Tesla

[Feniks]

Has anyone tested this on a tesla yet ? Just wondering.
If so, is it worth the investment ?

[YoonKyung]

Has anyone tested this on a tesla yet ? Just wondering.
If so, is it worth the investment ?

The performace of it would be same to GeForce cards which has same GPU.
But it would be nice for rendering large and or complex scenes.

[iknowmedia]

But didn't have Tesla cards way more cuda cores?

[radiance]

But didn't have Tesla cards way more cuda cores?

They are essentially the same cards but with more memory and they lack a display connector,
hence they are only for compute / CUDA, not for gaming or displaying a desktop.

Radiance

[iknowmedia]

Thanks for the info!

So I can buy a Tesla and put it next to my gtx480 for extra raw power and memory for octane?

[radiance]

if you need the memory and you have lots of money to spend, be my guest
although i think a GTX480 is sufficient for most uses, and way cheaper.

Radiance

[Feniks]

It might be worthed trying with a small tesla farm once the multiGPU is stable imho.
I'm just curious about how many you'll need to get it rendering at delayed realtime if possible at all.

[radiance]

It might be worthed trying with a small tesla farm once the multiGPU is stable imho.
I'm just curious about how many you'll need to get it rendering at delayed realtime if possible at all.

I would definately consider purchasing a tesla cards that's based on the fermi chip, eg a GTX480 tesla equiv.
besides more speed, it has many other advantages...

Radiance

[EricNV]

I would definately consider purchasing a tesla cards that's based on the fermi chip, eg a GTX480 tesla equiv.
besides more speed, it has many other advantages...

Radiance

Like multi-GPU render farms

EricNV

[YoonKyung]

It might be worthed trying with a small tesla farm once the multiGPU is stable imho.
I'm just curious about how many you'll need to get it rendering at delayed realtime if possible at all.

If you want to build a real time render farm, you must fix two arguments, size of 'render target' and 'max samples per second'.

if render_target_size = HD (width 1280, height 720);
and max_samples_per_second = 256(medium quality);

real_time_fps = 25;

render_target_pixels = 1280 x 720 = about 1M;
tesla_samples_per_second = 4M;
tesla_frames_per_second = tesla_samples_per_second / render_target_pixels = 4 fps;

num_teslas = (max_samples_per_second / tesla_frames_per_second) x real_time_fps = (256 / 4) x 25 = 64 x 25 = 400;
num_pcs_has_4_teslas = num_teslas / 4 = 400 / 4 = 100;

So, if you have 100 pcs which has 4 tesla cards in it, you can render HD size render target in real time.
And if you decrease the render_target_size to half then 50 pcs can do that.
Anf if you increase the max_samples_per_second upto double then 200 pcs will be needed to do that.

EDIT: fixed -> fix