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!

 

[Timmah!]

I think its given its going to be overall superior CPU to vanilla ones, simply based on the fact it trades 5 percent worse clocks (which even might be somewhat fixable by PBO stuff) for 2x bigger L3 cache.

 

[Hail The Brain Slug]

Can't wait to see folks buy a $699 CPU, only to turn off the CCD without the V-cache

That's why I bought a license for process lasso.

 

[nicalandia]

That's why I bought a license for process lasso.

Why do you have to pay for it?

 

[Hail The Brain Slug]

Why do you have to pay for it?

So it doesn't pop up an intrusive "buy me now!" prompt at every boot

 

[deasd]

sonuds like the new chipset driver was not tested. That may result in what, some non-cache-sensitive games still use V-cache cores?

https://twitter.com/x/status/1629116333991272449

 

[utahraptor]

sonuds like the new chipset driver was not tested. That may result in what, some non-cache-sensitive games still use V-cache cores?

https://twitter.com/x/status/1629116333991272449

That depends on what extent it is bios (see the new Asus bios reports: Beta BIOS Lets You Prioritize CCDs on AMD's 7000X3D CPUs) rather than driver dependent. I find it unlikely a reviewer guide would be created using settings that would hamstring their performance.

 

[Mopetar]

sonuds like the new chipset driver was not tested. That may result in what, some non-cache-sensitive games still use V-cache cores?

https://twitter.com/x/status/1629116333991272449

Ooo... Last hype train leaving the station. Last hype train! All aboard!

 

[Hail The Brain Slug]

That depends on what extent it is bios (see the new Asus bios reports: Beta BIOS Lets You Prioritize CCDs on AMD's 7000X3D CPUs) rather than driver dependent. I find it unlikely a reviewer guide would be created using settings that would hamstring their performance.

It's not unusual for reviewers to receive samples early on and perform preliminary benchmarking only to have updated drivers sent to them just days before launch and embargo lift, forcing them to rush to rebenchmark everything.

 

[utahraptor]

It's not unusual for reviewers to receive samples early on and perform preliminary benchmarking only to have updated drivers sent to them just days before launch and embargo lift, forcing them to rush to rebenchmark everything.

I agree with that statement, but I believe the numbers the tweet is referencing were provided by AMD for reviewers rather than by a reviewer.

 

[Timmah!]

Ooo... Last hype train leaving the station. Last hype train! All aboard!

Yeah, AMD classic. Its incredibly annoying. Even when they have good product and dont need this, various kinds of deranged people will keep hyping it up to the eleven for no reason.

 

[Joe NYC]

When the 5800x3d came out, it was compared against cache limited 11th gen Intel products and Alder Lake products with less cache than the current Raptor lake products. Current Raptor lake has 36MB of L3 and a lot of L2 coupled with a notably faster memory subsystem than previous processors. I wouldn't expect the large vCache to make a huge difference across the board any more, and only those few games that have moderate working sets that fall in the 50-100MB of size will strongly favor AMD. Anything larger will swing back to Intel.

Exactly. Often overlooked fact, that Raptor Lake high end SKUs owe a big part of their gaming performance to the increased L3. Plus, the clock speed.

 

[LightningZ71]

conversely, AMD owes a big part of their being able to be competitive with Intel in the past few years to having a much larger L3. Now that Raptor Lake has comparable cache to an AMD CCX, we can see that AMD's chips aren't faster anymore. When we look at their mobile chips, which are based on the same cores, they fare worse against Intel's comparable parts.

 

[biostud]

Looking forward to see the difference with optimized memory timings in games.

 

[utahraptor]

Looking forward to see the difference with optimized memory timings in games.

As am I. I have heard multiple people at multiple places saying 3d cache is not really memory dependent based on 5x series, but I don't recall any of them providing a benchmark to back that up. It really opened my eyes when hardware unboxed posted their roided memory video the other day. I was unaware that the xmp/xpo profiles simply are unable to contain enough data to save sub timings that are that impactful. I hope they get that fixed later in the DDR5 cycle or in DDR6. Its non ideal to leave it to the consumers to fill in the missing values. At an absolute minimum there should be an insert in the package with the missing values so they can be input by hand.

 

[biostud]

I'll be damned. The reviews are out February the 27th, my 20 year anniversary as an AT member. Is this a sign?

 

[JoeRambo]

Its non ideal to leave it to the consumers to fill in the missing values. At an absolute minimum there should be an insert in the package with the missing values so they can be input by hand.

Indeed! It is AMD's and Intel's job to provide facilities for proper memory training and then motherboard makers to make sure AUTO values are proper, not brain dead.
Right now incentitives are stacked the other way - memory and MB vendors have to avoid RMAs and they slacked secondaries and tertiaries as much as possible to make things work with EXPO/XMP.

I can't blame them, with the way things are on AMD right now, wrong step with timings -> MB that does not work without battery removal = average Joe sending it back. Can't blame vendors when platform owner is run by incompetent idiots.

 

[Joe NYC]

Looking forward to see the difference with optimized memory timings in games.

EXPO or XMP timings are more indicative of what overwhelming majority of users will experience.

 

[Mopetar]

conversely, AMD owes a big part of their being able to be competitive with Intel in the past few years to having a much larger L3. Now that Raptor Lake has comparable cache to an AMD CCX, we can see that AMD's chips aren't faster anymore. When we look at their mobile chips, which are based on the same cores, they fare worse against Intel's comparable parts.

I don't think the L3 matters that much for most applications. We already see that the extra v-cache does nothing for a lot of applications. Up until Zen 2 or 3 (I forget which) AMD had the same 2 MB of L3 per core that Intel's architectures had.

AMD was primarily successful in catching up with Intel in single core performance due to being able to iterate on their original Zen design rather successfully. But I think most of their success comes down to them being able to offer more cores at a lower price and only now has Intel been able to curb that advantage.

If Intel had t dropped the ball as hard as they did, I'm not sure if Zen 3 would have been the surprise that it was. I don't know if it would have mattered either as Zen 2 was a major success for AMD even though Intel still had a solid lead in single core performance and gaming at the time. But as I said, AMD was offering significantly more cores and a lot more MT performance. The Ryzen 5 CPUs offering 6C/12T for $200 or less was a hell of a deal for consumers at a time when Intel had to disable SMT and wanted $250+ for a 6C/6T CPU.

 

[Schmide]

I'll be damned. The reviews are out February the 27th, my 20 year anniversary as an AT member. Is this a sign?

Well 27th anniversary gift is sculptures (i did not know this offhand) and 3d cache could be interpenetrated as sculptured ?

edit: read date as year oops

 

[A///]

So it doesn't pop up an intrusive "buy me now!" prompt at every boot

Terrible reasoning. You're supporting the developers. That's a better reason.

 

[A///]

Ooo... Last hype train leaving the station. Last hype train! All aboard!

Hasan and BS. Name a better combo.

 

[A///]

conversely, AMD owes a big part of their being able to be competitive with Intel in the past few years to having a much larger L3. Now that Raptor Lake has comparable cache to an AMD CCX, we can see that AMD's chips aren't faster anymore. When we look at their mobile chips, which are based on the same cores, they fare worse against Intel's comparable parts.

My only issue right now is what the f is this snappy feedback people say about zen 4 on windows 10 and 11 over Intel. People say it but there's no real info about it. Maybe I'm getting too old to notice stuff like this.

 

[biostud]

EXPO or XMP timings are more indicative of what overwhelming majority of users will experience.

Yes, but we are not those, are we?

 

[A///]

Yes, but we are not those, are we?

XMP and EXPO seem to do well being enabled at higher speeds with minimal to no user meddling. V different from years ago on intel.

 

[alexruiz]

That depends on what extent it is bios (see the new Asus bios reports: Beta BIOS Lets You Prioritize CCDs on AMD's 7000X3D CPUs) rather than driver dependent. I find it unlikely a reviewer guide would be created using settings that would hamstring their performance.

You need both.
The UEFI updates allows for the proper CCD selection, the updated chipset divers will direct the OS to do it.