Speculation: Ryzen 4000 series/Zen 3

[amrnuke]

Makes me (again) wish for all benchmarks to include the amount of joules consumed.

What's clear is that Apple throws a lot of energy at the speed problem, and tosses efficiency out the window at those times.

What's not clear is how often a normal user pushes the Lightning core to its max, or whether people are largely using the Thunder cores when they "Like" their high school acquaintance's word-image hymn post on Facebook, making the power consumption numbers for Lightning largely useless if that's the case.

 

[Carfax83]

In AMD's arguable most important market, datacenters, this is by far not as clear cut. There the combination of high frequency plus high IPC is not as applicable, instead it's pure IPC since lower frequency translates to higher effciency. Apple in this regard is no direct competitor, but a good target to strive for for all chip designers.

I agree with this, but my point was that as far as raw performance goes, high IPC along with high frequency is the ultimate combo. Also, as amrnuke already illustrated several pages ago, Zen 3 manages to increase both performance and efficiency simultaneously on the same node, so they definitely know what they are doing and won't cede ground to ARM easily.

AMD has shown that you can have your cake and eat it too, by manufacturing a high frequency, high IPC and high efficiency CPU. Actually, Intel used to nail all three of those attributes but their struggles with their process node advancements have clearly stymied them in that regard.

 

[Kenmitch]

What does all the ARM crap have to do with Zen 3 speculation thread?

I'm probably not the only one that has to scroll back up and see what thread I'm actually in. I thought I accidently clicked on the wrong thread again.

 

[amrnuke]

What does all the ARM crap have to do with Zen 3 speculation thread?

I'm probably not the only one that has to scroll back up and see what thread I'm actually in. I thought I accidently clicked on the wrong thread again.

One of the key server/datacenter benchmarks for the Zen 3 chips is going to have to be IPC, efficiency, etc. By calculating Zen 3's estimated IPC and efficiency and lining that up against server/datacenter competition (namely, what ARM is positioning themselves to do with highly power-efficient designs, albeit with less top-end performance), we can provide some speculation as to were Zen 3 might position itself overall.

That goes for comparisons to Apple (AMD's Zen 3 laptop competition next summer, possibly) as well.

 

[Carfax83]

What does all the ARM crap have to do with Zen 3 speculation thread?

I'm probably not the only one that has to scroll back up and see what thread I'm actually in. I thought I accidently clicked on the wrong thread again.

I agree that we shouldn't have ARM stuff in a Zen 3 thread generally speaking, but now that we have a good idea of Zen 3's performance attributes (which we never had until the official Zen 3 announcement and various benchmark leaks), comparisons to other CPUs and other architectures are inevitable.

A CPU's worth is ultimately decided by how it compares to the competition, and ARM is definitely positioning itself as a competitor to AMD and Intel.

 

[Ajay]

I agree that we shouldn't have ARM stuff in a Zen 3 thread generally speaking, but now that we have a good idea of Zen 3's performance attributes (which we never had until the official Zen 3 announcement and various benchmark leaks), comparisons to other CPUs and other architectures are inevitable.

A CPU's worth is ultimately decided by how it compares to the competition, and ARM is definitely positioning itself as a competitor to AMD and Intel.

Yeah - most ARM threads die off pretty quickly - so the conversations wind up in some x86 thread.

 

[Kenmitch]

Yeah - most ARM threads die off pretty quickly - so the conversations wind up in some x86 thread.

Maybe they die off for a reason.

Most people would expect to see something like this posted.

AMD-Based Finnish Supercomputer Will Be One of World's Most Powerful in Mid-2021

At 0.55 ExaFLOPS, EuroHPC's LUMI will challenge world's most powerful systems

www.tomshardware.com

It's kind of WTF? when you see the ARM stuff.

 

[soresu]

Maybe they die off for a reason.

Most people would expect to see something like this posted.

AMD-Based Finnish Supercomputer Will Be One of World's Most Powerful in Mid-2021

At 0.55 ExaFLOPS, EuroHPC's LUMI will challenge world's most powerful systems

www.tomshardware.com

It's kind of WTF? when you see the ARM stuff.

That's not the reason.

The average person here might think a supercomputer is interesting for a bit, but how many of us will ever comes across an actual supercomputer by modern standards in person?

The reason is that ARM platforms are still a mishmash of a mess and the more easily customisable SBC's lag new core announcements by years - it's kind of sad that we got all excited for RPi4 being A72 based in 2019 isn't it?

We are still waiting on the first A76 based SBC's 2 and a half years later, even streamers only just started seeing A55 cores with the new Chromecast.

Not to mention that there probably still isn't an SBC that can even match the 5 1/2 year old Tegra X1 for graphics horsepower, let alone a GPU from any modern flagship phone SoC that could smite the TX1 with ease.

If there were a relatively simple way to disassemble a state of the art smartphone to reuse the SoC/memory as an SBC then we could really get the ball rolling tout suite.

 

[Antey]

5900X 4,5GHz all cores

System manufacturer System Product Name - Geekbench

Benchmark results for a System manufacturer System Product Name with an AMD Ryzen 9 5900X processor.

browser.geekbench.com

 

[amrnuke]

5900X 4,5GHz all cores

System manufacturer System Product Name - Geekbench

Benchmark results for a System manufacturer System Product Name with an AMD Ryzen 9 5900X processor.

browser.geekbench.com

That would lead current single-core chart, and between the Intel 10920X (12/24) and TR 2950X (16/32) on multicore. Seems there are still some kinks to workout on multi-core.

 

[leoneazzurro]

Multicore is probably starting to be limited by interconnect and bandwidth available per core. A IF clock/RAM scaling bench should give an answer to this.

 

[majord]

Multicore is probably starting to be limited by interconnect and bandwidth available per core. A IF clock/RAM scaling bench should give an answer to this.

honestly having poured over the slides, I'm expecting lower SMT yield with Zen 3 vs Zen 2. But so hard to tell without knowing exact clockspeeds under all core load.. this locked 4.5Ghz result sort of backs it up, but OTOH the whole idea is a bit in contrast to AMD's perf/watt claims, which typically would be for total throughput.

 

[Magic Carpet]

I hope, the new cpu design will be able to run this better.

 

[Carfax83]

Multicore is probably starting to be limited by interconnect and bandwidth available per core. A IF clock/RAM scaling bench should give an answer to this.

Or perhaps, Geekbench is just a terrible benchmark (for desktop CPUs at any rate) that doesn't scale well across that many cores/threads? I'm probably just biased against Geekbench, and I've only ran it once. During that time, I noticed that it barely taxed my CPU (didn't even get my aging overclocked 6900K CPU into turbo boost mode) and it didn't recognize that my CPU had quad channel memory. So after my experience, I think Geekbench would be more reliable for single core testing rather than a high core count multicore/thread CPU like a 5900x.

 

[Gideon]

I hope, the new cpu design will be able to run this better.

View attachment 32263

A bit OT but I was really looking forward for that, but they majorly screwed up the remake. So-much that IMO it's better to just play the original.

Digital Foundry has a good review of the problems (and the few good parts).

In this current form - made from the console port - it has a number of glaring issues:

• the VTOL mission has been removed
• a lot of graphical options missing that were in the original (many alpha-particle effects totally missing, etc)
• while parts and pieces have improved (Global Illumination, shadows) they completely screwed up the color-scheme and feel of the original, over-saturating stuff.
• The models haven't been touched at all (except the nanosuite) all are the same as in the original
• New textures are a hit-and-miss with some straying quite far from the original (snow missions having a lot more dirt in them)
• Even the added Raytracing features are buggy and cause hiccups enough that even Digital Foundry recommends disabling the feature
• What's worse, despite latest Cry-engines (Crysis 3) being excellently multithreaded, the one in this is pretty much as bad as the original, thus only loads one thread.

I mean just look at this image:

Spoiler: Cysis vs Crysis Remastered

It's the original Crysis on the left, while the remaster on the right. If it weren't for the improved nanosuite textures I'd think the latter is original running @ low (it looks even worse in motion btw)

Overall, in it's current state the Crysis Remastered it's a terrible cash-grab. Ignore it and rather play the original (with perhaps carefully selected HD Texture pack mods).

 

[Hans de Vries]

That would lead current single-core chart, and between the Intel 10920X (12/24) and TR 2950X (16/32) on multicore. Seems there are still some kinks to workout on multi-core.

The single core result here (1525) is highly constrained by the fixed 4.5 GHz clock: https://browser.geekbench.com/v5/cpu/search?q=5900x


The single core result will be well above 1700 with the new Geekbench 5.3 coming out November 2nd just before the launch. The updated version uses vector EAS instructions just like Ice Lake and Tiger Lake currently do.

 

[Antey]

Ryzen 5900X vs 3900X

https://browser.geekbench.com/v5/cpu/4306615 (5900X, max freq 4.50 GHz, CAS Latency 22)

https://browser.geekbench.com/v5/cpu/4314849.gb5 (3900X, max freq 4.45 GHz, CAS Latency 18)

they are not exactly equal but well...

improvements over 3900x:

ST Crypto workloads:

AES-XTS: +18%

ST int workloads:

Text Compression: +22%
Image Compression: +15%
Navigation: +15%
HTML5: +23%
SQLite: +32%
PDF Rendering: +20%
Text Rendering: +13%
Clang: +24%
Camera: +2%

ST Floating point workloads:

N-Body Physics: +6%
Rigid Body Physics: +16%
Face Detection: +17%
Horizon Detection: +17%
Image Inpainting: +14%
HDR: +21%
Ray Tracing: +17%
Structure from Motion: +24%
Speech Recognition: +17%
Machine Learning: +17%

 

[Kenmitch]

The single core result here (1525) is highly constrained by the fixed 4.5 GHz clock: https://browser.geekbench.com/v5/cpu/search?q=5900x
View attachment 32266

The single core result will be well above 1700 with the new Geekbench 5.3 coming out November 2nd just before the launch. The updated version uses vector EAS instructions just like Ice Lake and Tiger Lake currently do.

Does it give others the impression that the all-core boost was around 4.4GHz or so during the non 4.5GHz runs?

 

[leoneazzurro]

Or perhaps, Geekbench is just a terrible benchmark (for desktop CPUs at any rate) that doesn't scale well across that many cores/threads? I'm probably just biased against Geekbench, and I've only ran it once. During that time, I noticed that it barely taxed my CPU (didn't even get my aging overclocked 6900K CPU into turbo boost mode) and it didn't recognize that my CPU had quad channel memory. So after my experience, I think Geekbench would be more reliable for single core testing rather than a high core count multicore/thread CPU like a 5900x.

Yes, Geekbench especially under Windows is simply terrible and in fact Linux scores are well above those under Windows.

 

[Kedas]

They could have shown the 5950X but no they added the cheapest one to make it even more embarrassing

PassMark CPU Benchmarks - Single Thread Performance

Benchmarks of the single thread performance of CPUs. This chart comparing CPUs single thread performance is made using thousands of PerformanceTest benchmark results and is updated daily.

www.cpubenchmark.net

 

[Ajay]

honestly having poured over the slides, I'm expecting lower SMT yield with Zen 3 vs Zen 2. But so hard to tell without knowing exact clockspeeds under all core load.. this locked 4.5Ghz result sort of backs it up, but OTOH the whole idea is a bit in contrast to AMD's perf/watt claims, which typically would be for total throughput.

This tends to happen when the mis-predict rate goes down (fewer pipeline flushes). There are fewer stalls and threads must compete more competitively for resource use. Net throughput goes up, but the gains from SMT go down. Reduced memory/cache latency would also reduced thread stalls (mem waits). There could be other reasons, once Ian gets to do a deep dive on Zen3, we'll get a better idea.

 

[itsmydamnation]

But if they increased load and store width and queue depths, one of the biggest SMT bottlenecks will be relieved. So i wouldn't be surprised to see SMT yeild increase or stay the same. I also wouldn't be surprised to see memory bandwdith , IO die etc to be a bottleneck when scaling workloads.

The interesting rumors arethat AMD are doing both 12nm and 7nm IOD for EPYC, if so that would be a very interesting comparison and give good insight to what warhol/DDR5 might bring.

 

[Carfax83]

The interesting rumors arethat AMD are doing both 12nm and 7nm IOD for EPYC, if so that would be a very interesting comparison and give good insight to what warhol/DDR5 might bring.

So Warhol is a real thing then? I thought it was just a rumor.

 

[soresu]

I'm starting to wonder whether Rembrandt being 6nm means that we might see 6nm Zen3 CCD's for Warhol even though the roadmap said 7nm.

Same for RDNA2, it doesn't make sense to shrink both CPU and GPU uArch's just for an APU.

 

[moinmoin]

I'm starting to wonder whether Rembrandt being 6nm means that we might see 6nm Zen3 CCD's for Warhol even though the roadmap said 7nm.

Same for RDNA2, it doesn't make sense to shrink both CPU and GPU uArch's just for an APU.

Isn't N6 reusing N7's design rules?