V-Ray Benchmark 1.0 release

[cmjohncheng]

hmm, weird. Try that one for xeons:

https://mab.to/gzhO9j2FH

this one can download, just tested mine, look like same, timer stop counting when the first bucket become very small.

[vlado]

And as my friend said, scenes looks like 10 years old.

The one in Cinebench looks 20 years old, and it's still a popular benchmark. While it is a bonus if the scene looks nice, it is not quite relevant to the purpose of the benchmark.

Scene for CPU has only 4 objects or so, dont think it represents powerful and modern renderer as Vray should be, but its an opinion.

Making the scene more complicated or slower to render will not make the actual hardware comparisons any different in the vast majority of cases.

But real problem is seconds counting, I think we can put all results in a trash.

This is true, once the results get into the seconds, they tend to be not very reliable. We are thinking of ways to improve this for future versions of the benchmark.

Best regards,
Vlado

[Jiri.Matys]

The one in Cinebench looks 20 years old, and it's still a popular benchmark. While it is a bonus if the scene looks nice, it is not quite relevant to the purpose of the benchmark.

Making the scene more complicated or slower to render will not make the actual hardware comparisons any different in the vast majority of cases.

This is true, once the results get into the seconds, they tend to be not very reliable. We are thinking of ways to improve this for future versions of the benchmark.

Best regards,
Vlado

About Cinebench - you are right, its really popular benchmark and I like it too. But I think VRaybenchmark should make a wow efect to a newbies, like oh my god, I can render this with vray, I need it right away You know what I mean, its VRay - most popular renderer for last 10 years, so I think it should represent it right, thats all. And I know scene is irelevant, for comparision scene doesnt matter and it works of course, but if you look at it only this way, you can put a few spheres in a scene and thats it. When it render two minutes on i7, it will be enough for comparision too.

[varosi]

hmm, weird. Try that one for xeons:

https://mab.to/gzhO9j2FH

V-Ray Benchmark is measuring only V-Ray rendering time. You could verify that by checking CPU load in Task Manager. Currently Benchmark UI is slower and we'll fix that soon.

[alkhall]

Downloaded, but the benchmark will not run, it flashes for a second, then closes.

Win7 U x64 SP1.

[daniel.taskoff]

Downloaded, but the benchmark will not run, it flashes for a second, then closes.

Win7 U x64 SP1.

It's probably the issue mentioned in this thread.

We'll release a minor version which fixes that problem, but the current solution is to install the Microsoft's VC14 runtime,
which can be downloaded from here (x86 version).

If you have more questions on the matter please post them in the other thread..

[vlado]

But I think VRaybenchmark should make a wow efect to a newbies, like oh my god, I can render this with vray, I need it right away

Ok, point taken. We will consider other scenes for future versions.

Best regards,
Vlado

[vlado]

However, the cpu test seems to stop the timer before the actual image is complete though - is this normal? Maybe related, but it also seems like my CPU usage goes down significantly after most of the full size buckets are done rendering. It still has all buckets in use unitl the end apparently but CPU use goes from 100% down to ~10% (this is on a 72 core system). Is this normal?

This issue is now fixed in the 1.0.2 version of the benchmark on our website. However like I mentioned, the benchmark results themselves are not actually affected by it as it was a display problem only.

Best regards,
Vlado

[im.thatoneguy]

Ok, point taken. We will consider other scenes for future versions.

Best regards,
Vlado

Not to be too cynical, but there is no way to know what's actually rendering. After all, you already have a grayed out version in the BG being revealed. You could consider just throwing up a beautiful random render and use the existing scene (if you've found it's reflective of general vray performance).

But it would be great to see all features tested in a future test: volumetric god rays, SSS, instancing, motion blur etc. Get a bit more of a cross section of the code-base features tested.

It's going to be hard though to keep this as a long-term standard since it would be most useful for actual vray users if you keep updating the code to reflect the latest optimizations in your code.

[Tim Van]

I see that 2 gp100's seem to perform almost about as fast as 4 1080's ti's. they are just that good or is there a scaling issue somewhere? thnx!

[vlado]

they are just that good

They are that good.

Best regards,
Vlado

[Tim Van]

They are that good.

Best regards,
Vlado

thanks vlado, good to hear. Pure fp32 compute they are a little below titanX/P6000 so the gains are mostly HBM related?

in any case that should be good news for better priced devices with HBM in the future

[im.thatoneguy]

Just out of curiosity is it just the faster HBM memory that makes it so much faster than the 1080 ti?

Both have 3584 cores.
Both are around 1.4GHZ
Both are around 250W

The 1080 Ti on paper actually claims higher FP32 TFLOPS.

But the GP100 looks to be about 33% faster.

[savage309]

Just out of curiosity is it just the faster HBM memory that makes it so much faster than the 1080 ti?

Both have 3584 cores.
Both are around 1.4GHZ
Both are around 250W

The 1080 Ti on paper actually claims higher FP32 TFLOPS.

But the GP100 looks to be about 33% faster.

TFLOPs can be compared only between same architecutres. Although both 1080ti and GP100 are called "Pascal" architecture, in fact they have different one.
In fact GP100 architecture is quite different to all other Pascal GPUs (all other Pascal GPUs are much more similiar to Maxwell).
The HBM2 for sure plays a huge part, but the GP100 architecture improvements most likely help as well. I can't say which (architecture/HBM) helped by exactly how much though ... but the GP100 is for sure crazy fast.

Best,
Blago.

[Streetwise]

Just tried to run 1.0.2. The CPU runs great (1:30) but then for GPU, nothing happens for about 10-15 seconds at which point, a hard re-boot is triggered. Not sure what that's all about. GTX780 6GB card.