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

[RussianSensation]

DX12 Math Dispatcher => AVX2(256-bit Integer) or AVX(128-bit Integer).

It is also why the dual-core and quad-core part is about as fast as the 8-core Piledriver with 2 128-bit packed integer units per module.

How many DX12 slated to come out in the next 5 years will use AVX/2?

As I said above, what if an 8-core Zen is priced at $350-$545, and can overclock from 3.2 to 4.2Ghz? All of a sudden the $1089 6900K doesn't look so hot, does it? This forum sometimes compares products as if prices don't matter. It's ridiculous like last generation the usual suspects would recommend $550 980 over a $280 290X as if price was 100% irrelevant. Things don't work like that in the real world. We also knew from Lisa's statements that Zen was not a 1 shot hit wonder. It was a start of a new CPU architecture that would be improved upon over time.

Of course it's a legit negative if Zen bombs in AVX/2 but it doesn't seem as if many programs take advantage of these instructions yet.

 

[bystander36]

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.

Was there not some Fallout 4 results which showed insane increases on Skylake earlier this year? Even going to Haswell, you can see massive differences on that game:

The tricky part, is finding an official CPU benchmark that is using high enough settings to not bottleneck the high end that includes the 6700K and 2500K.

 

[NostaSeronx]

How many DX12 slated to come out in the next 5 years will use AVX/2?

It is automatic. If you have the latest DirectX Redist. It will select the best codepath for your processor.

SIMD C++ => AVX2 if you have AVX2, AVX if you have AVX, FMA4 or FMA3 if you have FMA4 or FMA3. etc.

 

[eternalone]

Still waiting on some non ES Zens benchmarks from multiple sources until then we cant be sure, that being said price/performance is my bottom line and no matter what I hope they dont price these at $350-699.

 

[Carfax83]

DX12 Math Dispatcher => AVX2(256-bit Integer) or AVX(128-bit Integer).

It is also why the dual-core and quad-core part is about as fast as the 8-core Piledriver with 2 128-bit packed integer units per module.

This is a great point. I remember I made a thread a while back about AVX2 instructions being used in games, and someone (might have been you) said that AVX2 was already being used in the DX12 math.

More DX12, the better

Though I have a question. Does this only apply to DX12 games? DX11 games can use the latest DirectX compilers as well right, so they should also benefit as well.

Also do you know if Vulkan's compiler uses advanced instruction sets like AVX2? If it doesn't, then that could give DX12 an edge over Vulkan in performance.

 

[majord]

Interesting. Buy why is the i5 6400 so slow compared to 6700 in that graph? a 43% higher clockspeed + SMT is giving the i7 6700 a 122% lead . If throughput of AVX execution is the key, the i5 and i7 have the same resources

 

[Sweepr]

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.

The CPU core scaling test was done at 1080p 'Crazy' while OP benchmarks are 1080p 'High'.

 

[NTMBK]

Urgh, I really hope that this doesn't turn out to be another Phenom- an interesting architecture, crippled by clock speeds. Maybe a "Zen+" in a year's time with higher clocks will come along, but I won't hold my breath.

 

[Insert_Nickname]

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.

Well said, well said. I could certainly find a few uses for an 8C/16T CPU for $350*. I really wasn't expecting more then SB level per-core performance (maybe IB on a good day) from Zen anyway. Which is more then enough for what I plan to do with it when it has 8 physical cores to play with.

*Can hope, can't we?

 

[itsmydamnation]

This is a great point. I remember I made a thread a while back about AVX2 instructions being used in games, and someone (might have been you) said that AVX2 was already being used in the DX12 math.

More DX12, the better

Though I have a question. Does this only apply to DX12 games? DX11 games can use the latest DirectX compilers as well right, so they should also benefit as well.

Also do you know if Vulkan's compiler uses advanced instruction sets like AVX2? If it doesn't, then that could give DX12 an edge over Vulkan in performance.

Its a minor improvement, AVX2 over AVX isn't the issue for Zen its the 256bit ops vs >=haswell, but it is just extra width, you can get the same "extra width" for more cores ( with some additional overhead) . Also remember he is just talking about DX12 infrastructure not games themselves. So long as intel doesn't have AVX end to end in its stack and so long as consoles are at 128bit, 256bit vectors aren't going to see usage in games.

Urgh, I really hope that this doesn't turn out to be another Phenom- an interesting architecture, crippled by clock speeds. Maybe a "Zen+" in a year's time with higher clocks will come along, but I won't hold my breath.

For consumer it might very well be, Zen is a server part, 24 to 32 cores are its target and 2-3ghz clock is perfect for that. The question for the consumer part, is it clock limited or its it TDP/clock limited, im more then happy to run at what ever power consumption if it means i can push clocks. There is one benchmark.........wait now there are actually two benchmarks i am waiting for.

1. spec int 403.gcc, this is the benchmark to determine "IPC" increase. The apps people have been using in Zen threads to guess or determine IPC increases have been pretty poor. 403.gcc is repeatable, hard for compilers to cheat/optimize and branchy, thats exactly where your CPU's ability to predict, prefetch and recover will be tested.
2. geekbench 4 ,dev asked for input on RWT and everyone including Linus (very outspoken on benchmarks particularly geekbench 3..lol) was impressed with what they have cooking.

 

[cytg111]

Am I the only one who thinks that it doesnt look half bad? I can only remember one incident in history where ES performed better than final silicon and that was haswell (overclocked better).. other than that clocks and features have always been gimped. How old is this ES? Could be a year old for all we know.
The only thing I take away from this is that Zen MOST likely wont be worse than this. Thats a good thing imo. "This.Is.Then" <- Thats almost threadcrapping. Just kidding .

...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....

- 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.

 

[Erenhardt]

Is this really Zen? How does it compare to other amd CPUs?

 

[scannall]

I wish there were more information here. Lots of bandwagon jumping, without knowing some really relevant things. Which rev ES? An early BIOS? A late one? My own predictions have been pretty modest, somewhere around IB. Give me 8 core, 16 threads at a reasonable price with around IB IPC then I'm fine with it. I think the more exciting product next year will be the Zen based APU's though.

 

[know of fence]

I don't think people are seeing just how bad this is.

When I look at those two screenshots I am seeing the 4670K being more than 60% faster than Zen! What am I not seeing? This is such an epic disaster I'd be dropping 6 figures on AMD puts right now if I didnt think the market was totally broken. AMD stock should be down 80% tomorrow, because they are donezo. But being in the twilight zone that we are, I wouldnt be surprised if it went up 20% just for kicks. Swiss Central Bank > fundermentals.

You've asked: "What am I not seeing?" You did present very comparable benchmarks with the same RX 480, Standard[1080p], Version 1.30.21168.0. And I hope this was a genuine question, before you went on to jump to quantify your conclusions.

When confronted with such a spay of data, the mistake/oversight that both OP and you fell for is to cherry pick single results. With a reliable benchmark this would be fine, after all it is direct comparisons we are after. But you need to make sure the data is consistent first. I deleted the Intel screenshot from your quote, instead let's look at another screeny WCcfTech hosts. Same thing - different score. Which actually lines up with the other results for "high" and "extreme" CPU-framerate. Which is how WCCFtech arrived of Zen ahead of the Haswell i5@3.4 GHz behind the Devils Canyon i7@3.6.

Typically websites would run the benchmarks several times and either throw out or average in such outliers.

To be fair ffleader1 already linked this screenshot on the previous page, they are the same game version just 3 days apart it seems according to the date. The rest of the Zen ES benches are also from the 9th of August.

 

[podspi]

It isn't really possible to draw any conclusions from one data point.

While I am sure most of us would have preferred an early leak that implied greater performance, even if we had it we couldn't draw any conclusions on a single, incomplete, probably noisy data point. We:

1) Do not know if this is final silicon or not.
2) Do not know if the clock speed is reported accurately.
3) Do not know the rest of the system configuration.
4) Do not know the motivations of the person who posted the benchmark.
5) Do not know the power usage of the system.
6) Have to take into account benchmarks for this game are all over the place - even for similar system configurations.


Hopefully this is just the first of many leaks, or you know, maybe AMD will paper launch tomorrow. That would be even better, but I doubt it

 

[cytg111]

Hopefully this is just the first of many leaks, or you know, maybe AMD will paper launch tomorrow. That would be even better, but I doubt it

.. Yes! It is leak time, finally something is happening again.

 

[frozentundra123456]

You've asked: "What am I not seeing?" You did present very comparable benchmarks with the same RX 480, Standard[1080p], Version 1.30.21168.0. And I hope this was a genuine question, before you went on to jump to quantify your conclusions.

When confronted with such a spay of data, the mistake/oversight that both OP and you fell for is to cherry pick single results. With a reliable benchmark this would be fine, after all it is direct comparisons we are after. But you need to make sure the data is consistent first. I deleted the Intel screenshot from your quote, instead let's look at another screeny WCcfTech hosts. Same thing - different score. Which actually lines up with the other results for "high" and "extreme" CPU-framerate. Which is how WCCFtech arrived of Zen ahead of the Haswell i5@3.4 GHz behind the Devils Canyon i7@3.6.

Typically websites would run the benchmarks several times and either throw out or average in such outliers.

This is also the source that said Skylake was going to be a Conroe like jump in performance. I agree, and said so in another post, I would wait for other benchmarks before making any conclusions. I view AoTS as an extremely unreliable benchmark, although that hasnt stopped it from being used extensively in other threads when it fit the poster's agenda.

Be that as it may, even if one accepts the WCCF tech results, I disagree with their conclusion. At first glance, yes, it does not seem too bad. But then stop to think about it. Performance in a DX12 game, which is supposed to utilize multi-cores well, is only slightly above a 2 generation old i5, and below a 2 generation old i7. 8 cores vs 4, in a game that is supposed to love cores, and it still does not dominate. And their projections of 4ghz seem highly "optimistic", especially at anything close to 95 watts.

 

[Nothingness]

1. spec int 403.gcc, this is the benchmark to determine "IPC" increase. The apps people have been using in Zen threads to guess or determine IPC increases have been pretty poor. 403.gcc is repeatable, hard for compilers to cheat/optimize and branchy, thats exactly where your CPU's ability to predict, prefetch and recover will be tested.

Agreed but we'll still need to get results using the same compiler on both Intel and AMD machines, and not icc.

And an i4790 is faster at 403.gcc when compiled with gcc 5.2 vs icc, so that couldn't be considered as a problem by anyone

2. geekbench 4 ,dev asked for input on RWT and everyone including Linus (very outspoken on benchmarks particularly geekbench 3..lol) was impressed with what they have cooking.

Yes, it definitely is much better. Better choice of benchmarks, datasets identical for mobile and desktop.

 

[guskline]

I was going to hold off for Zen as a third build but came upon a great deal on a 6700k and Asus mb to couple with the new RX 480 I just snagged.
From the sound of this thread I think I made a good move.

 

[Killrose]

What about the chipset/motherboard performance for Zen? These benches could be gimped by poor performance of the chipset/motherboard.

 

[know of fence]

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.

 

[swilli89]

I don't understand why everyone is worried. What are the system configs ? what video cards are used ? Why are there no CPU benchmarks ?

This appears to be a very heavily tweaked bunch of stats to make a product look bad. I used to work in statistics, and you can make anything look good or bad if you design the test badly.

When a real site benchmarks it, talk to me, until then this is just flame fodder.

Agreed with the above post. I want to see the actual game at actual settings people would play it at. What if one CPU from Company A gets you 95 FPS and one CPU from Company B gets you 88 FPS; CPU from Company A costs $350 and cpu from Company B costs $250, what would you purchase? I know personally that since there is zero tangible difference between the two I'd rather save the cash and get the exact same experience.

 

[Boze]

Are these posts a joke?

Why would any of you ever think that AMD was going to be the competitive gaming chip? Its been what? 10? 11? years since that was the case. You didn't honestly believe Jim Keller alone could come in and make a gaming chip that would lay low the mighty Intel with its R&D budget that's what, 900% more than AMD's?

I guess hope really does spring eternal for AMD fans.

The only saving grace this chip can claim is it might be a contender for mid-range workstations. A $300 price point with 75-80% of the i7-6900K performance would be a huge win for AMD.

 

[Arachnotronic]

What about the chipset/motherboard performance for Zen? These benches could be gimped by poor performance of the chipset/motherboard.

Chipset hasn't influenced performance in a long time since all of the performance impacting bits are integrated on the CPU.

 

[Phynaz]

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.

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