Trade in Quadro K4000 for 2X GTX970

[Vizioen]

Is it a good idea to trade in my quadro K4000 for a setup with 2X GTX970? I know 3dsmax doesn't support SLI, but VrayRT does support mutliple GPU's. And also Octane benefits of multiple GPU's.

[vlado]

Two 970's would almost certainly be faster for GPU rendering than a single Quadro K4000. Plus, they have more VRAM (4 GB each). However, I have not done any direct comparisons...

Best regards,
Vlado

[Vizioen]

Two 970's would almost certainly be faster for GPU rendering than a single Quadro K4000. Plus, they have more VRAM (4 GB each). However, I have not done any direct comparisons...

Best regards,
Vlado

One 970 would be already faster than one single quadro K4000 wouldn't it? If I check the benchmarks it beats all cards except the 980. Also it has more CUDA cores with better technology.

[Vizioen]

Two 970's would almost certainly be faster for GPU rendering than a single Quadro K4000. Plus, they have more VRAM (4 GB each). However, I have not done any direct comparisons...

Best regards,
Vlado

Vlado, do you have a GTX970 at your disposal? If so, could I prepare a 3DSmax (2014) scene for you and see how it performs in the viewport (or anybody else on the forum). I'm certain that the 2GTX's would outperform the quadro. But as 3dsmax itself doesn't support SLI, I don't know if 1GTX will hold against millions of polygons scenes in comparison to the quadro K4000.

[Neilg]

I have a gtx670 which can handle 10+ million polys in the viewport with no slowdown. chokes a little when i switch on all my 5k textures but that's to be expected. This gets 5k points compared to the 970's 9k, i think you'll be fine.
Just get one first and wait for someone else to try them in SLI if you're worried.

[vlado]

Vlado, do you have a GTX970 at your disposal?

Nope, we don't have one, unfortunately. We will probably get a couple of 980's, but it's a bit different than the 970's. Even so it might be weeks before it reaches this part of the world.

Best regards,
Vlado

[Vizioen]

I have a gtx670 which can handle 10+ million polys in the viewport with no slowdown. chokes a little when i switch on all my 5k textures but that's to be expected. This gets 5k points compared to the 970's 9k, i think you'll be fine.
Just get one first and wait for someone else to try them in SLI if you're worried.

Yeah I might start with one, but SLI doesn't work in Max does it? I've googled it a lot over the past few years and it doesn't seem to have changed.

Also when it comes to looking at the specs of a card, for performance in max or rendering in VrayRT which ones are important? (for instance: pixel rate; memory bus; shading units; texture mapping units; output processors;...) Or all these only important for gaming and does it only come down to VRAM and clock speed?

[Neilg]

i think the pixel rate, memory bus and textures probably make a difference but it's all kind of meaningless and impossible to figure out how much they contribute. it's new, faster and cooler/silent.

[Vizioen]

Alright will keep you guys posted

[vlado]

I got a GeForce 980 today for testing; it runs fine, but I seem to be spoilt by the quadro K6000 - the 980 is about 16% slower than the K6000.

Best regards,
Vlado

[Morne]

I got a GeForce 980 today for testing; it runs fine, but I seem to be spoilt by the quadro K6000 - the 980 is about 16% slower than the K6000.

Best regards,
Vlado

I'm guessing that's for RT?
How do they compare in viewport alone?

[Vizioen]

I got a GeForce 980 today for testing; it runs fine, but I seem to be spoilt by the quadro K6000 - the 980 is about 16% slower than the K6000.

Best regards,
Vlado

That's not that much I might say if you compare the prices. And I'm indeed wondering about viewport performance.

[vlado]

That's not that much I might say if you compare the prices. And I'm indeed wondering about viewport performance.

Well, the K6000 also has 12 GB of VRAM I don't know how to measure the viewport performance though - any pointers?

Best regards,
Vlado

[Neilg]

It's got 2880 cuda cores compared to the 980's 2048, but it gets 10k to the quadro's 7k on passmarks speed test so might be faster in the viewport.

I think the only way to test is make a scene with millions of polys until breaking point and pan around it looking at the fps...

[duke2]

Hopefully they release a version down the road that has, say, 6gb ram (like the 580gtx got the 3gb version) as that seems to be the bottleneck, but in general NVIDIA has been pulling back the CUDA capabilities of the consumer cards since the 6xx series, trying to push the Tesla's.