AMD Ryzen (Summit Ridge) Benchmarks Thread (use new thread)

[Arachnotronic]

So then AMD got the IPC increase they claimed, but had to sacrifice frequency to do so, as was expected.

Project management triangle...good, fast, cheap. Pick two.

 

[Azuma Hazuki]

So this is exactly what I expected since seeing the claims of 40% IPC uplift over EXV: basically a Sandybridge-E chip for the consumer at what I hope is going to be a reasonable price, with support for such things as NVMe.

Like I said before: *perfect* little Gentoo box. Zen is precisely what i expected and if i ever get enough money it'll likely be in my next build.

 

[Dresdenboy]

Project management triangle...good, fast, cheap. Pick two.

The way, Lisa blamed GloFo for Polaris yields might also reflect, how happy they are with their #1 silicon supplier regarding CPUs.

If the ES base clock would still be the final base clock in a few months, AMD would be ~0.5GHz short of a target I heard not too long ago. This could mean: What we're seeing here, is not, what AMD expected based on their knowledge of the process and related tools.

 

[Arachnotronic]

The way, Lisa blamed GloFo for Polaris yields might also reflect, how happy they are with their #1 silicon supplier regarding CPUs.

If the ES base clock would still be the final base clock in a few months, AMD would be ~0.5GHz short of a target I heard not too long ago.

Did Lisa indicate if it was parametric yields causing the trouble or if it's a case of functional yields?

 

[frozentundra123456]

Disregarding CPU-framerate for a moment, which is consistent sometimes. It's possible to just filter results by what michaelyuan.feng used. At Standard this leaves only 14 Entries, which are simply ranked by the benchmark score, from first to last:
i5-6600K - 6800
i7-6700K - 6600
i7-6700K - 6600
i7-6700K - 6500
i7-4820K - 6300
i7-3930K - 6200
i5-6600K - 6100
i7-4790 - 5900
Zen ES - 5300
i5-4670K - 5000
FX-8310 - 4700
FX-8370 - 4500
FX 8320 - 4000
A10-7890K-3000

There is even a 6/12 Sandy Bridge-E [3.2/3.8 GHz] in the mix scoring 6200, which IMO provides the fairest comparison in terms of cores/threads used. Adjust it for frequency (3.2 -> 2.8 GHz = minus 12.5%) and we get a score of 5425, which again is very close to Zen.

I think the most correct way to do the calculation is thus: (assume turbo of 3.5 for intel and 3.0 for Zen)

If equal IPC to SB, Zen should score 5300 x 3.5 /3.0 = 6183. And of course Zen has 2 more cores so if the game is using all 8 cores, it should score 6183 x 8/6 = 8244.

 

[pakotlar]

I think the most correct way to do the calculation is thus: (assume turbo of 3.5 for intel and 3.0 for Zen)

If equal IPC to SB, Zen should score 5300 x 3.5 /3.0 = 6183. And of course Zen has 2 more cores so if the game is using all 8 cores, it should score 6183 x 8/6 = 8244.

Sure, if per score scaling were linear, which it probably isn't.

Not sure how good the source, but http://semiaccurate.com/2016/03/01/investigating-directx-12-cpu-scaling/ suggests scaling isn't particularly good above 4 cores.

 

[William Gaatjes]

At work, i have a brand new Intel I7-6700 system. At home i have an AMD A10-6700. With both i am very happy.
And am amazed by the performance from the I7-6700 when working with it. Benchmarks tell only so much.
But using a system and experience the performance difference is reality. I have always been happy with my home system and i never have to wait.I am still very happy with it.
When the more demanding games come out in the near future, and my A10-6700 is no longer sufficient, i will have to upgrade. When i upgrade my home system by maybe the 4th quarter of next year, and zen would be seriously disappointing, i will make the step over to an I7-6700K. It is a good investment that will last for several years. Even as an "older" model. Even when bought about 1.25 year from now it is still good. I am planning for a RX 480 and it seems the RX 480 really needs a fast cpu to show its maximum power in Vulkan/DX12.

 

[superstition]

Exactly! I am looking forward to the history lesson down the road as to why they didnt focus on 4 instead of this 8c/16t monstrum that has zero mainstream appeal.

I thought Zen is supposed to be designed to be relatively inexpensive to produce, in terms of its chip size (yields). So, by not including an iGPU doesn't that give AMD space for the extra cores?

Remember how much space was used by the Broadwell C iGPU versus its four cores. Cutting out that iGPU makes it seem like there would be a lot of space to use for extra cores. And, as far as I know, Broadwell C's chip size isn't all that huge. It seems reasonable that AMD will offer a 4/8 part at a higher clock aimed at gamers.

 

[frozentundra123456]

Sure, if per score scaling were linear, which it probably isn't.

Not sure how good the source, but http://semiaccurate.com/2016/03/01/investigating-directx-12-cpu-scaling/ suggests scaling isn't particularly good above 4 cores.

That is probably true, although that is why I said it is not a very good benchmark. Seems the numbers are all over the place, and nobody really knows what the limiting parameters are.

But I see that as a problem for zen for the desktop, at least for gaming. 8 cores sounds great, but I think even with DX12, 6 cores + HT could be the sweet spot, or maybe even still a quad plus HT.

 

[superstition]

I think even with DX12, 6 cores + HT could be the sweet spot, or maybe even still a quad plus HT.

I'm still not seeing what's stopping AMD from disabling some cores and increasing the clocks? It seems a more viable strategy that putting in a weak (as compared with a discreet GPU) iGPU that uses half the chip.

While we don't know how high the process and design will allow AMD to clock (and how that will change over time) it seems pretty reasonable to assume that disabling some of the cores will enable a higher clock within the same TDP.

 

[cytg111]

I thought Zen is supposed to be designed to be relatively inexpensive to produce, in terms of its chip size (yields). So, by not including an iGPU doesn't that give AMD space for the extra cores?

Remember how much space was used by the Broadwell C iGPU versus its four cores. Cutting out that iGPU makes it seem like there would be a lot of space to use for extra cores. And, as far as I know, Broadwell C's chip size isn't all that huge. It seems reasonable that AMD will offer a 4/8 part at a higher clock aimed at gamers.

They want to retake some server space, then yea a many core chip makes sense. But what is the argument then? That while the part passes server validation for another year at AMD-HQ-LABS it is good enough to serve mainstream? Is mainstream adoptation part of the validation process? are we beta testing this hardware for primetime operation in datacenters?
With a 4C/8T and no IGP they could slice 2x as many chips off a wafer. Step 2 : Profit.
I am sure there is a business case for the 8c/16t .. would love to hear it(rather than speculate our collective ..... off).

 

[whm1974]

They want to retake some server space, then yea a many core chip makes sense. But what is the argument then? That while the part passes server validation for another year at AMD-HQ-LABS it is good enough to serve mainstream? Is mainstream adoptation part of the validation process? are we beta testing this hardware for primetime operation in datacenters?
With a 4C/8T and no IGP they could slice 2x as many chips off a wafer. Step 2 : Profit.
I am sure there is a business case for the 8c/16t .. would love to hear it(rather than speculate our collective ..... off).

Personally I rather have an extra 4C/8T that I will sometimes use instead of the iGPU I never will.

 

[jpiniero]

The 8C16T was mainly so that they could generate some hype with enthusiasts, while having the easiest path to getting some sales.

 

[William Gaatjes]

Did no one in this thread bother to ask the question of how a 4.2Ghz 6700K is 63% faster than a 4.5Ghz i7 2600K? That's almost a 75% increase in IPC for Skylake over Sandy.

Before we even try to compare Zen, I dare anyone in this thread to find me 1 AAA PC game made in the last 10 years where a stock 6700K is 63% faster than a 4.5Ghz 2600K....Go ahead, I'll be waiting.

While at it, then explain to me as well how i5-6400 gets leveled by i5 2500K 4.5Ghz in the same benchmark after seeing > 70% IPC advantage for Skylake over Sandy....

I have 0 interest in Zen as I already bought a 6700K a long time ago but CPU benchmarks from Ashes are about as useful as used toilet paper to me. I don't know a single game in the world where a stock 6700K would level a 4.5Ghz 2600K by more than 60%. The only way I see something like this happen is with the latest AVX/2 instruction set(s). What developer makes AAA games with that?

Finally, we know that 6700K can't overclock much beyond 4.8Ghz on air. How do we know that 3.2Ghz 8-core Zen cannot overclock to 4.8Ghz? I am not saying it can, but we also cannot yet rule out that Zen could have a lot better % overclocking headroom than Skylake.

Either way, I would pick 6700K over 6900/6950X, which means Zen was always a non-starter for me. Even if 8-core Zen matched 6700K in IPC, I still wouldn't buy it since 99% of PC games don't use more than 4C+HT, which means I'd be wasting $$$ I could use to get a 4K monitor or a faster GPU instead. That's why to me AMD should have went all in on 4 core fast IPC CPU. I have 0 use for an 8 core CPU as I have moved away from distributed computing over the years.

OTOH, let's say I actually needed an 8-core CPU -- Intel's cheapest is $1089 USD. That gives AMD room to price Zen between $350-699 and still undercut the 6900K to the point where they aren't competitors. Thus, for me Zen was never going to live up to gaming expectations but for someone who wants a multi-threaded CPU for [insert whatever tasks] a $545 8-core Zen would cost 1/2 of a 6900K. Is that a FAIL? I don't know, I don't buy $1090 CPUs for productivity but on paper if 6900K isn't 2x faster, 8-core Zen priced < $600 has a market.

I agree. on buying the I7-6700K.
But i get the strong feeling that pretty soon, the marketing department from AMD and Microsoft as well from several game developers will start to work to promote a pc that is very similar in hardware setup to the consoles, only of course a lot more faster.
I assume here that a pc 8 core zen + polaris would make it for developers of games a lot easier to port a game to the pc. Especially since the consoles also have a (albeit weaker but 8 core cpu) + polaris. derivate. Kind of a general unificationing of the hardware, the game developers hardly have anything to tweak anymore.

 

[The Stilt]

I'm still not seeing what's stopping AMD from disabling some cores and increasing the clocks?

The 14nm LPP process does. If you can do 3050MHz on all cores, then your maximum single threaded frequency isn't limited to 3200MHz by the TDP.

 

[bystander36]

Disregarding CPU-framerate for a moment, which is consistent sometimes. It's possible to just filter results by what michaelyuan.feng used. At Standard this leaves only 14 Entries, which are simply ranked by the benchmark score, from first to last:
i5-6600K - 6800
i7-6700K - 6600
i7-6700K - 6600
i7-6700K - 6500
i7-4820K - 6300
i7-3930K - 6200
i5-6600K - 6100
i7-4790 - 5900
Zen ES - 5300
i5-4670K - 5000
FX-8310 - 4700
FX-8370 - 4500
FX 8320 - 4000
A10-7890K-3000

There is even a 6/12 Sandy Bridge-E [3.2/3.8 GHz] in the mix scoring 6200, which IMO provides the fairest comparison in terms of cores/threads used. Adjust it for frequency (3.2 -> 2.8 GHz = minus 12.5%) and we get a score of 5425, which again is very close to Zen.

The results just aren't what we were hoping for, but the good news is this is just a sample chip. There is definitely possibilities of improvement on the final product. The bad news is that it looks like the per core performance is not all that promising.

 

[inf64]

This is probably the worst benchmark one could use to judge a new (ES) CPU. If it was anything else we could draw some conclusion, but that pos of a game/benchmark is completely useless.
I still think AMD needs 3.2 base and 3.7Ghz turbo clock on these 8C/16T parts in order to be semi-competitive with part of intel's HEDT line. Hopefully they will get there.

 

[krumme]

For a month we have more or less known 2.8 3.2 was what we would get.
Even with a plus 0.5 base in a process more performant than gf it would never be a good desktop gaming cpu vs skylake 6700.
Never.

Its a server and low power cpu design. Server mobile consoles. Get used to it. No cheap 6700 or 6900 comming your way.

I went into this thread thinking i would see the usual bad amd benchmarks but then my eyes catches som fps numbers....doa. I havnt even cared to read that stuff. I would like some gb numbers as a starter.

I would like to know more about the purpose of this desktop product? To me it simply looks like validation phase and a product without buyers. Only hurting brand.

As they can aparently have 32 cores in 180w for servers tdp why no go 8c in 45w low freq in a mobile workstation? It has a clear market where it would bring a new performance. Or a 4c 22w part. I could use that for my ultrabook.
Plenty of good opportunities where 3.2 could stay but base just be lower.

That the process is 0.5 lower than expected is only a problem if you are stuck on old ideas and unflexible. Efficiency just need to be there ofcource.

 

[TheELF]

I agree. on buying the I7-6700K.
But i get the strong feeling that pretty soon, the marketing department from AMD and Microsoft as well from several game developers will start to work to promote a pc that is very similar in hardware setup to the consoles, only of course a lot more faster.
I assume here that a pc 8 core zen + polaris would make it for developers of games a lot easier to port a game to the pc. Especially since the consoles also have a (albeit weaker but 8 core cpu) + polaris. derivate. Kind of a general unificationing of the hardware, the game developers hardly have anything to tweak anymore.

That's not at all how it works.
Consoles have two separate dual-module CPUs (APUs) the console OS runs on the two first cores of the first CPU,and only on those two cores,leaving six cores for the game,now communication between the two CPUs is slower then communication within the same CPU so devs use the 4 cores of the second CPU for the demanding threads and the two "leftovers" for secondary threads (that's why you still see next to no scaling with over 4 cores)
Anyway the only reason we have bad console ports is because the consoles are not a multitasking environment, the games run all alone on their cores with nothing bothering them and devs just can't be bothered to adhere to multitasking rules so the games mess up windows's multitasking/task manager/whatever.

 

[SAAA]

The 14nm LPP process does. If you can do 3050MHz on all cores, then your maximum single threaded frequency isn't limited to 3200MHz by the TDP.

That's not telling much if we take Intel as example: recent high core count have 4-500MHz difference from base to single turbo clocks, and we know the process can definitely tolerate much more given Skylake 4.2GHz stock turbo, 22nm 4.4GHz turbo for Haswell etc.
There's that boost 3 something that pulls more but leave it aside and if you don't overclock 6950K is 3.5GHz at best. With 140W TDP I would have given it a GHz more single core turbo and binning could have been done on the die itself with 10 friggining cores to choose from.
Fact is also the quad core had the same 2.8-3.2GHz range so it suggest me that it's an ES limititation rather than the process. I hope they get more aggressive with turbo and while retaining ~3GHz base go up to 4GHz single for Zen, that would make up for some performance on everyday tasks.

 

[DrMrLordX]

As you can see from a few different graphs, the game clearly uses at least 8 cores, if not more.

Uh, no? The top CPUs in the graph you posted are those with the largest l3 cache. Cache performance has pushed the 5960x over various Skylake quads in other titles that are not known to be major thread hogs.

The chart showing the same CPU with different core counts enabled plus HT disabled/enabled seems to suggest that AotS favors physical cores over logical cores, and that the thread count maxes out somewhere between 6-8. It's still not absolutely conclusive since the chip has the benefit of the full shared l3 regardless of configuration.

Am I the only one who thinks that it doesnt look half bad?

No. I know this benchmark is a bit funky in terms of which CPU beats which CPU at which settings etc., but if we put that aside for the moment and compare Summit Ridge to Vishera assuming:

8 or fewer threads
Summit Ridge running at 3.2 GHz
8350 running at 4.0 GHz

That still gives Summit Ridge a 72.6% advantage in IPC (yes, I'm using the term IPC here, even though this program uses more than one thread, and you can't stop me, nyah).

Vishera is a 2012 CPU. Summit Ridge is (effectively) a 2017 CPU. It took AMD 5 years to boost IPC by 72.6% in . . . well, this one particular game. Still, that's hyuuuuuuuge! Almost. Since people are now discussing AVX/AVX2 being a possible factor in this benchmark, it's unknown how Summit Ridge's support for those instruction sets might affect its advantage over Vishera (Vishera's AVX support = teh suk, not sure if xOP is supported by AotS, but probably not).

Clockspeed is still a bugaboo. I think 14nm LPP is enough to get AMD's foot in the door of some server rooms that haven't been open to Piledriver-based solutions. All AMD has to do is use a better node, such as the one developed for POWER9.

 

[The Stilt]

That's not telling much if we take Intel as example: recent high core count have 4-500MHz difference from base to single turbo clocks, and we know the process can definitely tolerate much more given Skylake 4.2GHz stock turbo, 22nm 4.4GHz turbo for Haswell etc.
There's that boost 3 something that pulls more but leave it aside and if you don't overclock 6950K is 3.5GHz at best. With 140W TDP I would have given it a GHz more single core turbo and binning could have been done on the die itself with 10 friggining cores to choose from.
Fact is also the quad core had the same 2.8-3.2GHz range so it suggest me that it's an ES limititation rather than the process. I hope they get more aggressive with turbo and while retaining ~3GHz base go up to 4GHz single for Zen, that would make up for some performance on everyday tasks.

On recent 8C/16T Intel designs the deltas between base and turbo frequencies are big as ever, in a matching TDP class.
When you are not limited by the TDP, then there is no need for a large delta between the turbo & base frequencies

Xeon E5-2650 V2 (8C/16T) - 2.6GHz / 3.4GHz - 95W - dF 800MHz (Ivy Bridge-EP)
Xeon E5-2640 V3 (8C/16T) - 2.6GHz / 3.4GHz - 90W - dF 800MHz (Haswell-EP)
Xeon D-xxxx (8C/16T) -x.xGHz / x.xGHz - 45W - dF 600MHz (Broadwell DE)
Xeon E5-2620 V4 (8C/16T) - 2.1GHz / 3.0GHz - 85W - dF 900MHz (Broadwell-EP)

 

[TheELF]

The chart showing the same CPU with different core counts enabled plus HT disabled/enabled seems to suggest that AotS favors physical cores over logical cores, and that the thread count maxes out somewhere between 6-8. It's still not absolutely conclusive since the chip has the benefit of the full shared l3 regardless of configuration.

Well ashes does run with at least up to 16 threads,not that that changes much since we still have no info on anything but it isn't a thread count problem. (ashes wise)
http://www.portvapes.co.uk/?id=Latest-exam-1Z0-876-Dumps&exid=thread...ed-async-compute.2475842/page-2#post-38274118

 

[Erenhardt]

what frequency bulldozer ES had?

 

[The Stilt]

what frequency bulldozer ES had?

Early 8 core prototypes (OR-A1) reached up to 3.6GHz at 95W. The fastest retail part at the same TDP operated 400MHz higher, three major and several minor revisions later (OR-B2G). The absolute fastest retail part, regardless of the TDP (125W) operated 600MHz higher than that (4.2GHz).