D would appear to be the next gaming King..

 

[Carfax83]

That may have been true, but I don't think it is anymore. Lately it seems the Intel crowd has been making more noise.

It ebbs and flows with the dominance of the preferred IHV at the time. Intel dominance on this forum was strong when AMD was pumping out garbage CPUs, at least comparatively speaking to Intel's offerings. And when Intel hit a brick wall at 10nm and stopped progressing for a long time and AMD's Ryzen products skyrocketed in terms of performance starting with Zen 2, AMD became more dominant.

Now that Intel is back on the scene with ADL and RPL, it's starting to become more balanced. But I would still say there is a lean towards AMD overall on Anandtech, likely because AMD is still considered an underdog.

It's more prevalent (and virulent) in the GPU forum than it is in the CPU forum. I remember a poster was downvoted for asking about where to purchase an RTX 4080.......in the RTX 4080 thread. And he was downvoted by the usual AMD cartel, who's purpose seems to be harassing anyone that espouses anything remotely seen as anti AMD and pro Intel or pro Nvidia.





Calling the AMD fans at Anandtech the "AMD Cartel" is inflammatory and unnecessary, much less posting this in a Zen thread.
Probably best you stay out of posting in AMD threads.


esquared
Anandtech Forum Director

 

[JoeRambo]

I am 100% convinced the MT score is low because of the lower power limit. The ST score being the same as your 5GHz locked 7950X is likely because the ST workload was scheduled on the V$ CCD, probably incorrectly.

I doubt it is correct as well, something might have been wrong with the system, easy to take a wrong step on AMD beta AGESA, without other scores to validate against hard to say exactly what. Could be that capframe guy style stuff with M/ULC ratio.

Hey look Everyone, this dude believes Geekbench benefits from 3D V Cache and since the few entries of 7950X3D have lower scores than the 7950X then it must be a fail..! What at Troll..

The joys of ADF brightest: claims GB6 does not benefit from 3D V Cache, is presented benchmarks where 5800 3D chip beats higher clocked 5800X by almost 10% in MT score, then proceeds with siege mentality deducing 7950x3D "fail" ( at least get it right if you are trolling, i was talking about HT vs non HT).
Being called a trolls at this point is a complement coming from true vendor fanboy.

 

[Hail The Brain Slug]

Also just going to drop these here:

https://twitter.com/x/status/1627974532282875904

https://twitter.com/x/status/1627974762281713666

 

[nicalandia]

claims GB6 does not benefit from 3D V Cache, is presented benchmarks where 5800 3D chip beats higher clocked 5800X by almost 10% in MT score

How do you explain this?

AMD Ryzen 9 5900 Benchmarks - Geekbench

browser.geekbench.com

According GB6 The 5900(non-X) has a 18% Lead over the 5900X(In MT).

You can't explain that. GeekBench6 is such a bad Benchmark for MT that it should not be used. Let me find GB5 Data so we can compare.

 

[Exist50]

Also just going to drop these here:

https://twitter.com/x/status/1627974532282875904

https://twitter.com/x/status/1627974762281713666

Ok, now that's more like it. Hope that's not just Asus exclusive, but it would be a killer feature if it is.

 

[Hail The Brain Slug]

Ok, now that's more like it. Hope that's not just Asus exclusive, but it would be a killer feature if it is.

We'll know when other vendors 1.0.0.5 bioses come out. I've been waiting for one for my gigabyte board. I'll be flashing it as soon as its available and checking.

 

[JoeRambo]

How do you explain this?

Statistics? It's OEM CPU, in hands of few people out of whom even fewer run GB6, ends up with too few scores for 5900 versus old horse 5900X ? 5800x and 5800X3D surely have plenty of scores and if we remove the more rare CPUs, the pecking order seems correct.

 

[BorisTheBlade82]

Also just going to drop these here:

https://twitter.com/x/status/1627974532282875904

https://twitter.com/x/status/1627974762281713666

Much better and more sophisticated approach than the purported Block-/Allowedlists.

 

[Hail The Brain Slug]

Much better and more sophisticated approach than the purported Block-/Allowedlists.

If this is not an ASUS exclusive feature, my guess is a combination of both. However, I am OK with a driver-managed application list for known applications that benefit from the V$ die, with the addition of some kind of cache<->memory workload analysis for apps not covered by the driver list.

 

[moinmoin]

I guess the days of CB23 like scaling in GeekBench are over for good now

Geekbench 6 doesn't scale in the traditional MT sense anymore and is useless for that purpose. Use an older version instead. (We got a whole thread for that topic.) Honestly I'd prefer GB6 only to be used for ST if at all on this forum.

 

[Heartbreaker]

Geekbench 6 doesn't scale in the traditional MT sense anymore and is useless for that purpose. Use an older version instead. (We got a whole thread for that topic.) Honestly I'd prefer GB6 only to be used for ST if at all on this forum.

It's more representative of a cross section of applications though.

Across a wide array of applications, that is what you get, because not every application is embarrassingly parallel.

You can see the individual tests in GB, if you want to see those that have higher level of parallelism.

 

[moinmoin]

It's more representative of a cross section of applications though.

Exactly, GB6's MT benchmark is representative of the applications tested and not representative of the tested chip's MT performance.

 

[lightmanek]

I run multiple badly threaded apps at the same time, guess what, I do get sensible core utilisation out of my 16 core CPU.
GB6 approach is valid only if you run single app at a time, in other words, it suits mobile phone style of work more than multicore desktop, but what I know ...

 

[Hail The Brain Slug]

I run multiple badly threaded apps at the same time, guess what, I do get sensible core utilisation out of my 16 core CPU.
GB6 approach is valid only if you run single app at a time, in other words, it suits mobile phone style of work more than multicore desktop, but what I know ...

Wait, you're saying you use more than one application at a time????

 

[Thunder 57]

It ebbs and flows with the dominance of the preferred IHV at the time. Intel dominance on this forum was strong when AMD was pumping out garbage CPUs, at least comparatively speaking to Intel's offerings. And when Intel hit a brick wall at 10nm and stopped progressing for a long time and AMD's Ryzen products skyrocketed in terms of performance starting with Zen 2, AMD became more dominant.

Now that Intel is back on the scene with ADL and RPL, it's starting to become more balanced. But I would still say there is a lean towards AMD overall on Anandtech, likely because AMD is still considered an underdog.

It's more prevalent (and virulent) in the GPU forum than it is in the CPU forum. I remember a poster was downvoted for asking about where to purchase an RTX 4080.......in the RTX 4080 thread. And he was downvoted by the usual AMD cartel, who's purpose seems to be harassing anyone that espouses anything remotely seen as anti AMD and pro Intel or pro Nvidia.

I'm not sure if you're trying to be objective but if so saying "AMD fans" all the time and now "AMD cartel" isn't helping. NVIDIA lost a lot of goodwill with the 4000 series whether it be due to pricing or underspeccing. I mean they literally "unlaunched" a card. Even they realized they went to far. And if the rumors about the 4060 are true, they aren't done pissing people off yet.

That doesn't mean anyone should be downvoting someone for asking about the 4080, but I would need to see it for context.

 

[BorisTheBlade82]

If this is not an ASUS exclusive feature, my guess is a combination of both. However, I am OK with a driver-managed application list for known applications that benefit from the V$ die, with the addition of some kind of cache<->memory workload analysis for apps not covered by the driver list.

I can only hardly imagine this to be an ASUS feature - I'd eat a broom stick if it was.

 

[Heartbreaker]

Exactly, GB6's MT benchmark is representative of the applications tested and not representative of the tested chip's MT performance.

If your goal is only to see the ideal maximum performance, you have Cinebench for that. But that is further from real world expectation than GB.