Btw, anybody can tell me if the GTX 590 or the 6990 are seen as an unique OpenCL device or, on the contrary, as two independent OpenCL devices, pls?
Printable View
Btw, anybody can tell me if the GTX 590 or the 6990 are seen as an unique OpenCL device or, on the contrary, as two independent OpenCL devices, pls?
I still don't see any improvements with my 3 560 GTXs. The benchmark does look like cinebench now but it doesn't act like it as i obvisiouly don't get a better time score. To me it seems that all the GPUs are used all the time but not all are used at 100%.
What version are you using? Try to use the latest one ( 0.5.0 ), pls.Quote:
Originally Posted by soya_crack
If you render with (1 GPU vs 3 GPUs) or (1 GPU vs 1 CPU) you should notice any difference for sure :)
Are you performing hybrid rendering? Render only with the GPUs, disable all the CPU devices and see if that helps.
I've uploaded the 0.5.1. Fixed several problems with Catalyst 11.3
Concerning Post Number 128.
I do of course see a difference comparing 3 Cards with 1 Card. But there is no performance difference between v.0.5 and v0.4.9.
In Version 4.9 all GPUS were used at 100% Usage but the lasr frame was always only rendered by 1 card.
In version 5 all GPUs render all the time but not at 100 %, that is why there is no difference between 5 and 4.9
Thanks for the detailed explanation, soya.
Well, the 0.5.0 performs AA and the 0.4.9 not, so if their render times are equal then then 0.5.0 is much faster in practise because it's taking much more samples !
However, the last tile is a bit problematic: as the GPU requires tons of data to perform optimal parallel execution, I cannot make it much smaller. The 0.4.9 used bigger tiles but the 0.5.0+ uses sightly smaller ones. In theory, that should help SLI synchronization but, in practise, reducing the data size affects performance so the net win might be annulled.
It's hard to solve it. All the parallel tasks must wait at some point to perform synchronization. It's near impossible to get a 100% scaling. In a CPU renderer the tiles can be smaller but GPUs need lots of data to amortize the ALUs and the big-latency memory fetches. I neither can avoid to use tiles because, then, it will be almost impossible to render big images without being out of memory.
It is support as plugin in 3ds max 2012?
3dsmax 2010, 2011 and 2012, yep.Quote:
Originally Posted by realbabilu
I plan to support Maya soon too.
im not going to pretend that i read this thread all the way so it may be covered.
the rv770/790 work with this so why are they not supported officially on the list of working cards (both detect as the rv770)
http://www.ratgpu.com/verify.aspx?k=...B6373916752916
its not quick but it was not using the cpu (that was an rv790, 4890 not a 770 4870)
Oh, nice to see a 4XXX running well.
The main reason for not supporting the 4XXX is that I lack a card to test and that the AMD APP SDK does not support images/multiple UAV buffers for the 4XXX.
It's strange that your 4XXX supports all this but maybe your drivers are patched to emulate that functionality in some way.
i think that amd dose not support it since its slow, but the box said openCL.Quote:
Originally Posted by jogshy
can u change the site to say that it may work on the rv770/90 as i think that those are the only ones with openCL support that works, or it might be a change from the 770 to 790 like the change is shader cashe and power trenching and shader timings
HD4670 (RV730) 1GB DDR3. Windows 7 with Catalyst 11.3 .
http://www.ratgpu.com/verify.aspx?k=...0BE2CC16752999
Will test with my 4870 later....
I've uploaded the 0.5.2:
- Added Maya 2012 support ( Windows and Mac ).
- Added support for ATI Radeon HD 4XXX cards.
- Optimized a bit the precomputation phase.
- Fixed several installation problems under linux.
- Removed the need for an Internet connection and the expiration date.
- Improved rendering speed for ATIs a lot ( 33%-100% ) ... BUT you'll need Catalyst 11.5 or above or ratGPU may hang.
OMG 4xxx support, looks like ill be testing this when i get home tonight
great work!
stock 4850: 268 seconds
http://www.ratgpu.com/verify.aspx?k=...C12D1C16752996
Nice, thanks for testing.
NVidia Quadro 600 : 299.56s
http://www.ratgpu.com/verify.aspx?k=...CE1E0B167527D9
HD4870 650MHz Ratgpu 0.52 beta - 252.197548s
http://www.ratgpu.com/verify.aspx?k=...D925141675297B
Catalyst 11.5 hotfix
HD 6970 910 core, 1450 mem clock
83.505 seconds
http://www.ratgpu.com/verify.aspx?k=...1FEFDA16752E1D
Nice!
Btw, anybody with a Fusion E-350 to test, pls? I'm curious.
4870 (1Gb) 775/935 clocks - 190.264 secs
http://www.ratgpu.com/verify.aspx?k=...B3434E1675293E
I was curious if any platform differences would come into play, and also if modded 6950 is somewhat a bit slower (mem timings ?).So i did a test with same settings.Quote:
Originally Posted by xPliziT
Appears all is same.
Cat 11.5b
HD6950@6970 910 core, 1450 mem
83.484
http://www.ratgpu.com/verify.aspx?k=...76867316752EFE
Cpu scores :
APP 652.338
Rat Gpu Cpu C++=1750.830
ill try again to convince my brother to run it,Quote:
Originally Posted by jogshy
and ill leave out the part about how long it will probably take when does the cpu :D
or a Fusion A8-3850 :p:
I got my card back from warranty land, so here goes a 4890.
284 seconds. Catalyst 11.6
http://www.ratgpu.com/verify.aspx?k=...A2323F167529A1
EDIT: I think my scores are pretty bad because I'm running a simulation right now (and a lot more queued)... I'll try again in a few months when I get a newer computer, lol.