PDA

View Full Version : NVIDIA Tesla C2050 x i7 processor How do I calculate?



silviotoledo
06-10-2010, 04:51 PM
Need a help here.

How do I calculate the amount of benefit while rendering from this N Vidia Card and a i7 Processor?

NVIDIA Tesla C2050 x i7 processor

How much more teraflops or processing power the new nvidia videocard has in comparison to an i7 processor???????

I'm assuming that Lightwave will support Cuda soon, and that we can use Octane Render while it does not support.

cresshead
06-10-2010, 05:06 PM
i think in the near future [3 to 5 years] graphics card chip driven rendering will start to be used on some productions but for the time being it's just pie in the sky dreaming for realworld needs on production and not simplistic or custom coded demo's trying to showcase such stuff.

Lightwolf
06-10-2010, 05:31 PM
What cress said...

I've recently red a report on the top 500 list of super computers, one of which is using GPUs as well.
Using standard benchmarks (Linpack in this case) it seems that CPUs can leverage up to 90% of their theoretical floating point performance, GPUs reach around 30%.
So much for the theory.

When it comes to rendering it's also extremely hard to compare as other rendering methods are used on the GPU (and need to be used to be effective in the first place). CPUs in general are a lot more flexible and thus also allow for a lot more code trickery when rendering.

Cheers,
Mike

silviotoledo
06-10-2010, 05:37 PM
hmmm... so let's wait the future again :(

I just thought this math would be interesting:

Core i7 = more than 98 gigaflops

NVIDIA Tesla C2050 = 515 Gigaflops per GPU and

has 448 CUDA nucleos ( Is each nucleo a GPU? )


So, would this Nvidia card ( that costs only $

3.500 ) get 2.300 times more render power while

rendering with Octane render as example?

Titus
06-10-2010, 06:11 PM
i think in the near future [3 to 5 years] graphics card chip driven rendering will start to be used on some productions but for the time being it's just pie in the sky dreaming for realworld needs on production and not simplistic or custom coded demo's trying to showcase such stuff.


I've advised a studi to try Machstudio. They are rendering an animated feature film, they had time contrain problems using an existing render farm and the report now is their rendering power increased up to 20x, they bought only 3 or 4 licenses.

Lightwolf
06-10-2010, 06:16 PM
Core i7 = more than 98 gigaflops

NVIDIA Tesla C2050 = 515 Gigaflops per GPU and

has 448 CUDA nucleos ( Is each nucleo a GPU? )

No, the 515 GFlops are the theoretical maximum of all cores on the GPU.

However, if you break down according to the 90%/30% ratio you end up with 90 vs. 154 (roughly) .... and GPUs aren't well suited for every kind of processing.

Machstudio is interesting in that regard as it seems to be a hybrid renderer that is also OpenGL/D3D based as opposed to a pure GPU renderer (that just uses the GPU as if it was a CPU - CUDE/OpenCL - not using the realtime graphics APIs - OpenGL/D3D).

Cheers,
Mike

nolme
07-23-2010, 04:35 AM
Hi,

I'm looking for the same type of informations.

Today, LW9.6 don't support GPU acceleration :(.
Tomorrow, LW10 seems to support it but when ?

Does somebody have informations on it ?