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

[PPB]

No.
It has been compiled with Intel Compiler, which according to some automatically makes it to favor Intel. Intel compiler automatically adds a dispatcher to each and every binary or library it produces, however Cinebench has not been compiled with settings which would generate a different code path for Intel CPUs. Because of that, despite the dispatcher is present in the binary removing it makes no difference. It can be tested by either modifying few instructions found in the binary, or running it in VM and spoofing the CPU information (vendor, model, etc).

Regardless, based on my own tests the microarchitectures which receive the largest gain in FP code from using Intel compiler (compared to MSVC or GCC) are AMD microarchitectures. Their gains are generally significantly larger than the gains for Intel's recent µarchs.

When comparing the last four (PD, SR, XV and Zen) µarchs against Haswell or newer, Cinebenches are highly favorable workloads for AMD (especially for Zen). All of the current versions are legacy scalar workloads (up to SSE3 in R15) so newer Intels can never utilize their larger resources. Despite that Cinebench represents pretty well the CPUs capabilities in the current consumer workloads.

Cinema 4D R15 on which Cinebench R15 is based on supports Embree, however Cinebench R15 uses Maxon's own rendered.
Cinebench R18 should be out shortly, so it will be interesting to see if they continue using their own renderer or swap to Embree.

They dont use it for the benchmark because Embree Raycaster should remain a ON/OFF setting on most renderers since:

- It uses AVX and not every processor out there still supports it.
- It uses FP32 which can lead to image artifacts, specially on complex scenes, compared to FP64 which is the standard here.

On Vray it is a setting because of this, even if it is ON by default.

Cinebench is a benchmark for people who dont own the software to know how good can they setup perform on a yet to buy software. I would think it should have the option too to use Embree, but it shouldnt be the only Raycaster available because of what I explained.

Vray devs estimate an up to 30% performance uplift from switching to the Embree Raycaster. Considering the precision loss and possibility of arctifacts in the framebuffer, it is nothing extraordinary per se, in Vray it can quite help using Brute Force/Quasi Monte Carlo as both main and secondary GI.

 

[SlickR12345]

People were saying to sell AMD stick when it was low and before they released Pascal and without considering Vega and Zen either. I think up till now and even right now its a good purchase opportunity, AMD is even now so low, that there is huge room to grow.

If you bought their stock at like $5 and you bought quite a few of it about a year or so ago then you are golden now. Even today I'd say its worth buying AMD stock, in facts lots of it.

 

[lobz]

Uhm, from what i see 4.8Ghz 3930k does only 163. How come?

obviously he's a magician

 

[lobz]

People were saying to sell AMD stick when it was low and before they released Pascal and without considering Vega and Zen either. I think up till now and even right now its a good purchase opportunity, AMD is even now so low, that there is huge room to grow.

If you bought their stock at like $5 and you bought quite a few of it about a year or so ago then you are golden now. Even today I'd say its worth buying AMD stock, in facts lots of it.

I'd certainly buy an AMD stick

 

[inf64]

People were saying to sell AMD stick when it was low and before they released Pascal and without considering Vega and Zen either. I think up till now and even right now its a good purchase opportunity, AMD is even now so low, that there is huge room to grow.

If you bought their stock at like $5 and you bought quite a few of it about a year or so ago then you are golden now. Even today I'd say its worth buying AMD stock, in facts lots of it.

If I had bought Apple stock when it was worth nothing 30 years ago I would be owning a chip company right now

edit: meant Apple, wrote IBM

 

[KTE]

Back to Zen.
Has this been posted before?
https://www.nordichardware.se/nyheter/ryzen-presterar-battre-an-vantat.html

OT: Is Andreas still around?

Sent from HTC 10
(Opinions are own)

 

[.vodka]

just did 3 runs in a row with my OC i7 2600k at 4.8ghz and got 173 first time and then two times 172...thats less then 1% margin of error...

You sure about that?

4.5GHz - 156 ST
4.8GHz - 166 ST
5GHz - 172 ST

 

[hotstocks]

Turbo is kind of ridiculous. Just find out what your max overclock is stable and peg it at that. I have all 4 cores at 4.7ghz stable 24/7 for 6 years. I don't run it at 3.8ghz and turbo to 4.7ghz. I really could care less if one or two of my cores could go to 4.9ghz, the point is leave them all at the same speed and stable max. Leave it pegged at its max stable speed, or do you really care about $3.09 in electricity a year when you are buying a $2000 PC.

P.S. Ok, I get turbo. It is for LAPTOPS on BATTERY. So the processor can be at 2.8ghz or lower when idling to save battery and then turbo higher when gaming, but the turbos are then limited much lower anyways so it defeats the purpose unless you are just going to web browse or watch a movie. I have a Kaby Lake i7-7700HQ laptop and set it to performance. It never runs at the stock 2.8Ghz, hell the lowest I see it is at 3.1GHz and it is usually at 3.5GHz

 

[inf64]

OT: Is Andreas still around?

Sent from HTC 10
(Opinions are own)

I'm not sure, haven't noticed anything recent from him but according to his Linkedin page he is still member of staff at Nordic HW (even though he has another job as well).

There is a new article by Nordic HW, dunno how much truth is there to it:
https://www.nordichardware.se/nyhet...-nog-for-att-gora-titan-x-till-flaskhals.html

Rough headline (google) translation:

AMD Ryzen powerful enough to make the Titan X become the bottleneck

 

[zinfamous]

Well yes, but the last month has been slowed,

it's actually up about 30% over 30 days.

If any of this is true about performance, then the real gains will probably start showing up after Q2 numbers are reported in June. That will have a full quarter of Ryzen sales, but then by June you probably have Vega releasing as well as Naples--Naples being the real money-maker for AMD, so if things are only ever positive, Q3 could be nuts.

at 600% gains over a year, a lot of the hype and rumor is feeding this gain, so I honestly don't expect much more movement until after Q2 is reported. Bear in mind, though, that I don't know anything.

 

[hotstocks]

Oh, I get your point. I have never bought a non-K or non overclockable cpu in 20 years, lol

 

[vissarix]

You sure about that?

4.5GHz - 156 ST
4.8GHz - 166 ST
5GHz - 172 ST

Here it is...just did the test

 

[OrangeKhrush]

AMD is back, let's not get it twisted, AMD is back and matching INTEL and there's no other way to look at it. Sure you can nit pick here and there, but grand scheme of things, put up an INTEL CPU at 3.4GHz and compare it to 1800X and AMD will win. I can prove this, I have seen this and you all will this coming week.

AMD has extremely good memory bandwidth efficiency, better than INTEL's even in dual channel mode. So I wouldn't say the IMC is weaker per se, yes it's over 1GHZ lower than what INTEL can manage on air cooling (4266MHZ DDR4 vs 3200MHz max on AMD), but weaker is to undersell what AMD has managed here.
So in dual channel mode @ 3200MHZ, CL15 you can get up to 50GB/s memory writes in AIDA 64 which is damn near theoretical limit and that's at CL15.
AMD IMC always runs a command rate of 1T/1N. You can't change it at all and you also can't use really use the tight timings that we can right now on INTEL platforms. Those 12-12-12-26 B-die timings are not going to happen, but no matter, you'll still whip most 7700K systems that aren't tuned to death.

Overclocking this platform is crazy complicated and unrewarding. The clock speeds are low and if you have dreams of a 4.5GHz 24/7 setup, best forget it. In fact most of us will run 3.8 to maybe 4GHz on all cores if we're lucky for 24/7. 1.4V at 4GHz = 90'C load temp with air cooler, so be mindful of that. We need new coolers for the most part because the CPU height is not the same as previous CPUs amongst other changes.

For Haswell-E 5960X to match or beat 1800X (3.6~ 3.7GHzGHz) you'll need about 4.5GHz in the multithreaded programs as Ryzen is ridiculously fast at this beating 6950X sometimes.

 

[vissarix]

Another run..

 

[Crumpet]

I'm happy with 3.8-4 on all cores.

8 cores at 3.8 makes me pretty happy.. to be fair.

 

[formulav8]

 

[OrangeKhrush]

I'm happy with 3.8-4 on all cores.

8 cores at 3.8 makes me pretty happy.. to be fair.

You won't have much choice unless you have high end cooling

 

[Crumpet]

You won't have much choice unless you have high end cooling

And that counts as? I was looking at 280mm AIO.

 

[OrangeKhrush]

And that counts as? I was looking at 280mm AIO.

H100+ or equivilant, though watercooling is your best option to get 4.2+ 24/7

 

[.vodka]

Here it is...just did the test

Another run..

I tried bclk overclock, same as yours, got 167 ST again at 102.3*47 = 4.8GHz. It can't be your 2600k's extra 2MB L3 cache because the other link has a 4.8GHz 3930k scoring 163 ST with even more L3 cache for a single thread test. What are your RAM speed and timings?

Your 2600k is unusually strong, well outside margin of error (1-2%).

 

[Dave2150]

Anyone seen Sweepr lately?
He last updated this thread on 23rd Dec 2016. Since then zero updates and we have had some very positive signs for Zen's performance. I wonder if the topic would have been updated if we saw some crappy results .

I too wish he would update this thread with as much love as devotion that he puts into his beloved Skylake/Kabylake thread. I suspect he thought Zen would end up being rather poor; now that it's very likely going to upset the Intel monopoly he has stopped posting.

I assume he's financially involved with Intel, what a sad state of affairs.

 

[swilli89]

Wooooooooow this is absolute gold

 

[OrangeKhrush]

My source is internal and knows his stuff, he is trustworthy and he says that Ryzen is a beast but a beast that will not be good for high frequency, it is a low frequency high performance silicon uarch

 

[vissarix]

I tried bclk overclock, same as yours, got 167 ST again at 102.3*47 = 4.8GHz. It can't be your 2600k's extra 2MB L3 cache because the other link has a 4.8GHz 3930k scoring 163 ST with even more L3 cache for a single thread test. What are your RAM speed and timings?

Your 2600k is unusually strong, well outside margin of error (1-2%).

Tried to run them higher but they would crash after a while no matter what voltage or timing i use...so just end up using these settings.

 

[swilli89]

I too wish he would update this thread with as much love as devotion that he puts into his beloved Skylake/Kabylake thread. I suspect he thought Zen would end up being rather poor; now that it's very likely going to upset the Intel monopoly he has stopped posting.

I assume he's financially involved with Intel, what a sad state of affairs.

Of course he won't update this thread with positive Ryzen news. As speculated, the entire reason he made this thread under the false pretense of "news" is that he could better direct the narrative about Ryzen should any bad news surface. What's very interesting is this is textbook ACTUAL "schill" strategy used on reddit and many message boards. Companies will hire "PR" firms that employ thousands of registered users to promote their brand/products.

Its just awfully strange that while he claims to be unbiased he hasn't updated this thread. He has some leaks that have been proven to be blatantly wrong still on the first post! I say the first post should be reported as suspecting schilling/trolling.

Insulting other members is not allowed.
Markfw
Anandtech Moderator