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!

 

[eek2121]

Igor’s Lab has an article on thermal and power management of Raphael. 2 things of interest that I did not see before:

1. Internal documents call the 16-core part an APU.
2. Internal documents show 120W and 170W TDP SKUs.

 

[Ajay]

Igor’s Lab has an article on thermal and power management of Raphael. 2 things of interest that I did not see before:

1. Internal documents call the 16-core part an APU.
2. Internal documents show 120W and 170W TDP SKUs.

There are no more secrets here and the changes compared to Zen 3 are more of a minor evolution, but the actual basic principle remains the same. So you can also infer Zen 3 from a lot of things. By the way, it’s interesting that AMD now also refers to the 16-core internally as an APU, which hides a graphics unit in the I/O die. But now we come to the power management on the next page, because the waste heat and the temperatures as a consequence have to come from somewhere.

So, yes, we've already seen these reported. Apparently, enthusiast grade CPUs are headed for high-end Consumer desktops; seems a bit odd, but okay.

 

[A///]

Wasn't it last week we were told the so-called 170-175 watt TDP part was a fabrication? Interesting to see it back on the menu but I'm also curious what's causing that bump in power. Is it the iGPU? The bump to 120 watts from a baseline of 105 watts for the 5800X and 5900X and likely also the 5600X would point towards the entire range getting iGPUs and people who tossed the idea away even as recently should take note of this revelation. Though the 5950X and 5900X share the same TDP, so I'm a little confused about the jump for the 16c part and not the 12c, unless both are getting the 170 watt treatment?

 

[Hougy]

Gracemont performs better than Zen 2 and worse than Zen 3. Throw 16 of them in a chip along with the Raptor Cove, and assuming Intel isn't power limited, the chip will be 40-50% faster than a 5950X. That means a 6950X will need to be 40%-50% faster to tie with the Intel chip. That is why AMD needs a 24 core SKU. Note that Raptor Lake is currently rumored to launch BEFORE Zen 4. If Raptor Cove is power limited, then AMD doesn't need to make up such a wide gap.



125W PL1, 241W PL2. That is from the Gigabyte leak. Raptor lake apparently has a lower PL2 (I would have to dig up the Igor's lab article).

The flagship Raptor lake could perform better than the flagship Zen 4 chip in multithreaded benchmarks if it has more cores but these benchmarks are useless to most consumers. What matters to games and the apps most people use is single threaded performance.

You are way overestimating the performance of Gracemont because you're only considering IPC and not the fact that it clocks lower and doesn't have SMT.

Zen 4 could have better single threaded performance, much lower power consumption and be cheaper, making it a huge win.

 

[eek2121]

So, yes, we've already seen these reported. Apparently, enthusiast grade CPUs are headed for high-end Consumer desktops; seems a bit odd, but okay.

Yes, many of us have known it for months, but some were in denial.

Wasn't it last week we were told the so-called 170-175 watt TDP part was a fabrication? Interesting to see it back on the menu but I'm also curious what's causing that bump in power. Is it the iGPU? The bump to 120 watts from a baseline of 105 watts for the 5800X and 5900X and likely also the 5600X would point towards the entire range getting iGPUs and people who tossed the idea away even as recently should take note of this revelation. Though the 5950X and 5900X share the same TDP, so I'm a little confused about the jump for the 16c part and not the 12c, unless both are getting the 170 watt treatment?

There appear to still be 105W parts. My initial thought was the 170W part is a 24c part and the 120w part is the 16c part, with the rest being 105W parts, but the APU could also be the differentiator.

 

[Mopetar]

So, yes, we've already seen these reported. Apparently, enthusiast grade CPUs are headed for high-end Consumer desktops; seems a bit odd, but okay.

I don't think it's odd at all. First not everyone who can benefit from a 16 core CPU needs a HEDT platform. Second, not everyone needs a particularly powerful GPU either and with the market the way it is right now being able to get an APU with 16 cores could save a lot of money.

 

[jpiniero]

Wasn't it last week we were told the so-called 170-175 watt TDP part was a fabrication? Interesting to see it back on the menu but I'm also curious what's causing that bump in power. Is it the iGPU? The bump to 120 watts from a baseline of 105 watts for the 5800X and 5900X and likely also the 5600X would point towards the entire range getting iGPUs and people who tossed the idea away even as recently should take note of this revelation. Though the 5950X and 5900X share the same TDP, so I'm a little confused about the jump for the 16c part and not the 12c, unless both are getting the 170 watt treatment?

Most likely a post-Raphael product that has more than 16 cores.

 

[Ajay]

I don't think it's odd at all. First not everyone who can benefit from a 16 core CPU needs a HEDT platform. Second, not everyone needs a particularly powerful GPU either and with the market the way it is right now being able to get an APU with 16 cores could save a lot of money.

Good point. Like scientific computing but with fairly simple output (graphs, less complex 3D models, etc.). Actually, video editing, if the APU has some acceleration for popular codecs.

 

[eek2121]

Good point. Like scientific computing but with fairly simple output (graphs, less complex 3D models, etc.). Actually, video editing, if the APU has some acceleration for popular codecs.

There is also the fact that the higher core count parts are typically binned better.

I can’t wait for Zen 4 TBH so I can spend more money. AMD is robbing me of every penny I have at this point. I went from a 1950X to a 3900X to a 5950X. That doesn’t even count the other machines in my household. Hopefully Zen3D isn’t an outrageously good product as well…

 

[Ajay]

There is also the fact that the higher core count parts are typically binned better.

Well, one of the two chiplets (at least with Zen2).

 

[Mopetar]

Well, one of the two chiplets (at least with Zen2).

Both chiplets hit the minimum bin requirements, but they aren't paired in any way beyond that.

Depending on the actual distribution in terms of silicon quality and where they set the bins it's pretty unlikely to get two evenly matched chiplets.

The vast majority of CPUs sold will have a disparity between those chiplets of at least a certain threshold.

 

[A///]

Most likely a post-Raphael product that has more than 16 cores.

Might have something mixed up here, but I distinctly remember that Zen 4 has a 170-175 watt TDP product, not AM5. If it's post Raphael, then it's not Zen 4, but an AM5 socket product.

 

[prtskg]

Might have something mixed up here, but I distinctly remember that Zen 4 has a 170-175 watt TDP product, not AM5. If it's post Raphael, then it's not Zen 4, but an AM5 socket product.

I was under the impression that it was max sustainable power consumption of AM5 socket, like 140W is for AM4.

 

[eek2121]

I was under the impression that it was max sustainable power consumption of AM5 socket, like 140W is for AM4.

This implies otherwise. Specs were given, though that does not mean AMD will release a part with those specs. They are basically saying such a SKU is a possibility.

 

[DrMrLordX]

I was under the impression that it was max sustainable power consumption of AM5 socket, like 140W is for AM4.

142W isn't actually the max sustainable power for AM4. It can handle over 300W. Nothing AM4 pulls that kind of power by default, but it's possible.

 

[Abwx]

142W isn't actually the max sustainable power for AM4. It can handle over 300W. Nothing AM4 pulls that kind of power by default, but it's possible.

OEMs can even do a 500W sustainable, the 142W sustained capability is the requirement from AMD for AM4 validation of the boards, they can increase this number for next boards the same way they did increase the power requirements from AM3 to AM3+.

 

[eek2121]

142W isn't actually the max sustainable power for AM4. It can handle over 300W. Nothing AM4 pulls that kind of power by default, but it's possible.

Yeah I thought the quote was kind of odd. On my 3900X I had PBO turned on and set to 175W. I never had heat issues (except my office when CPU and GPU was active lol)

 

[prtskg]

142W isn't actually the max sustainable power for AM4. It can handle over 300W. Nothing AM4 pulls that kind of power by default, but it's possible.

I should have said default power consumption limit of AM4 processors.

 

[eek2121]

Zen 4 leaks, this time about cooling: https://videocardz.com/newz/first-c...n4-genoa-and-am5-zen4-raphael-sockets-spotted

What I find curious is that there isn’t a separate section for Threadripper, and some AM5 examples are above 200W.

 

[Joe NYC]

Any theories on this one?

https://twitter.com/x/status/1438564239933853698

 

[leoneazzurro]

Isn't MI300 the next CDNA compute GPU?

 

[Joe NYC]

Isn't MI300 the next CDNA compute GPU?

Yes, but in a server socket?

MI100 and (presumably) MI200 are PCIe cards

 

[leoneazzurro]

Yes, but in a server socket?

MI100 and (presumably) MI200 are PCIe cards

I am following some leakers and the idea seems that MI300 is something like an APU. Probably something like Epyc+CDNA3 on the same socket, probably with enhanced interconnects thanks to stacking. I wonder which advantages this kind of configuration will have.

 

[xilli_fiberbit]

Yes, but in a server socket?

MI100 and (presumably) MI200 are PCIe cards

MI200 from what we know is in OAM card, i don't know if there will be too PCIe version

 

[LightningZ71]

I am following some leakers and the idea seems that MI300 is something like an APU. Probably something like Epyc+CDNA3 on the same socket, probably with enhanced interconnects thanks to stacking. I wonder which advantages this kind of configuration will have.

Hmm, carrying this idea forward, this could be the core of a highly converged product, very similar to an APU, but, in an EPYC-like MCM with an IO die, a pair of Zen CCDs for flow control, a pair of CDNA GPUs chips, and four HBM2E or maybe even 3 stacks to feed those GPUs. I don't think that the 8-12 channels of DDR5 ECC RDIMMS will have near enough bandwidth to directly feed the GPU die, but, if we imagine the HBM chips in the same role of the infinity cache, at 8-16GB per gpu chip, and considerably more through the IO die, maybe we have something interesting. Instead of remaining at just 2P systems with a ton of PCIe lanes to expansion cards, you have quad socket systems with sufficient IO for system connectivity.