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

[BeepBeep2]

About half his post,is the Intel principle engineer badmouthing a competing product. He seems spooked for some reason,to go on such a massive rant on Twitter. If you look at the scores for the FX8350 in Blender,Intel CPUs absolutely destroy it.

If the AMD chip is as bad as he thinks it is,he would have no need to actually rant like that,since the benchmarks at launch would tell the full story. You never saw a similar rant like that before Bulldozer was launched,right??

I know who Francois Piednoel is, lol. He "rants" on all kinds of subjects, some unrelated to x86. He just doesn't think there is enough proof yet to prove the performance of Zen, but yeah I would imagine they are a bit worried about the competition as well.

 

[sirmo]

You know there is relatively easy way to check his theory. His theory if I understand him correctly is that Ryzen could have a poor branch predictor, and because Blender is not very branch predict difficult workload it's looking better than it normally is.

We have the benchmark, we can check and see how Bulldozer or a BD derived chip does in it. From my recollection people have posted FX results and they weren't anything to write home about. And we know BD cips have a long pipeline and not so great branch predictor. So this should be one of BD's best benchmarks.

I don't understand his point about the game loading. Rene was waiting on a game countdown to join a game, the wait wasn't CPU load based. Also would a ray tracer really not be branch prediction dependent? I would imagine ray tracer by it's pure nature has a lot of variables and different branches to take as it passes through different materials and different reflections it has to calculate.

 

[KTE]

They'll most likely under price it just to attract people like me who have felt burned by AMD's chips lately.
I'm still concerned about their partners motherboard quality, hopefully Ryzen has some serious quality behind it.

Yea despite having something solid, winning the partners and their product/execution is key for sure. All the DT, Mobile and Server market rests on them.

Working in that field, I know it's the most difficult aspect. Client relationship management. As they are a key client, you cannot upset them and must gauge your words wisely, being not just diplomatic but very lax.

When they do something wrong, it is difficult to let them know just how badly it is affecting you.

Sent from HTC 10
(Opinions are own)

 

[USER8000]

I know who Francois Piednoel is, lol. He "rants" on all kinds of subjects, some unrelated to x86. He just doesn't think there is enough proof yet to prove the performance of Zen, but yeah I would imagine they are a bit worried about the competition as well.

You know there is relatively easy way to check his theory. His theory if I understand him correctly is that Ryzen could have a poor branch predictor, and because Blender is not very branch predict difficult workload it's looking better than it normally is.

We have the benchmark, we can check and see how Bulldozer or a BD derived chip does in it. From my recollection people have posted FX results and they weren't anything to write home about. And we know BD cips have a long pipeline and not so great branch predictor. So this should be one of BD's best benchmarks.

I don't understand his point about the game loading. Rene was waiting on a game countdown to join a game, the wait wasn't CPU load based. Also would a ray tracer really not be branch prediction dependent? I would imagine ray tracer by it's pure nature has a lot of variables and different branches to take as it passes through different materials and different reflections it has to calculate.

AMD used Blender for a very good reason - the FX8350 absolutely sucks in it compared to Intel,and this is why he is talking like that. Whether or not AMD has better SMT,lower IPC,etc the fact is AMD has managed to catch up in that benchmark,with a huge per core/per thread improvement in performance.

 

[sirmo]

I just browsed some of Blender render code, and there is tons of checks and branches in it, like check_intersect_solid etc.. https://github.com/dfelinto/blender/blob/master/source/blender/render/intern/raytrace/rayobject.cpp So I am not buying his theory really.

 

[sirmo]

AMD used Blender for a very good reason - the FX8350 absolutely sucks in it compared to Intel,and this is why he is talking like that. Whether or not AMD has better SMT,lower IPC,etc the fact is AMD has managed to catch up in that benchmark,with a huge per core/per thread improvement in performance.

This is why I think he's completely off base.

 

[bjt2]

You know there is relatively easy way to check his theory. His theory if I understand him correctly is that Ryzen could have a poor branch predictor, and because Blender is not very branch predict difficult workload it's looking better than it normally is.

We have the benchmark, we can check and see how Bulldozer or a BD derived chip does in it. From my recollection people have posted FX results and they weren't anything to write home about. And we know BD cips have a long pipeline and not so great branch predictor. So this should be one of BD's best benchmarks.

I don't understand his point about the game loading. Rene was waiting on a game countdown to join a game, the wait wasn't CPU load based. Also would a ray tracer really not be branch prediction dependent? I would imagine ray tracer by it's pure nature has a lot of variables and different branches to take as it passes through different materials and different reflections it has to calculate.

And the +9% speed in handbrake? video encoding is simpler to predict?

 

[KTE]

This is why I think he's completely off base.

Running CodeAnalyzer on a preZen AMD chip would reveal the required details, and just where the older chips struggled in this benchmark.

Any volunteers?

Sent from HTC 10
(Opinions are own)

 

[USER8000]

This is why I think he's completely off base.

I had a look on Overclockers UK,and a 4.3GHZ Core i7 4770K was doing the AMD Zen render in half the time a 4.5GHZ FX8350 was doing it in.

So you would need a 16 core AMD Piledriver CPU to get a similar time to a Core i7 4770K if both were at similar clockspeeds.

That would mean AMD has easily doubled performance PER THREAD over the FX8350 in this test.

Now,nobody is saying AMD has doubled IPC,but that hints at Ryzen having a much more even performance profile.

 

[KTE]

About half his post,is the Intel principle engineer badmouthing a competing product.

The Intel guy, I don't agree with as I stated (he had some AMD gurus replying to him!).

If you read, I debunked his fear mongering so that it couldn't be used as a source.

Sent from HTC 10
(Opinions are own)

 

[Ajay]

They'll most likely under price it just to attract people like me who have felt burned by AMD's chips lately.
I'm still concerned about their partners motherboard quality, hopefully Ryzen has some serious quality behind it.

I think AMD will have some solid partners behind them for AM4 motherboards - so long as that basic design & chipset is solid. Poor chipsets and motherboard quality has hurt AMD in the past with prospective IT buyers that I knew.

 

[sirmo]

It will be interesting to see. But I did not quite expect Ryzen to have a 19 stage pipeline. I am quite confident on the clocks of this thing now. Just based on what we know about Excavator pipeline on 28nm planar and the fmax it could reach there. I think just shrinking Excavator to 14nm FF would yield a 25% clock boost, based on what we've seen from GCN. And Ryzen with its long pipeline could reach those high clocks as well. If they really got a great branch predictor in it, this is going to be one helluva competitive chip. Very excited to see the final benchmarks with Boost in the play.

Maybe we now know a reason for the Skylake/Kaby Lake respin.

 

[USER8000]

The Intel guy, I don't agree with as I stated (he had some AMD gurus replying to him!).

If you read, I debunked his fear mongering so that it couldn't be used as a source.

Sent from HTC 10
(Opinions are own)

Fair enough - I will be personally surprised if AMD gets to Broadwell level IPC,but if they could even get to Haswell level and improve performance/watt noticeably over what they had before,that would still do them in good stead overall.

 

[Ajay]

The Intel guy, I don't agree with as I stated (he had some AMD gurus replying to him!).

If you read, I debunked his fear mongering so that it couldn't be used as a source.

Sent from HTC 10
(Opinions are own)

Thanks! That clears things up. I thought you were siding with Francois - who seems to have gone crazy.

 

[Ajay]

It will be interesting to see. But I did not quite expect Ryzen to have a 19 stage pipeline. I am quite confident on the clocks of this thing now. Just based on what we know about Excavator pipeline on 28nm planar and the fmax it could reach there. I think just shrinking Excavator to 14nm FF would yield a 25% clock boost, based on what we've seen from GCN. And Ryzen with its long pipeline could reach those high clocks as well. If they really got a great branch predictor in it, this is going to be one helluva competitive chip. Very excited to see the final benchmarks with Boost in the play.

Maybe we now know a reason for the Skylake/Kaby Lake respin.

A point of interest is that the branch predictor seems to have some capacity for learning - hence it should do better with frequently used programs. This should be especially true for design professionals who rely mainly on a small set of development tools. Oh, and servers, which often run just one main application (at least in large data centers).

 

[MajinCry]

Running CodeAnalyzer on a preZen AMD chip would reveal the required details, and just where the older chips struggled in this benchmark.

Any volunteers?

Sent from HTC 10
(Opinions are own)

Ehm, this thing right? http://www.codeanalyzer.teel.ws/

Only gives a summary of the files themselves, not the code.

 

[sirmo]

A point of interest is that the branch predictor seems to have some capacity for learning - hence it should do better with frequently used programs. This should be especially true for design professionals who rely mainly on a small set of development tools. Oh, and servers, which often run just one main application (at least in large data centers).

Aye, Lisa spent quite the moment there to praise it. She even went as far to say that it's responsible for 1/4th of the IPC gain we're seeing. Now it is a CEO praising its own product which is to be taken with a grain of salt, but she really doesn't strike me as a CEO who overhypes things, at least she's much better than the AMD's past leadership.

 

[zinfamous]

some guys live in the dream?
I wont buy their product with high price. no way.
a company with fail product? A company is gone but the earth is still turning.

this is the reason:
1. for now all of benchmark, I still thinking Intel better(if drop the pirce should be nice)
2.BMW vs Ford, anyone agree if Ford have high price than BMW? Brand value different.
if any company just push garbage few years and asking customer pay for it. did you accept?

you guys really crazy fans, I was an AMD fans with lot of AMD CPU.
now...go hell...the damn price with mouthful of lies.
I just bought RX 480 with fully regret, bye bye Vega no more again.
Terrible Watt kill the polar bear and pls follow PCI-SIG.
you cant design something without criterion.

the only good is the driver, no more crap CCC no more blue screen, nothing more.
pascal driver really shit this year.

damn, you killed a polar bear and blame AMD?

I mean, wat?

 

[Ajay]

Aye, Lisa spent quite the moment there to praise it. She even went as far to say that it's responsible for 1/4th of the IPC gain we're seeing. Now it is a CEO praising its own product which is to be taken with a grain of salt, but she really doesn't strike me as a CEO who overhypes things, at least she's much better than the AMD's past leadership.

Agree. As I see it, AMD is making use of extra xtors in new multi billion xtor CPUs to add intelligence and flexibility at appropriate points in the architecture to improve performance. A particular point is to be more flexible so that code optimized for Intel will run well on AMD also ( a long time sticking point for AMD). I think AMD is making some smart moves. Now, Intel isn't going to sit still, but it's looking like AMD has created a new start for themselves. We'll know much better once Ryzen is released. 2017 is the first time in something like 12 years that I'm thinking of building an AMD system (I was going to build Intel). I'll have all the proof I need soon enough.

 

[zinfamous]

I know who Francois Piednoel is, lol. He "rants" on all kinds of subjects, some unrelated to x86. He just doesn't think there is enough proof yet to prove the performance of Zen, but yeah I would imagine they are a bit worried about the competition as well.

agreed, but that doesn't justify monopolizing 80% of thread space with obnoxious twitter images which can be summarized in 6 words. No one wants to read that. It is "gamergate-level" posting. Immediate turn-off.

 

[sirmo]

Here is a bit more about Ryzen's "neural net" branch predictor:

It is interesting that Samsung has broken ranks by publicly declaring that its prediction engine is a neural network. David Kanter, Linley Group's microprocessor analyst, told us today's state-of-the-art branch prediction systems are based on neural-network-like designs: for example, AMD's Jaguar and Bobcat predictors use similar technology. Burgess, now a veep at Samsung, was formerly chief architect of AMD's Bobcat microarchitecture.

AMD's Zen architect Mike Clark confirmed to us his microarchitecture uses a hashed perceptron system in its branch prediction. "Maybe I should have called it a neural net," he added.

http://www.theregister.co.uk/2016/08/22/samsung_m1_core/

So it seems that the idea for it came from Jaguar/Bobcat cores.. which from my recollection were pretty decent chips for their size/power footprint.

 

[zinfamous]

The Intel guy, I don't agree with as I stated (he had some AMD gurus replying to him!).

If you read, I debunked his fear mongering so that it couldn't be used as a source.

Sent from HTC 10
(Opinions are own)

Fair enough--the other issue is that the start of that post is borderline/primarily personal-attacky. There is basically zero tolerance of that in these hardware forums, unlike some other AT forums.

...ask me how I know.

 

[Phynaz]

In the server market, yes. They simply couldn't make enough to reach 50+% overall. That shouldn't be a problem this time, so long as the demand is there.

Retail is 9% of the total PC market.

 

[Phynaz]

AMD stock price will allow them to get more in loans. AMD has cut huge amounts of their debt, and are prime for investment. AMD could very easily offset a reduction in margins and use the investments to supplement things like R&D. AMD could likely sell Zen for $600-$700 at the top end and still make very good margins. If AMD can gain enough market share, then they can leverage that for further investment to supplement cash.

You are thinking in small business terms, and not in corporate terms. If AMD goes after the cash pop, then that means higher prices. Higher prices directly reduce consumption at the higher price point vs a lower price point. That limits market capture which allows Intel to come back if they so choose.

What is important for AMD is to gain market share and lock people into their system. If AMD can get a win in the server market and the next gen is equal or slightly behind Intel, that is a win. That is because the cost of moving platforms is expensive and is only worth while if the cost of moving is worth the benefit. If Intel comes out with its next gen and its only a few % faster than AMDs next gen, then many still upgrade to AMD because that means less hardware to buy. Big companies may have the capital to move, but many will not.

All of this is completely wrong.

Stock price has nothing to do with the ability to get loans.

And for the last time, AMD isn't going for a market share grab at the cost of profits. They have said it over and over, I don't know why you don't believe them.

 

[KTE]

Ehm, this thing right? http://www.codeanalyzer.teel.ws/

Only gives a summary of the files themselves, not the code.

Sorry mate. 'Code Analyst' is correct.

It's this: http://developer.amd.com/tools-and-sdks/archive/compute/amd-codeanalyst-performance-analyzer/

Sent from HTC 10
(Opinions are own)