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

[Abwx]

If by "all FP instructions" you mean "nothing beyond scalar SSE2", then sure. Otherwise it wouldn't run on the Pentium G3258. Povray doesn't do code dispatching.

PoVray use up to AVX and FMA, this was already discussed in this forum, now we can look at professional softs ultra optimised for Intel HW and later , and the picture is the same, nowhere you ll see a 2600K matching a FX8350, even the 3770K is somewhat lagging, so CB R15 is quite an exception...

http://www.hardware.fr/articles/946-5/performances-applicatives.html

 

[majord]

I don't think AMD would do an event like Horizon to show off its new CPU before launch if the actually made a "ZEN-Dozer".

Additionally the performance discrepancy between the shown Blender/Handbrake and those leaks is just too large for the same CPU

Not really, what makes you think because 2 selected benchmarks show roughly equal throughout/clk that all others will be So close? When has this ever been the case for any AMD Vs Intel uarch (including the last time they were similar, I.e k7 Vs p6)?? .. I'll answer my own question - Never.

 

[jhu]

PoVray use up to AVX and FMA, this was already discussed in this forum,

No it does not, especially not when your provided graph points to an RC4 build from several years ago. Point to me where in the source AVX/FMA is used. Like I said, it otherwise wouldn't run on the Pentium, and Povray doesn't do code dispatching.

My point is, Povray isn't really the best thing to use to use as a CPU benchmark. It's only somewhat useful if you actually use Povray, and even then maybe not so much since the benchmark scene doesn't even utilize commonly used features.

 

[AtenRa]

Blender and Handbrake are Int or fp ???

 

[itsmydamnation]

Ahhh,so in a nutshell it just stores results for instructions that it deems might be used again?And smart cache then distributes them to where they are needed?
Or am I totally wrong again?
Could that be why zen is so fast in those benches?It finds a result once and reuses it over and over?

Wrong again but you asked

What predictors do is not keep the result but try to guess the way a branch will fall(branch) or what data might be needed soon (prefetch) depending on either what it currently observes or what happened previously. Now a neural net predictor does the same thing but uses a network of simple Perceptron's maybe with feed back mechanism to learn the codes behavior and thus hopefully getting more predictions right.

I guess you could say a neural net branch predictor is about learning behavior dynamically without a base ruleset and a traditional predictor makes predictions based of a rule set ( the rule set being the really secret sauce we never hear about).

AMD will likely have both types and some sort of weighting or confidence factor, like if the neural net is in early learning for this behavior use the traditional predictors result, as confidence grows use the neural net.

Blender and Handbrake are Int or fp ???

Blender SIMD FP
Handbrake int and SIMD int

 

[Abwx]

My point is, Povray isn't really the best thing to use to use as a CPU benchmark. It's only somewhat useful if you actually use Povray, and even then maybe not so much since the benchmark scene doesn't even utilize commonly used features.

You once said that you re more interested by Blender tests, or are you saying that Cinebench is better..?.

And what about 3DS MAX since it s much more used than Cinema 4D..?.

Blender and Handbrake are Int or fp ???

Handbrake is Integer, Blender is FP 32bit precision, Povray is FP 64 bit precision, this make those two renderers measuring different FP throughputs from a CPU.

 

[jhu]

You once said that you re more interested by Blender tests, or are you saying that Cinebench is better..?.

And what about 3DS MAX since it s much more used than Cinema 4D..?.

I'm more interested in Blender because that's what I use. I've come to distrust Povray benchmarks because the results don't match actual performance in usage. As for the commercial packages (Cinema 4D, vray, etc.) I'm not too interested in those because they don't line up with Blender results all that well either, which doesn't help when making purchasing decisions (ie looking at these Ryzen benchmarks is really interesting).

Honestly, Cinebench is only really useful to anyone who actually uses Cinema4D. For everyone else, it's just another way to slightly increase the room temperature.

 

[Dresdenboy]

Let's not put too much effort into decoding 2 rows of pixels. There could still be a smart 15 year old having fun keeping lots of forum users busy.

 

[blublub]

I find that hard to believe. If Ryzen is shipping in February, even a paper launch, they need to have started final production by now.

The other option could be that Ryzen's final silicon is still in the fab right now and will finish in the the 1st two weeks of January - this would leave enough time for validation binning and packaging for a late q1 release.

 

[BeepBeep2]

Let's not put too much effort into decoding 2 rows of pixels. There could still be a smart 15 year old having fun keeping lots of forum users busy.

Maybe I like being busy.

My guess is that we'll see more leaks leading up to launch, probably closer to launch. Until then, I've said everything I think I can say on this "leaked" Cinebench score.

 

[coercitiv]

Food for thought: a really bad piece of news for Zen would have been a good Fritz score and a bad CB score, both from the same source.

 

[AtenRa]

Blender SIMD FP
Handbrake int and SIMD int

Handbrake is Integer, Blender is FP 32bit precision, Povray is FP 64 bit precision, this make those two renderers measuring different FP throughputs from a CPU.

thx

 

[Nothingness]

Because it does not line up with two (different) workloads we already have scores on. Blender and Handbrake are completely different types of workloads and both put Zen at the same clock roughly at the same level (or slightly ahead) of Broadwell-E.

Though different, Blender and Handbrake have at least two important similarities:

1. they scale well with the number of cores
2. they both are relatively few branch instructions and lowish branch mispredictions per thousand instructions.

That doesn't mean both workloads are the same, but they certainly are not "completely different".

 

[bjt2]

Let's not put too much effort into decoding 2 rows of pixels. There could still be a smart 15 year old having fun keeping lots of forum users busy.

If we would ask AMD if this is bullshit, you think they will answer "no comment"?

 

[krumme]

Broadwell-E does exceptionally well in Cinebench R15, about 10% uplift clock-per-clock over Haswell-E.

What??? For a fp bm ! Lol. This then goes to show the information gained by CB R15 is either borked or bend to a purpose making bw shine vs prior gen. Such a ipc result to me - beeing of by aprox 300% vs the results eg AT get over a broad portfolio of apps - just points to this beeing a marketing tool for sellig the new generation of cpu.

But yeaa your post made me look at
http://images.anandtech.com/doci/9482/IPC Over Sandy.png?_ga=1.59123949.1968553451.1464761965

Cb r15 shows by far the biggest gain of all.

Unfortunately marketing can be dangerous because it the pull down the performance of your next gen skylake

http://images.anandtech.com/doci/9483/01 - Gains over Sandy.png?_ga=1.132277069.258125519.1478557456

http://www.anandtech.com/show/9483/intel-skylake-review-6700k-6600k-ddr4-ddr3-ipc-6th-generation/9

But i guess thats what a new CB R18 should adress. Lol. There is work for all; Rinse and repeat.

 

[TheELF]

What??? For a fp bm ! Lol. This then goes to show the information gained by CB R15 is either borked or bend to a purpose making bw shine vs prior gen. Such a ipc result to me - beeing of by aprox 300% vs the results eg AT get over a broad portfolio of apps - just points to this beeing a marketing tool for sellig the new generation of cpu.

Don't get too upset, povray,in the pic you posted, also has huge gains and that's one of the go to apps for amd supporters as does dolphin which is completely independent as well,far as I know neither one uses intel compilers.
Look at dolphin improvement from ivy to haswell and look at what ipc improvement all the sites state between them.
Some apps just have code that runs better on some archs,or maybe it's the opposite some archs just run some code much better.

 

[krumme]

Don't get too upset, povray,in the pic you posted, also has huge gains and that's one of the go to apps for amd supporters as does dolphin which is completely independent as well,far as I know neither one uses intel compilers.
Look at dolphin improvement from ivy to haswell and look at what ipc improvement all the sites state between them.
Some apps just have code that runs better on some archs,or maybe it's the opposite some archs just run some code much better.

You can reat asurred i am no to upset. Dont worry . I am not into what apps amd supporters support or what compiler is used.

But nearly each year i have to read about the huge ipc gains a new cpu gets. But when i compare over a span of 6 years those gains is less than a third.

In the house we use handbrake a lot and some heavy rendering. But thats often done by the gpu but its kids competence anyway lol. In general i think the focus on fpu is a tad overdone.

I am all for better branchpredictors and smart stuff but when i look at cpu today and its 4 wide design i think its kind of brutal use of ressources to get progress. We need some more itanium like thinking imo and more radical solutions. Zen is not it. And its not amd that is gone a do it but it seems to me there is a business opportunity here. To frame it positively.

 

[PhonakV30]

Hello, I'm new here. I come here to just clarify the screenshot of cinebenchR15 and Fritzchess of Zen. This forum registration makes me headache that verification code doesn't display that I can't resigter for quite a while.

Both CBR15 and fritzchess screenshot are from Chinese forum Baidu, the original thread has been deleted for unknown reason. Most likely deleted by poster. I asked my friends which read that thread before got deleted on that forum, they said poster would 'post more bench' after a little while, but then thread deleted and poster disapeear. This poster hold a shop on chinese online shop Taobao and sell rigs, this is suspicious that a little ordinary trader could have final silicon or even ES.

As you can see he has no CPU-Z screenshot, CBR15 and fritzchess score looks quite suspicious and looks seems to not from a same CPU or even same rig. Afer I talked to my friends and went over on that forum, we decided to regard these screenshot as a fake. very likely to be a fake. And that poster has reasonable intent to make a fake to help selling his toy at his shop.

You can ask some question here but I'm not sure I can give reasonable answer to you. I'm here to try my best to break language barrier between world and Chinese. Thanks a lot.

Please don't trust any thing from Baidu, there's so many troll thread and wrong information on that forum

source

 

[inf64]

Thanks behrouz. It makes sense now .

 

[TheELF]

I am all for better branchpredictors and smart stuff but when i look at cpu today and its 4 wide design i think its kind of brutal use of ressources to get progress. We need some more itanium like thinking imo and more radical solutions. Zen is not it. And its not amd that is gone a do it but it seems to me there is a business opportunity here. To frame it positively.

Changing hardware in a radical way will make all available software run very bad which means that a new technology will look very bad,that's what happened to itanium as well.
You need a new tech that is 5-6 times faster then current tech so it can run all current software at the same speed to convince people to change,this is almost impossible to do.

 

[DrMrLordX]

source

Figures. Thanks for the clarification.

 

[Gikaseixas]

We don't know if this is real Zen ES benchmarks, if they are we don't know which SKU (Could be 4c/8t) and we pretty much know Turbo is disabled on these as they aren't finished validating that part yet.

Tip for Future leakers: SHOW US CPU-Z AT LEAST!

If it looks bad for a certain company it's good enough to post and spread it as much as one can, even if the source is a website many blast as not reliable source when two other certain companies benches don't look so good either

 

[vissarix]

Something is very strange here. Amd showed us Ryzen beating an i7 6900k on blender, here we have Ryzen barely faster then an i7 7700k overclocked on cinebench and also slower on fritz chess...

Now i understand Amd trying to show Ryzen on the best light possible, probably cherry picked the best benchmarks which shows Ryzen performing better then an i7 6900k....but there is no way they lied to us so much...i mean we are talking about more then 35% performance difference from blender benchmarks to cinebench...

I dont put to much trust on these leaks, but if they are true then Zen is another Bulldozer...

 

[Veradun]

Something is very strange here. Amd showed us Ryzen beating an i7 6900k on blender, here we have Ryzen barely faster then an i7 7700k overclocked on cinebench and also slower on fritz chess...

Now i understand Amd trying to show Ryzen on the best light possible, probably cherry picked the best benchmarks which shows Ryzen performing better then an i7 6900k....but there is no way they lied to us so much...i mean we are talking about more then 35% performance difference from blender benchmarks to cinebench...

I dont put to much trust on these leaks, but if they are true then Zen is another Bulldozer...

Let me put it on perspective: can you link a pre-release benchmark posted by AMD in which Bulldozer was faster than the top of the line Intel processor?

 

[frozentundra123456]

Let me put it on perspective: can you link a pre-release benchmark posted by AMD in which Bulldozer was faster than the top of the line Intel processor?

Yes, I can actually. They showed gpu limited gaming benchmarks showing BD as fast as the top of the line Hedt intel cpu.