GTX 1080 vs Quadro M4000 for FP32/64 CUDA work

[Chaotic42]

Hey all, I'm thinking about upgrading my K1200 to a more robust card for more intensive CUDA performance and it occurs to me in my ignorance that a GTX 1080 has more cores for a lower cost. I'm not using any special software like CAD or Iray or anything, just CUDA development.

Is there any reason to spring for the M4000 over a 1080? Am I missing something here?

 

[Timmah!]

Quadros do come with professional drivers, which are great for stuff like 3D modeling in software like 3Ds MAX or Maya and god know what more. I can only tell from my own experience, with regular Geforce cards this kind of work can be really painful. Rotating camera around your high poly-count model, zooming in and out, turning wireframe mode....all of that can become pretty slow and stutter like hell, even with high end Geforce. Not to mention jaded edges and blinking surfaces... I never had Quadro, but from my understanding, its meant to deal exactly with this stuff and turn this kind of work into dream.

Now, regarding CUDA stuff, i have no idea whether Quadros have anything over Geforces, which would be helpful for CUDA development - honestly i cant even imagine, what "CUDA development" means :-D. I can however say, for things like iRay, GTX1080 would be vastly superior performance-wise. M4000 seems to be akin to GTX970 based on their equal CUDA core numbers, possibly even slower cause of lesser chip frequency. With Octane Render, which is similar to iRay, M4000 performs like this:

https://render.otoy.com/octanebench/summary_detail_item.php?systemID=1x+Quadro+M4000

55 points on average, actually to my surprise quite a lot worse than gtx 970, which gets 80.

GTX 1080 gets about 140, if you OC to 2-2,1 GHz then you look at 160-170 score. So at least 3x as much.

Since M4000 is based on Maxwell, its probably equally capped at FP64 performance, so it does not even that going for it compared to 1080. Thats just my assumption, though.
All in all, i cant be 100 percent sure, but i would believe 1080 is a better choice.

 

[David_k]

If you want REAL FP64 Performance, If I'm not mistaken, your ONLY option is the SUPER EXPENSIVE GP100 based Pascal Tesla, the GP102/104 Tesla's and the GP104 Quadro's won't do, the chip is designed to deliver 1/32 Perfrormance in FP64 mode, so your only option for that aspect exept selling your car is to get a Compute Capable Kepler, while the GP100 is a 1/2 FP64 rated card, GK110 was a 1/3 Rated chip, but only on the Titan&Titan Black, and the quadro K6000, this will give you 1.5Tflops-1.7TFLops on FP64, even a Titan XP will give you only 0.3TFLops in FP64 Mode thanks to the 1/32 rate design..

But if you really don't need FP64, the 1080 will runs miles around the GTX970 Based quadro M4000.. should be 3x times more powerfull than the M4000, and MILES between your K1200..

 

[Chaotic42]

If you want REAL FP64 Performance, If I'm not mistaken, your ONLY option is the SUPER EXPENSIVE GP100 based Pascal Tesla, the GP102/104 Tesla's and the GP104 Quadro's won't do, the chip is designed to deliver 1/32 Perfrormance in FP64 mode, so your only option for that aspect exept selling your car is to get a Compute Capable Kepler, while the GP100 is a 1/2 FP64 rated card, GK110 was a 1/3 Rated chip, but only on the Titan&Titan Black, and the quadro K6000, this will give you 1.5Tflops-1.7TFLops on FP64, even a Titan XP will give you only 0.3TFLops in FP64 Mode thanks to the 1/32 rate design..

But if you really don't need FP64, the 1080 will runs miles around the GTX970 Based quadro M4000.. should be 3x times more powerfull than the M4000, and MILES between your K1200..

Yeah, that's what I was afraid of. I can live without FP64, the car not so much.

Thanks for the replies folks.