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!

 

[Tuna-Fish]

Really? We have known that for Months now, heck I even draw many core diagrams with half L3$ Trying to understand how AMD will double the cores on the same package size... Come on Dude, please keep up with the news and this thread.

I fully expected the same kind of slimming down that the PS5 FPU saw, that is, remove the FADD pipes and only do FADD in the FMA pipes (at higher latency). It seems like they got more results out of using density-optimized libraries than I'd have expected.

 

[nicalandia]

I fully expected the same kind of slimming down that the PS5 FPU saw, that is, remove the FADD pipes and only do FADD in the FMA pipes (at higher latency). It seems like they got more results out of using density-optimized libraries than I'd have expected.

I even made a drawing of that by making the FPU with a 40% reduction in size as the PS5, I am sure you saw my drawing right?

 

[jpiniero]

MLID so I would take with a large grain of salt, but Bergamo sounds like the better choice for most people. And the rest would want the vcache models.

 

[nicalandia]

Most people would not mind a 32C/64T Zen4c on AM5 on a desktop CPU if it could at least hit 4 Ghz all cores.

 

[Tuna-Fish]

MLID so I would take with a large grain of salt, but Bergamo sounds like the better choice for most people. And the rest would want the vcache models.

It really wouldn't be, imo. The clocks are more important than the cores, unless your workload is embarassingly parallel, which for most people it isn't.

There are people who would like more cores, but for real, most people on the desktop are better served by a 7600X than a hypothetical 32c Zen4c chip, and most server customers are better served by a 48-96 core Genoa than a 128 core Bergamo. (Even if your workload is embarrassingly parallel, as a rule a thread requires just as much memory regardless of how fast it is running, so if you drop your throughput by 25%, which that chart implies, you raise your memory expenses by 33%, which is probably more than the cost of your cpu.)

 

[Asterox]

There's a big chance that 7800X and 7300X actually do not exist.

https://twitter.com/BenchLeaks

I find it very suspicious that 6 unannounced ryzen cpus would be leaked on the same day, each using different motherboard.

You obviously have never heard of "non-existent AMD cpu models".

Page 29 - Discussion - AMD Cezanne/Zen 3 APU Speculation and Discussion

Page 29 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

Of course they exist, or explain which cpu model is being read by Geekbench?

Gigabyte Technology Co., Ltd. B650 AORUS ELITE AX - Geekbench

Benchmark results for a Gigabyte Technology Co., Ltd. B650 AORUS ELITE AX with an AMD Ryzen 3 7300X processor.

browser.geekbench.com

Why would someone fake a 4/8 Zen 4 CPU, what's so weird or exotic about it?There is no logic to fake it, and that's because the CPU really exists and will be used for a specific purpose.

 

[Saylick]

I fully expected the same kind of slimming down that the PS5 FPU saw, that is, remove the FADD pipes and only do FADD in the FMA pipes (at higher latency). It seems like they got more results out of using density-optimized libraries than I'd have expected.

This was my expectation as well, that Zen 4 c would have multiple aspects scaled back reduce the core area, such as a smaller FPU, smaller caches, maybe even less execution units. I thought that it would be a different core all together, so was pleasantly surprised that it's Zen 4 with half the L3 and with denser libraries.

 

[jpiniero]

There are people who would like more cores, but for real, most people on the desktop are better served by a 7600X than a hypothetical 32c Zen4c chip, and most server customers are better served by a 48-96 core Genoa than a 128 core Bergamo. (Even if your workload is embarrassingly parallel, as a rule a thread requires just as much memory regardless of how fast it is running, so if you drop your throughput by 25%, which that chart implies, you raise your memory expenses by 33%, which is probably more than the cost of your cpu.)

The standard Epycs mostly have boosted in the 3.5-3.7 range. So Bergamo is in theory not too far off from that.

 

[uzzi38]

Sorry, but this be a lesson to you all.

Chips and Cheese on Twitter: "Hello you fine Internet folks! I was not expecting to have an article for y'all today but, we are covering just how easy it is to fool CPUID and make legitimate looking leaks. Hope y'all enjoy! https://t.co/vvXeSriZUc" / Twitter

 

[nicalandia]

Sorry, but this be a lesson to you all.

Chips and Cheese on Twitter: "Hello you fine Internet folks! I was not expecting to have an article for y'all today but, we are covering just how easy it is to fool CPUID and make legitimate looking leaks. Hope y'all enjoy! https://t.co/vvXeSriZUc" / Twitter

Thanks, But NO THANKS..!

 

[Harry_Wild]

My ideal AM5 is AMD Ryzen 9 7950 65W CPU!😁💰👍

 

[nicalandia]

My ideal AM5 is AMD Ryzen 9 7950 65W CPU!😁💰👍

Just get the 7950X and put it on ECO Mode..!!

 

[moinmoin]

Zen 4C cores = Zen 4 cores except for cache? No one here ever said that. We said same ISA. Not the same thing.

I felt like a broken record repeating that to me Zen 4c/Bergamo appears to be AMD extending the TAM to servers/cloud for their previously mobile/APU only cores. Which as a reminder are exactly the same as desktop/server cores except higher density and half the cache.

Maybe you missed my posts like:

Page 165 - Discussion - Speculation: Zen 4 (EPYC 4 "Genoa", Ryzen 7000, etc.)

Page 165 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

Page 2 - Question - AMD Phoenix/Zen 4 APU Speculation and Discussion

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

 

[A///]

Thanks, But NO THANKS..!

While educative it was a stupid stunt. Any future data coming from that publication may be fabricated and should be scrutinized far more than any other publication. Faked cpuid uploads are not new phenomena but they've never been looked at positively when exposed by the leading party.

My ideal AM5 is AMD Ryzen 9 7950 65W CPU!😁💰👍

That's the spirit, Harry. You've found the missing 9 on your keyboard old boy.

Just get the 7950X and put it on ECO Mode..!!

I may have my numbers jumbled up here, but is the 65 watt eco mode not for the dies themselves and not the total package? @Markfw mentioned something about the IOD being a separate number but it only being 12-13 extra watts.

 

[moinmoin]

I may have my numbers jumbled up here, but is the 65 watt eco mode not for the dies themselves and not the total package? @Markfw mentioned something about the IOD being a separate number but it only being 12-13 extra watts.

TDP/PPT affects the whole package, ECO mode just changes that. Ryzen Master shows two separate power values, CPU Power and SOC Power which should relate to CCD and IOD respectively and peak at PPT when taken together.

 

[A///]

TDP/PPT affects the whole package, ECO mode just changes that. Ryzen Master shows two separate power values, CPU Power and SOC Power which should relate to CCD and IOD respectively and peak at PPT when taken together.

Thanks for the info. Is Ryzen master a lightweight program or are their lightweight alternatives?

 

[Markfw]

Thanks for the info. Is Ryzen master a lightweight program or are their lightweight alternatives?

Not sure what you mean, but I run it on my 5950x's and 7950x's. Oh, but I set ECO by setting the CO boost in PBO to -25 is bios. I just use Ryzen master to watch everything (like how that set PPT to 142)

 

[A///]

Not sure what you mean, but I run it on my 5950x's and 7950x's.

Is there a more basic tool that uses less system memory? Like inspector being a lighter weight option over afterburner for so many years.

 

[Markfw]

Is there a more basic tool that uses less system memory? Like inspector being a lighter weight option over afterburner for so many years.

It uses 128 meg ram on this 5950x. I use it then turn it off, you only need to monitor after startup. The Ryzen tool built into the CPU does all the hard work (keeping temp/power in check)

 

[moinmoin]

Thanks for the info. Is Ryzen master a lightweight program or are their lightweight alternatives?

Uh, I mostly use Linux so there are surely people better positioned to answer that than me, but to me Ryzen Master seem rather heavyweight (requires to disable Virtualization Based Security (VBS) for instance but also offers tweaking many CPU parameters in real time). HWiNFO seems to be more commonly used just for monitoring all possible hardware values.

 

[A///]

It uses 128 meg ram on this 5950x. I use it then turn it off, you only need to monitor after startup. The Ryzen tool built into the CPU does all the hard work (keeping temp/power in check)

Let me get this straight. You can set it to eco mode, turn ryzen master off or exit it and it releases that 120 megs but keeps the processor in eco mode until next startup? It doesn't do any scheduling right? 6pm to 12 pm full power and 6 am to 5 pm eco mode every day?

 

[Mopetar]

AMD just needs to concentrate on V-Cache and bring it to market ASAP, and create new SKUs by adding V-Cache to more products (7700/7800x, 7900x, 7950x, even 7600x)

V-cache is really only useful for gaming. If you look at the 5800X3D reviews it gets beat out by a regular 5800X in almost everything outside of gaming. Unless you primarily use a computer for one of those small number of other workloads, v-cache isn't going to give you a better experience.

Perhaps the one area where it shines, which isn't really captured by benchmarks, is improving general system and application responsiveness when running a mixed app workload. The extra last level cache means that other applications have more room in the cache. I doubt that too many people are going to be willing to pay the ~$100 extra for the v-cache model though.

AMD just needs to be more price competitive with Intel in the midrange, because the E-core spam on the i5 works well enough for the 13600K to pull ahead in many different thread heavy workloads and the chip does well enough in gaming that AMD doesn't have any real advantage there either.

 

[moinmoin]

Let me get this straight. You can set it to eco mode, turn ryzen master off or exit it and it releases that 120 megs but keeps the processor in eco mode until next startup? It doesn't do any scheduling right? 6pm to 12 pm full power and 6 am to 5 pm eco mode every day?

Keep in mind than any changes you can do in Ryzen Master you should be able to do in the BIOS as well. I'm not aware of a scheduling feature for either of the two though.

 

[A///]

Keep in mind than any changes you can do in Ryzen Master you should be able to do in the BIOS as well. I'm not aware of a scheduling feature for either of the two though.

Isn't that only if the motherboard maker implements hard power limits?

 

[moinmoin]

Perhaps the one area where it shines, which isn't really captured by benchmarks, is improving general system and application responsiveness when running a mixed app workload.

Which is quite a major area when thinking about it. Benchmarks are run in isolation to make the results comparable but are quite artificial that way. The more cores and cache a CPU has the more it is capable of running multiple of them at once. 5800X3D loses where pure frequency is the bottleneck, but as soon as it moves to cache limited multitasking it will shine. Games, often being unoptimized mix of different concurrent workloads, are just "better" at exposing that advantage through a single application.

Isn't that only if the motherboard maker implements hard power limits?

If you got a board capable of overclocking its BIOS should contain the relevant options.