OTOY Forums

Hey,

3.08 Ms/sec | 5.88 FPS

Nvidia : 590GTX
I7 2600K 3.40GHZ
8 GO de RAM.
Octane 2.46 (pathtracing).

luc255 - really? the 590 should be getting more like 6 Ms/sec....maybe there's something I'm missing, isn't the 590 a unique card somehow?

Yes :

MODEL : N590GTX-P3D3GD5
GPU : Geforce GTX 590
Memory size : 3072MB GDDR5 (1536MB X2)
Memory bus : 768 Bits (384BitsX2)
Output : Mini Display port + DVIx3.

(I am in toolkit 4.0, and driver set to be updated).

But What is the N590GTX in the photo (in attachments) ?
Because it's note same a 590gtx : http://pan.fotovista.com/dev/0/3/10094130/u_10094130.jpg

Thx.

matej wrote:I've done the benchmark with 2.47 and I've entered my username in the form, so my entry can be removed.

I'm not sure about the significant performance gains with CUDA 4.0, tho. I haven't noticed much difference

Oh yes. it is, between 3.0 and 4.0 version of Octane there is 30 percent speed difference (according to my tests). I would call it significant improvement.
This is my own post from the 2.47 dedicated thread to support my claim:



OK, little info regarding performance compared to 2.46b>
i tested so far only the trench benchmark scene, both with pathtracing and PMC and its interesting indeed>

tested with one core of GTX590 @700 MHz only as multiGPU does not work yet at 1920x1200>

- on 2.46b CUDA 3.0 i get with pathtracing 3,18 Ms
- on 2.47 i get with pathtracing 4,27 Ms!!! this is brilliant speed improvement indeed, probably down to CUDA 4.0, cant wait for the multi-GPU to be functional
- on 2.47 with PMC i get 3,29 Ms - its still faster than pathtracing on 2.46b CUDA 3.0 and it really seems to get away with fireflies faster, although this scene is not probably the best for testing new kernel. it does have a lot of fireflies in the beginning, as the walls of the trench are glossy be default.

going to test now those caustics and dispersion

in any case, well done, keep it up

colorlabs wrote:luc255 - really? the 590 should be getting more like 6 Ms/sec....maybe there's something I'm missing, isn't the 590 a unique card somehow?

It is obviously one core result only.

Yes, the screenshot shows one of the cores inactive.

@Timmaigh

yes there is some difference, 14% in my case:

2.46b with CUDA 3.2: 1.51 Ms/sec
2.47 with CUDA 4.0: 1.72 Ms/sec

I have been using CUDA 3.2 since it was possible. There is probably more speedup if you go from 3.0 to 4.0. That means the the benchmark should also differentiate between 2.46b for CUDA 3.0 and the one for CUDA 3.2

It would be probably better to do this benchmark when everyone will be on CUDA 4.0, with the same Octane version

matej wrote:@Timmaigh

yes there is some difference, 14% in my case:

2.46b with CUDA 3.2: 1.51 Ms/sec
2.47 with CUDA 4.0: 1.72 Ms/sec

I have been using CUDA 3.2 since it was possible. There is probably more speedup if you go from 3.0 to 4.0. That means the the benchmark should also differentiate between 2.46b for CUDA 3.0 and the one for CUDA 3.2

It would be probably better to do this benchmark when everyone will be on CUDA 4.0, with the same Octane version

Yes, i would think so, 2.48/2.5 with working multiGPU and CUDA 4.0 would make most sense. I was using 3.0 version obviously cause of the multi-gpu and there was indeed perf increase with 3.2 already, about 15 percent as you said.
Anyway, i just wanted to say, that everyone should stick to results from same Octane release with same CUDA version and with exactly same scene and paramaters, otherwise there is not much point in doing this benchmark. Colorlabs seemed to be rather benevolent with this, understandably though, as obviously he cannot order anyone to install 2.46 back, when the already use 2.47...

Yay someone posted a 560! Now we're just missing a 480 and a 465!!

I got the ASUS ENGTX570 DCII/

My Benchmark is:

Octane Render 2.46: 2.92 Ms/sec
Octane Render 2.47: 3.50 Ms/sec

I also have 2 Full HD screens connected to the same card... I have to test if I get more resources if I do the benchmark with only one connected...