GPU benchmarks

[super gnu]

ok after fiddling enormously with my new titan x (watercooled, modified bios, overvolt, overclocked core and ram)

1525mhz core 4040mhz ram.

my results with vray 3.20.03 and 353.06 nvidia drivers:


benchmark 1: 1m 22 seconds

benchmark 2: 8m 36 seconds


its not the highest result i got, (8m 22 seconds on benchmark 2, 1560mhz core, 4100mhz ram) but i was getting some major instability (sometimes driver crash on render end, sometimes not) so ive dialled things back until im stable.


interestingly, unlike in games, overclocking the ram seems to have as much, if not more effect than overclocking the core.

to be honest i was hoping for bigger gains considering the amount of fiddling ive done, but its possible my card isnt among the bet overclockers out there, i got the cheapest one i could find online.

[super gnu]

ok so it turns out i still had the cuda ptx optimisation manually disabled, from some earlier issues with another build.

with that env variable removed, my times are:

benchmark 1: 1m 15.5 seconds.

benchmark 2: 7m 54 seconds.

thats a bit more like it.

[eyepiz]

Nvidia K6000
Driver 3.53.82

3DS Max 2016 SP1
Vray RT 3.20.03

benchmark 2: 13m 8.4 seconds.

Seems freaking slow to me.

[3LP]

Well, quadro are not designed to render, they are only designed to be better at viewport, although this can be debated.

Stan

[savage309]

Nvidia K6000
Driver 3.53.82

3DS Max 2016 SP1
Vray RT 3.20.03

benchmark 2: 13m 8.4 seconds.

Seems freaking slow to me.

Are you using CUDA or OpenCL ?

[eyepiz]

Are you using CUDA or OpenCL ?

It was set to CUDA I just opened the scene ant hit render.

[auke]

As far as I know both GTX as Quadro series use the exact same CUDA protocols, so when using the GPU as a GPGPU device (what you do during CUDA processes) the optimisations should be identical.

[3LP]

if you look at this chart :
https://docs.google.com/spreadsheets...A0k/edit#gid=0

You'll see that the K6000 is somewhat the same as the Titan X and that with only one card, the benchmark 2.0 should be done in about 11m14.
Not sure why yours is doing 13m, it's a bit slower, might be the drivers that are a bit slower for that particular version, but you're still around the same ballpark, although for pure cuda computing, the K6000 is so much more expensive compared to the Titan X...

Stan

[savage309]

if you look at this chart :
https://docs.google.com/spreadsheets...A0k/edit#gid=0

You'll see that the K6000 is somewhat the same as the Titan X and that with only one card, the benchmark 2.0 should be done in about 11m14.
Not sure why yours is doing 13m, it's a bit slower, might be the drivers that are a bit slower for that particular version, but you're still around the same ballpark, although for pure cuda computing, the K6000 is so much more expensive compared to the Titan X...

Stan

In fact, it is K6000 (not M6000), which is expected to be ~30% slower compared to M6000 (according to NVidia).

[eyepiz]

So I found this interesting...I was looking up recommended drivers for 3DS Max 2016 on Autodesks website and see that the Quadro drivers are recommended and certified for the Nvidia Titan and Titan Z? could this be correct?

[AlexP]

4 x Titan X - 30s


edit:

8 X Titan X = 14s

Think maybe the scene is "too easy" to test with more power?

[allemyr]

1x 780 Ti 3gb
1x 580 3gb
Driver: 347.25

1min 17.5s



3gb vram is not much. Can't decide between Titan X or Titan Z unfortunately, needing more vram.
As I understand Titan Z is dual Titan Black? Will 6gb be enough?

[3LP]

If you can afford, go for Titan X, if you can't go for 980 i
Depends on the scene you'll work with either 6Gb or 12Gb

Stan

[3LP]

4 x Titan X - 30s
[ATTACH=CONFIG]26019[/ATTACH]

edit:

8 X Titan X = 14s

Think maybe the scene is "too easy" to test with more power?

Use benchmark 2.0, it has been designer to suit bigger GPU systems :
https://docs.google.com/spreadsheets...A0k/edit#gid=0

Should take about 1min to render with your 8 Titan X.
By the way, do you DR to get the 8 Titan X working? If so, from your wrkstn to the two nodes or from one node DR to the other?

Stan

[AlexP]

Use benchmark 2.0, it has been designer to suit bigger GPU systems :
https://docs.google.com/spreadsheets...A0k/edit#gid=0

Should take about 1min to render with your 8 Titan X.
By the way, do you DR to get the 8 Titan X working? If so, from your wrkstn to the two nodes or from one node DR to the other?

Stan

4 in my workstation, and 4 in another.

Ill check out the benchmark 2, will also test with 12 Titan X, out of interest.

So far seems to scale quite linearly I guess