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

[swilli89]

How sweepr is allowed to create threads on products we know he has a focused biased against is beyond me. That's why these mega threads should be created by a moderator.

The fact that the title was updated to include WCCF Forum sourced rumors from some completely unknown Asian source made to look just as reputable to the published AMD demoes is laughable. Sucks the conversation was forced to take such a negative slant with the inclusion of those remarks.


First, sweepr can create threads on any subject that applies to the forum he is posting in. Also, mods may create threads, but its as a poster. Mods "moderate" the users. This is your ONLY warning on this subject.
Markfw
Anandtech Moderator.

 

[Sweepr]

How sweepr is allowed to create threads on products we know he has a focused biased against is beyond me.

You do realize that all major forums and websites are commenting on this, right? If it wasn't me, obviously someone else would post. But of course you're blind to this fact and would rather shoot the messenger, maybe you should ignore every single rumor that doesn't match your expectations. Take it easy, we don't even know if it's not a fake (yet) - though the leaker says he will post more in the coming days.

 

[KTE]

On topic: Quite honestly, this leak doesn't have any data whereby one can ascertain its credibility or verification. You can't ascertain the latter with AMDs leak either, but you can know the first.

I am telling you guys since the beginning to not get attached to early leaks and get ready for BS from AMD and the leakers. Whether it is a good or bad product. But still users here are pontifying unbelievable miracles and data that does'nt add up when they suit them.

Off topic: All this crazy fanboyism over one leak. You add zero value.

If Sweepr is biased then so are those attacking him, and they are the same ones making crazy Zen hypes in this thread.

If you have a problem with the leak, whether that by AMD or by leakers, attack the data. Attacking the person shows you as a loony fanboy with a butthurt kneejerk reaction. That's when no one takes you seriously anymore.

Sent from HTC 10
(Opinions are own)

 

[Dresdenboy]

Here link
Now compare it to Blender score, Maybe you find some correlation.I can't find a good database for blender

Thx. I took the Reddit Blender score collection (with one confirmed RyZen score for 100 samples (25s) and one extrapolated by me for 200 samples (53s -> 2.1x as suggested in the table).
https://docs.google.com/spreadsheets/d/1O2Or6XOETZr3a4gLAuCuYEx8m5X_lv23LdwZiCZpBcM/edit#gid=0
Then I matched Blender scores with your CB scores based on CPU names alone (no checking for turbo).

So far, the Zen scores (yellow dots) don't lie on the regression line, but don't look like outliers either.

 

[BeepBeep2]

While I can't speculate about the CB R15 cropped screen score or its legitimacy or what CPU, the ES naming convention on quick glance seems to fit the naming convention (in terms of length) for the leaked Geekbench 32 core Zeppelin chip. There isn't enough there to be able to tell what any of the characters are though.

 

[rainy]

Take it easy, we don't even know if it's not a fake (yet) - though the leaker says he will post more in the coming days.

I didn't know that AMD is already shipping Zen.
Even if those results aren't faked, that guy could have only an ES.

 

[SAAA]

I really hope the next time someone leaks there's at least a single bench score like blender with the same setting of AMD event and a similar score, just to give a little more credibility. I wouldn't be surprised by a new architecture having some high and low showings depending on the bench/drivers etc. So better than broadwell in some workloads? Sure. Same as sandy in others? That's possible too.

Heck if you really have a sample and can run those tests all it takes is 10min to do everything and post passable infos, including clocks, features (turbo) in play and so on.

On a aftertought perhaps these lower scores are related to that bug someone else spoke of, or maybe it's all a lie, maybe not.
For sure Intel didn't react much to Zen till now and I don't think they've got 0 clues on its performance, so if Kaby is all they prepared as a response (including prices) it doesn't bode well for Zen.
With this I don't mean 8 cores won't be better than 4 but it's starting to look more like they won't compete well except on multithreaded loads. Saving grace is most software is starting to use more cores today so the difference should turn in Zen's favor over a few years.

 

[itsmydamnation]

They also said that it will work together with their "neuro network" and that that will have to learn to give the best improvements...maybe they will have a driver for the CPU in the future with better performance,they already said that oc will only be possible through windows so maybe it will be a package.

NO,NO,NO,NO,NO (ok a little carried away here)

AMD have had neural net predictors since Bobcat and piledriver. What happens when you have two parts of your core both using the neural net methodology for learning and you start sharing result to each other via a bus, thats right you now have a "network".

When they talk about learning behavior they are not talking over long lengths of time but over times of 100k cycle , 1M cycles, 10M cycles maybe even shorter. There will be a structure that can only store so many observed behavioral results, thus it will have to forget results (oldest least used or something like that).

 

[The Stilt]

Even if those results aren't faked, that guy could have only an ES.

Of course the CPUs out there are ES. Even AMD used ES CPU for the demos (both press and "New Horizon").

 

[sirmo]

They also said that it will work together with their "neuro network" and that that will have to learn to give the best improvements...maybe they will have a driver for the CPU in the future with better performance,they already said that oc will only be possible through windows so maybe it will be a package.

It's not AI it's neural networks or statistics. It doesn't learn over days or months, it learns over nanoseconds.

 

[rainy]

Of course the CPUs out there are ES. Even AMD used ES CPU for the demos (both press and "New Horizon").

What if he have 2.8 GHz ES when AMD demonstrated 3.4 GHz sample on "New Horizon"?
That's over 21 percent clock difference.

 

[The Stilt]

What if he have 2.8 GHz ES when AMD demonstrated 3.4 GHz sample on "New Horizon"?
That's over 21 percent clock difference.

The lowest clocked 8C/16T ES variants (known) had 2.8GHz base and 3.05GHz MACB, meaning < 11.5% difference. Regardless, a stock 6900K should sustain 3.5GHz in Cinebench as it is purely a scalar workload.

 

[Phynaz]

How sweepr is allowed to create threads on products we know he has a focused biased against is beyond me. That's why these mega threads should be created by a moderator.

Yes, only people you agree with should be allowed to post here

 

[coercitiv]

When they talk about learning behavior they are not talking over long lengths of time but over times of 100k cycle , 1M cycles, 10M cycles maybe even shorter. There will be a structure that can only store so many observed behavioral results, thus it will have to forget results (oldest least used or something like that).

In this case a picture says more than 100 marketing slides.

When we hear "neuro" for a CPU we are tempted to think of this

when in fact, "neuro" in CPU (reaction) time looks more like this

It's more like well tuned reflex mechanisms than complex learning/adapting behavior.

 

[superstition]

If Sweepr is biased then so are those attacking him

tu quoque fallacy

and they are the same ones making crazy Zen hypes in this thread.

another fallacy. (If X does something bad then Y is blameless for doing something bad.)

 

[lolfail9001]

So, you guys are telling me these Cinebench/Fritz scores came from Chinese forums? Well, they are probably correct then.

Oh and btw, 1.7Ghz rx480 rumor was entirely western, Orients had it correct pointing at 390x performance land.

 

[Glo.]

Off topic: All this crazy fanboyism over one leak. You add zero value.

If Sweepr is biased then so are those attacking him, and they are the same ones making crazy Zen hypes in this thread.

If you have a problem with the leak, whether that by AMD or by leakers, attack the data. Attacking the person shows you as a loony fanboy with a butthurt kneejerk reaction. That's when no one takes you seriously anymore.

Right, so AMD demoed Zen, at 3.4 GHz few days ago, matching and outperforming i7 6900K in: transcoding, rendering, gaming, and streaming.

Today: some anonymous poster from china posted score from anonymous CPU, that has supposedly 16 threads. And everybody jumped to conclusion that THAT is the TRUTH about Zen. And that is where Sweeper is making a mistake.

Something here is not right, my friend. Sweeper jumped to conclusion that it MUST be Zen, despite all the data that was provided in New Horizon stream. Yet, we have no f******* proof that it factually IS Summit Ridge 8C/16T.

Start thinking. New Horizon was not only about Blender, and Handbrake. It was also few other things, that people completely miss(Gaming!). Its also not a matter of attacking the messenger, which I already have said in this thread. Attack the message.

So, you guys are telling me these Cinebench/Fritz scores came from Chinese forums? Well, they are probably correct then.

Oh and btw, 1.7Ghz rx480 rumor was entirely western, Orients had it correct pointing at 390x performance land.

I thought it was supposed to be around GTX 1070 at that clock speed.

 

[TheELF]

NO,NO,NO,NO,NO (ok a little carried away here)

AMD have had neural net predictors since Bobcat and piledriver. What happens when you have two parts of your core both using the neural net methodology for learning and you start sharing result to each other via a bus, thats right you now have a "network".

When they talk about learning behavior they are not talking over long lengths of time but over times of 100k cycle , 1M cycles, 10M cycles maybe even shorter. There will be a structure that can only store so many observed behavioral results, thus it will have to forget results (oldest least used or something like that).

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?

 

[krumme]

Amd have pointed out they target the cloud server market.
I asume such a market will imply code loads that varies and will expose the branch predictor?
What is important on the cloud server market in ralation to cpu perf?
You dont get a cpu with good branch predictor just because you want it. But neither do you get fpu perf in blender similar to bwe. If anything 3 months ago i would rate the latter as more unlikely. Good perf on fpu heavy stuff is needed but certainly not to the degree we see now when your competitor have such fpu monsters and is expected to take the market where fpu plays the major part.

 

[sm625]

These people leaking benchmarks need to run the blender demo and confirm that their system is scoring 36 seconds. This must be done before running any other benchmarks. You would think this would be common sense at this point, no?

 

[inf64]

Why not?

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. Now you have 2 dubious benchmarks both showing Ryzen at unknown clock being 30% (in both cases! red flag) slower than Broadwell-E.

So either the leak is a fake or Ryzen in the leak was running at way below the Broadwell-E MT boost clocks(3.4-3.5Ghz on 6900K). This is of course my opinion on the matter.

 

[CatMerc]

So, you guys are telling me these Cinebench/Fritz scores came from Chinese forums? Well, they are probably correct then.

Oh and btw, 1.7Ghz rx480 rumor was entirely western, Orients had it correct pointing at 390x performance land.

I distinctly remember quite a few Chinese "leaks" that turned out to be bollocks in the RX 480 era. The clockspeed leak was Western in origin, but the R9 Nano perf at stock clocks that was touted by WCCFTech was oriental.

 

[krumme]

Well the problem is all is taking as this is a "leak" yet its nothing.
Its a leak when we get more info that can comfirm its an es cpu.
Its a problem this clickbait is presented wrapped with tons of graphs to the degree everyone except a few is accepting it as a leak.

If this clickbait was presented as such it would give a better discussion.

 

[Markfw]

Well the problem is all is taking as this is a "leak" yet its nothing.
Its a leak when we get more info that can comfirm its an es cpu.
Its a problem this clickbait is presented wrapped with tons of graphs to the degree everyone except a few is accepting it as a leak.

If this clickbait was presented as such it would give a better discussion.

I am posting as a moderator, when I say, this is constructive input, however, I don't like the term "clickbait"
Markfw
Anandtech Moderator.

 

[Sweepr]

From another forum, Fritz Chess scores for Xeon E5-2670 - 8C/16T Sandy Bridge-EP @ 3.0 GHz in MT loads:

Core i7-7700K (Stock)
35.52
Core i7-7700K (OC)
41.44
Xeon E5-2670 (Stock):
41.88
Core i7-6900K (Stock)
47.80
Core i7-6950X (Stock)
51.50
AMD RYZEN (8/16) - Rumor
36.86