Ryzen

[werticus]

Be very interesting to see if the new AMD cpu's that are due out on 3/3 can compete in vray!

http://wccftech.com/amd-ryzen-7-1800x-world-record/

Apparently the new 1800x now holds the world record for cinebench.

[vlado]

Yes, it will be interesting. It looks like we might be getting a configuration from AMD for testing, so we'll see...

Best regards,
Vlado

[Alex_M]

Does anybody know if Ryzen 1800X supports Embree and if not, how would this affect rendering performance with Vray? I was looking at replacing some CPUs and was initially looking at i7 6950x but they're expensive as heck. 1800X maybe won't be as fast but at least it's more than 3 times cheaper.

[werticus]

Vlado, you should put together a bench mark type of application with vray stand alone, so that the press can run vray benchmarks instead of cinebench - as a marketing avenue for you and something that will be very handy for your user base to make hardware purchase choices with

[vlado]

We do have a V-Ray Bench in the works. It should be released fairly soon actually.

Best regards,
Vlado

[Alex_M]

It appears that somebody with an engineering sample of a Ryzen processor did the Corona benchmark! (results attached below)

The engineering sample has the code name ZD3406BAM88F4_38/34_Y, which according to my Google search should be Ryzen 1700X (8 core/16 threads, 3.4-3.8 Ghz, 399 USD).

If this benchmark result is legit, this is very impressive to say the least. Even running at its base clock of 3.4 Ghz, this $400 chip is just 15% slower than the 10-core behemoth i7 6950x (when also running at its base clock frequency). Mind blown if true. We should point out that the 15% faster 6950x at the time of this post costs 4.25 times more than Ryzen 1700X - a whopping $1700!

As far as I know, Corona, like Vray, takes advantage of Embree enabled processors. So the question is, how much of this 15% advantage for 6950x is due to the native Embree support? It would be interesting if there was a test with Embree disabled on the 6950x because even with it enabled, the 10-core 6950x doesn't seem to be much faster than the 2nd best processor (1700X) in the Ryzen lineup. So I would also like to see benchmark results of the Ryzen flagship CPU (1800X) with boost clock turned on and of course with overclock. Hopefully we won't be waiting very long now. If Ryzen 1800X doesn't disappoint, I might build my first AMD based PC since 2004.

[vlado]

Embree has been working well for us even on AMD processors. We are using just the plain SSE2 version, no SSE4.1 or AVX.

Best regards,
Vlado

[Alex_M]

I received an email about a new reply to this thread from Vlado but I don't see the reply here. Even the link that leads to the thread is strange - https://forums.chaosgroup.com/node/932230 Is it possible that it didn't appear because of the froum maintenance yesterday?

Here's the message:

Embree has always performed quite fast for us, even on AMD processors. Most of the code is not written in a way that favors any particular processor. Further

So Embree is definitely supported by AMD? I'm asking because I'm seeing this reply from Ondra on the Corona forum:

I also tested embree on an AMD rig, and it was not significantly slower. I suspect Corona is slower there because I only develop and optimize on Intel. Maybe sometimes in the future I will try to optimize also on AMD.

So in terms of code optimization, Vray has been developed to be platform agnostic unlike Corona which is heavily biased towards Intel processors?

PS: Vlado's comment has just appeared. Strange, maybe some forum glitches.

[werticus]

that system only has 16gb ram too but not sure if that would matter in that particular bench mark... in any case, it looks promising!

[vlado]

So Embree is definitely supported by AMD?

Yes. AMD even included Embree in the CPU version of their FirePro renderer.

So in terms of code optimization, V-Ray has been developed to be platform agnostic.

Well, we also develop mostly on Intel processors although we have some AMD machines for testing. In general though, we have tried to avoid Intel-specific optimizations and instructions. Hopefully AMD can bring in some much needed competition here

PS: Vlado's comment has just appeared. Strange, maybe some forum glitches.

We wanted to upgrade to vbulletin 5 this morning and the upgraded version was visible for a few hours, unfortunately we found issues so we had to roll back the old version. Unfortunately some posts were lost in the process.

Best regards,
Vlado

[Vizioen]

We wanted to upgrade to vbulletin 5 this morning and the upgraded version was visible for a few hours, unfortunately we found issues so we had to roll back the old version. Unfortunately some posts were lost in the process.

Best regards,
Vlado

The /usercp.php doesn't work anymore and redirects to the main forum page. :/

[vlado]

The /usercp.php doesn't work anymore and redirects to the main forum page. :/

Will look into it.

Best regards,
Vlado

[RotemS]

We are using just the plain SSE2 version, no SSE4.1 or AVX.

I am curious; is it that the gains were not impressive enough, or that you want to avoid maintaining a separate build for SSE4.1/AVX deprived machines?

[vlado]

I am curious; is it that the gains were not impressive enough, or that you want to avoid maintaining a separate build for SSE4.1/AVX deprived machines?

We didn't see any performance improvements really. Judging by what other are reporting, even if there were any improvements, they would have been in the single-digit percentage range and it's not worth to maintain separate builds for a couple of percentage points. Compare this to the adaptive lights, which can bring ~50% speed improvement on any architecture

Best regards,
Vlado

[Alex_M]

Is this legit? The $399 Ryzen 1700X when overclocked is 25% faster than the $1700 10-core monster 6950X in multi-threaded applications. And that's 1700X, not even the 1800X. Is this even possible? Maybe it was overclocked on liquid nitrogen.