Cycles NVidia MAXWELL Benchmarks

yeah but im a gamer too.

my 750ti runs normally @GPU 1085mhz / RAM 1375 boost: 1163 (in cycles@~1228)
with some overclocking to 1408!! /1552 i get this results

BMW Scene : 49.46s


cornell bench: 1:55.02s


Pabellon bench: 2:19.03


This is the same speed like my GTX570.
the 750ti is cycles-stable but not furmark-stable
furmark stable @~1363/1500

1419MHz boost clock. RAM +200MHz offset
EVGA 750ti FTW
Card temperature reached 39 deg C.

BMW: 1:06

Core i3 4130T, 4GB RAM, Windows 7 64 bit using Rolf’s build.

I need to learn more about overclocking.

im getting a gtx750ti ! woot!
now, what brand ? I only know of MSI. They do great ATI cards. The popular ones seem to be Gainward and EVGA, but ive never heard good or bad reviews. I wont touch ASUS (i have my reasons)
Is there an nvidia card maker that it better than the others ? or any company i should avoid ?

Rolf, many thanks for the benchmarks :slight_smile:

BMW1M-MikePan.blend

1:11.97 - 128 x 64 (default setting, I also unchecked “Use Spatial Splits”)
1:04.43 - 128 x 128
0:59.58 - 256 x 256
1:09.24 - 512 x 512

Card is at stock settings, no plan to overclock it for now. :slight_smile:

Then you should wait for the other Maxvell cards.

holy crap those are great times

Thank you for this new build. But I think I must learn how to compile my own, to optimize for my AMD APU workstation.

Cool ill be holding out for a maxwell gpu for my next upgrade. Im happy i didnt go for sli with my 780ti

Does this mean that I can just plug in 2 gtx 750 tis, and they will work? Just like that? What about 3? Also, by running multiple.cards, will I be able to use 4 or even 6GB VRAM as opposed to the 2GB that the card supports?

Please forgive my questions. I don’t know much about.video.cards, or.even how SLI and multiple.cards work.

from what i have heard, the 750ti’s wont have more than 2gig of RAM so you would be limiting yourself. From what I remember being told, they cant share the RAM at the moment.

Hi, SLI is used in games for antialiasing, for example, splitting to two-four cards and send result back to “master” card over SLI bridge.
Render engines use cards separately and send result to display as tile or mixed result.
Therefor GPU memory is not added, the smallest memory limit the whole system.
If mainboard support it you can use several cards, they scale good.

Cheers, mib.

Hey, this seems to be a 64-bit build. Can we have a 32-bit version of the build you made?
Or if its already there, can you link me to it? Because the post was very big, so I couldn’t read through all of it :slight_smile:

EDIT: Sorry, doublepost.

Same 32bit request here.
Thanks!
:slight_smile:

sorry, i can´t upload new builds, i update the source code to 2.70 “RC2”
and since then cycles no longer works with the SM_50 kernel on 750ti :frowning:

Last test was on latest source code RC2 Hash:a861e55

Thanks anyway, Rolf.
Probably I’ll migrate my system to x64, and then your previous builds will be very usefull for me. I hope so…
Thanks for share!
:slight_smile:

Hi, there was a bug all cuda cards does not work but it is fixed now, update your sources. :slight_smile:

Cheers, mib.

Not my bug, the 570 works well, and my source is now only 4 hours old xD, i update my source everyday and tried it now since rc2 is out

you mean => Invalid value in cuMem Alloc(&device_pointer, size)

but i mean (in best case and a simple scene) => Not found in cuModuleGetTexRef(&texref, cuModule, name)

or totally crashed (on more complex scene)