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!

 

[Carfax83]

None of what you said explains why that benchmark had the 5600 faster than the 5900X, 5800X, and 5600X in Doom Eternal.

The graph suggests that they have a bit of a GPU bottleneck. The 12700K is also slightly faster than the 12900K.

Does anyone know what GPU they used for the test?

 

[Hans Gruber]

My concern about Zen 4 moving forward. The sweet spot for DDR5 is 6000mhz according to AMD. The problem with that sweet spot. There are 7200mhz kits already out on the market. Unlike with DDR4 where the speed of the sticks was a slower move forward. DDR5 is coming with high memory frequency out of the box. The latency is what is lagging behind with DDR5 kits. I said earlier that DDR5 6000mhz with 28-30 Cas latency sticks would be what you want. Obviously 28-28-28 sticks @ 6000mhz sticks will be a little bit further out.

Will the memory dividers be a wall for latency with DDR5 in AM5?

 

[CakeMonster]

Well, we don't know yet how much of an impact latencies will have, do we? I mean, it reasonable to assume, but it might not be the same as Z3. And its not necessarily the same primary/secondary timings that make the difference either.

 

[Markfw]

My concern about Zen 4 moving forward. The sweet spot for DDR5 is 6000mhz according to AMD. The problem with that sweet spot. There are 7200mhz kits already out on the market. Unlike with DDR4 where the speed of the sticks was a slower move forward. DDR5 is coming with high memory frequency out of the box. The latency is what is lagging behind with DDR5 kits. I said earlier that DDR5 6000mhz with 28-30 Cas latency sticks would be what you want. Obviously 28-28-28 sticks @ 6000mhz sticks will be a little bit further out.

Will the memory dividers be a wall for latency with DDR5 in AM5?

I got 64 gig of DDR5 6000 cl32 for $489. I will report anything I find about memory when I get my 7950x (hopefully Tuesday)

 

[Hans Gruber]

Well, we don't know yet how much of an impact latencies will have, do we? I mean, it reasonable to assume but it might not be the same as Z3. And its not necessarily the same primary/secondary timings that make the difference either.

We kind of already do know. The latency on DDR5 is pretty bad but the bandwidth from high frequencies completely offsets higher memory timings. If sticks hit 8000mhz+, then we get into the memory divider issue that plagued Zen 2 and Zen 3. AMD has said that Zen 4 would be a massive memory OC'er. The problem is when they say the sweet spot on Zen 4 is 6000mhz.

Here is what I am talking about with fast memory kits coming to market.

https://www.tomshardware.com/news/teamgroups-speedy-ddr5-7200-c34-ram-hits-retail-for-dollar350

 

[Asterox]

None of what you said explains why that benchmark had the 5600 faster than the 5900X, 5800X, and 5600X in Doom Eternal.

Again, significant doubt. Those benchmarks don't pass the smell test, period.

Test is a garbage, when you see that even the R5 5600 is ahead of the R9 7950X is some game test.

There is no normal or relevant test, which put this kind of nonsense(without verification and explanation)on the table to stand in the rain and frost.If the tester has a good reputation, it is impossible to put this kind of nonsense on any table.

In football terms, or in short this is a very bad and irregular match.

 

[Hans Gruber]

Test is a garbage, when you see that even the R5 5600 is ahead of the R9 7950X is some game test.

There is no normal or relevant test, which put this kind of nonsense(without verification and explanation)on the table to stand in the rain and frost.If the tester has a good reputation, it is impossible to put this kind of nonsense on any table.

In football terms, or in short this is a very bad and irregular match.

If AMD says the IPC of Zen 4 is +15% over Zen 3. It's at least +15%, they do not lie about performance numbers. They are either right on or a little higher. We have to wait for the official release of Zen 4. There have been a ton of leaks of Zen 4 with incomplete bios and engineering samples. We need the full retail chips to see what Zen 4 has to offer.

 

[itsmydamnation]

My concern about Zen 4 moving forward. The sweet spot for DDR5 is 6000mhz according to AMD. The problem with that sweet spot. There are 7200mhz kits already out on the market. Unlike with DDR4 where the speed of the sticks was a slower move forward. DDR5 is coming with high memory frequency out of the box. The latency is what is lagging behind with DDR5 kits. I said earlier that DDR5 6000mhz with 28-30 Cas latency sticks would be what you want. Obviously 28-28-28 sticks @ 6000mhz sticks will be a little bit further out.

Will the memory dividers be a wall for latency with DDR5 in AM5?

WHY DO PEOPLE KEEP BRING THIS UP?@?!?!@?#?@!#?@!

FFS JUST RUN LOWER TIMINGS

you really think your going to be bandwidth bound?

 

[Kuiva maa]

I tried to run the Mankind Divided benchmark at 1080p on my 5900X/6800XT system since I have the game installed already at all ultra (including 8xMSAA) but it returned a variance of 48-87 fps (dirty, all sorts of background processes running) so I won’t even try to guestimate the settings they ran,if it was win10 or even if they used the built in benchmark. As for Doom eternal I don’t know why a cezanne would perform so well there.

 

[Hans Gruber]

WHY DO PEOPLE KEEP BRING THIS UP?@?!?!@?#?@!#?@!

FFS JUST RUN LOWER TIMINGS

you really think your going to be bandwidth bound?

I already answered your statement in my post. I said 6000mhz low latency memory between 28-32. Our moderator Mark has no budget. He says he has 64GB of Cas 32 DDR5 sticks ready to go for his 7950x. People want to know what the maximum DDR5 memory speed AMD will support with a 1:1 memory divider. Once you hit the max speed wall, all you have is lowering the latency timings of your sticks.

With Zen 2 and Zen3, AMD said the sweet spot for DDR4 was 3733mhz. DDR4 sticks were not capable of that speed for several years prior to Zen 2 and the CPU's really were only good up to (Zen & Zen+)DDR4 3200mhz. The maximum speed for the memory controller is 3800-4000mhz for Zen2 and Zen3.

With AM5 and Zen 4, they said the sweet spot for memory speed is 6000mhz. We want to know what the maximum supported DDR5 memory speed will be for the Zen 4 memory controller. DDR5 memory kits are already going up to 7200mhz+ with a baseline DDR5 kit of 4800mhz. The baseline DDR4 kits started @ 2133mhz.

If you compare DDR4 2133mhz to 3800mhz being the max supported speed. I know 4000mhz is possible with 1:1 dividers. With DDR5 4800mhz x 1.78=8544mhz DDR4(2133mhz vs 3800mhz) You should see a maximum speed of DDR5 8544mhz DDR5 maximum speed with 1:1 dividers if the ratio from Zen 3 is the same as Zen4.

I am just ballparking numbers based memory speeds. Cas 14 for DDR4 if doubled should be 28 for top DDR5 kits equal to B-die from Samsung. Obviously a DDR5 28 kit @8000mhz would be a year or two away.

 

[eek2121]

I said several pages ago that the Achilles heel of Zen 4 would be it's memory controller, and I based that statement on preliminary memory benchmarks from Aida.

Gaming stresses the CPUs memory controller and interface more than any other consumer oriented workload, so ostensibly, having a weak memory interface would hurt gaming performance.

And it's funny because that Spanish preview used slower DDR5 for the 12900K than the 7950x and the former still outperformed the latter in the memory tests, which indicates that Alder Lake's memory controller is superior to Zen 4's, and also possibly that Alder Lake may be better at hiding latency than Zen 4.

We kind of already do know. The latency on DDR5 is pretty bad but the bandwidth from high frequencies completely offsets higher memory timings. If sticks hit 8000mhz+, then we get into the memory divider issue that plagued Zen 2 and Zen 3. AMD has said that Zen 4 would be a massive memory OC'er. The problem is when they say the sweet spot on Zen 4 is 6000mhz.

Here is what I am talking about with fast memory kits coming to market.

https://www.tomshardware.com/news/teamgroups-speedy-ddr5-7200-c34-ram-hits-retail-for-dollar350

The question you should ask yourself was if it was 12900k official JEDEC specs that the CPU supports vs. 7950x official JEDEC specs that the CPU supports. Also, the power limits provided by both Intel and AMD should have been quadruple checked.

I'm willing to bet that the answer lie in a convenient configuration issue that a certain pro Intel person decided wasn't important (even though it was). I'm also willing to bet they did it for clicks. Finally, I'll bet the farm their results won't be reproducible by reputable sites such as Gamers Nexus, AnandTech, etc.

At the end of the day, you have to make a call. Either you test them all at "supported" memory speeds and power limits, or you overclock the daylights out of the memory and push power limits to obscene levels for both. You can't mix and match. That is why most reviewers either test at JEDEC or use the same memory/timings for both platforms. It eliminates memory/bandwidth latency and speed differences.

 

[Markfw]

Tomorrow I hope reviews will be out so we can discuss this intelligently. If not, Tuesday for sure.

 

[deasd]

Greymon spoiler warning. If true this fits my personal expectation. Dunno what you guys thought.

https://twitter.com/x/status/1574270885766508544

 

[inf64]

Greymon spoiler warning. If true this fits my personal expectation. Dunno what you guys thought.

https://twitter.com/x/status/1574270885766508544

Also he posted these :

https://twitter.com/x/status/1574273803848650753

"With 300W power unlocked, 13900K productivity perf ties the default 7950X. "
=> Expected. 7950X will be way more power efficient, same performance with ( at minimum) ~25% less power draw, or same power as 13900K with ~20% higher productivity performance.

https://twitter.com/greymon55/status/1574263997536538624
"The 13900K has more gaming perf improvements than the other 13gen SKUs, and I think it will be a bit ahead of the 7950X. "
=> A bit ahead of 7950X does not sound as being comfortably ahead. My guess is maybe ~5% better at best (on the average) when both are running DDR5 6000+.

BTW, we have a leak for pricing in the UK for the 13th gen:


If these prices hold then it looks like AMD will be fine with their Ryzen 7000 pricing model.

 

[Racan]

He also said

https://twitter.com/x/status/1574265209870778368

Which is a bit worrying.

 

[Det0x]

Most interesting in this screenshot is the memory speed @ 7000MT/s..
Highest we have seen to date i believe

 

[Abwx]

Deadline for the reviews :

https://twitter.com/x/status/1574315800835792897

 

[tamz_msc]

Puget Systems review is up:

AMD Ryzen 7000 Series Processors Content Creation Review

AMD has launched their new Ryzen 7000 Series desktop processors (code-named "Raphael") based on the latest Zen 4 architecture. These CPUs support DDR5 and PCIe 5.0, with up to 16 cores with a peak clock speed of 5.7 GHz. Along with the increased frequencies and DDR5 support, AMD has touted a 13%...

www.pugetsystems.com

Archive link:

archive.ph

 

[beginner99]

He also said

https://twitter.com/x/status/1574265209870778368

Which is a bit worrying.

That one like had the 5800x3d firmly beating a 7950x at gaming. And since we know latency is important for gaming, the cache helps a lot with latency and DDR5 is rather bad in terms of latency, it just makes too much sense not be be true.

 

[Panino Manino]

Like we say in this hemisphere, "está saindo da jaula o monstro".
I really thought the embargo was a few days away, didn't expect to see the reviews today. NOW I'm more anxious having to count the hours.

 

[MadRat]

Maybe price on 5800x3D will increase if that's true.

 

[Racan]

That one like had the 5800x3d firmly beating a 7950x at gaming. And since we know latency is important for gaming, the cache helps a lot with latency and DDR5 is rather bad in terms of latency, it just makes too much sense not be be true.

I expected that Zen 4 would fall behind Zen 3D and Akderlake in games but Zen 4 falling behind plain Zen 3 in any game is a bad look. If that's the case they need to get Zen 4 3D out the door ASAP.

 

[Asterox]

As expected, same AMD VCN tech as with Rembrandt APU or no AVI encoding.

AMD confirms Ryzen 7000 integrated RDNA2 graphics features - VideoCardz.com

AMD Ryzen 7000 RDNA2 GPU features confirmed AMD Ryzen 7000 are the first desktop Ryzen non-G series to feature RDNA2 graphics. Ryzen 7000 iGPU has powerful video decoding and display output specs. AMD today reaffirms that the built-in RDNA2 GPU has two Compute Units. As mentioned before, the...

videocardz.com

Let's be realistic, today most people have no idea what AV1 is or where it can be used in general.

Page 31 - Zen 2 APUs/"Renoir" discussion thread

Page 31 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

It can be interesting, and yes you can live normally without an AV1 hardware decoder.

Hm what, that is weird if legit, lol.

"It is worth noting that even NVIDIA RTX 40 series do not support DP 2.0 standard, so NVIDIA, better take a note. This standard will be available with Intel ARC GPUs, though."

 

[nicalandia]

Puget Systems review is up:

AMD Ryzen 7000 Series Processors Content Creation Review

AMD has launched their new Ryzen 7000 Series desktop processors (code-named "Raphael") based on the latest Zen 4 architecture. These CPUs support DDR5 and PCIe 5.0, with up to 16 cores with a peak clock speed of 5.7 GHz. Along with the increased frequencies and DDR5 support, AMD has touted a 13%...

www.pugetsystems.com

Archive link:

archive.ph

Puget System is the only one that does a 10 minute loop on Cinebench where Intel loses 11% performance and AMD only 2%.

Extrapolating those numbers The 13900K will yield about 34,000 points after 10 minutes.

 

[TESKATLIPOKA]

As expected, same AMD VCN tech as with Rembrandt APU or no AVI encoding.

AMD confirms Ryzen 7000 integrated RDNA2 graphics features - VideoCardz.com

AMD Ryzen 7000 RDNA2 GPU features confirmed AMD Ryzen 7000 are the first desktop Ryzen non-G series to feature RDNA2 graphics. Ryzen 7000 iGPU has powerful video decoding and display output specs. AMD today reaffirms that the built-in RDNA2 GPU has two Compute Units. As mentioned before, the...

videocardz.com

Let's be realistic, today most people have no idea what AV1 is or where it can be used in general.

Page 31 - Zen 2 APUs/"Renoir" discussion thread

Page 31 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

It can be interesting, and yes you can live normally without an AV1 hardware decoder.

Raphael and Dragon Ridge will in most cases be paired with dGPU, so the missing AV1 won't be a problem. If even Phoenix was missing It, then that wouldn't be good.