ATI released OpenCL SDK with hardware support

Discussions related to GPU Acceleration in LuxRender

Moderators: Dade, jromang, tomb, coordinators

Re: ATI released OpenCL SDK with hardware support

Postby Dade » Fri Jan 08, 2010 7:37 am

SATtva wrote:
Dade wrote:Can I say that CPUs are starting to look a bit pathetic ? My poor Q6600 is totally owned by the 4870 :D The GPU is 9.83 times faster than the CPU :?

I know now how my next budget on the renderfarm upgrade will be spent. ^_______________^


Ssssshhh, don't tell my credit card, I ordered the components for a new PC yesterday ... I'm going to put so many GPUs inside that poor box :D
User avatar
Dade
Developer
 
Posts: 8311
Joined: Sat Apr 19, 2008 6:04 pm
Location: Italy

Re: ATI released OpenCL SDK with hardware support

Postby MikeJ » Fri Jan 08, 2010 1:52 pm

Dade wrote:Can I say that CPUs are starting to look a bit pathetic ? My poor Q6600 is totally owned by the 4870 :D The GPU is 9.83 times faster than the CPU :?


FINALLY after 25 pages of technobabble a post I can understand! Although I am suffering from a bit of nonbelief, and I need someone to check my math.

Dade's Q6600 should be roughly 2x the processing speed of my E6600. And his 4870 is almost 10x faster than his Q6600. So this means that a 4870 would be ~20x faster than my E6600?

Furthermore, from earlier in this thread:

Dade wrote:Today, I have seen a screenshot of SmallptGPU running on a ATI 5870, it is about 3-4 times faster than my 4870 ... what the hell :D


Does this mean that if I were to put a 5870 in my little E6600 box that it would render at 60 - 80x what my CPU can do?

This can't be right... Where is my mistake? These numbers are too huge.

Mike
Lux: 0.6.1
Blender: 2.49b
WinXP(32) - E6600 - 4GB
MikeJ
 
Posts: 221
Joined: Sun Nov 22, 2009 11:37 am

Re: ATI released OpenCL SDK with hardware support

Postby SATtva » Fri Jan 08, 2010 2:22 pm

Mike, don't forget that 5870 has 1600 stream processors on board vs 2-4 cores of mainstream CPUs. ;) They are not as versatile and fast as general-purpose CPU, but massively parallelizeable.
Linux builds packager
聞くのは一時の恥、聞かぬのは一生の恥
User avatar
SATtva
Developer
 
Posts: 7162
Joined: Tue Apr 07, 2009 12:19 pm
Location: from Siberia with love

Re: ATI released OpenCL SDK with hardware support

Postby refugee » Fri Jan 08, 2010 4:26 pm

yeah amazing speed..
I used to mess about with early Indigo on a single thread 2.66ghz Pentium and wait for 14-15 hours and I guess if I used two 5870 now the same render would be done in about a minute :shock:
User avatar
refugee
 
Posts: 11
Joined: Sun Apr 27, 2008 3:20 am

Re: ATI released OpenCL SDK with hardware support

Postby Sheltem » Fri Jan 08, 2010 4:58 pm

I just run the caustic.scn on a mac mini with nvidia 9400
Code: Select all
Usage: ./smallptgpu <use CPU/GPU device (0=CPU or 1=GPU)> <workgroup size (0=default value or anything > 0 and power of 2)> <kernel file name> <window width> <window height> <scene file>
Reading scene: scenes/caustic.scn
Scene size: 3
OpenCL Platform 0: Apple
OpenCL Device 0: Type = TYPE_GPU
OpenCL Device 0: Name = GeForce 9400
OpenCL Device 0: Compute units = 2
OpenCL Device 0: Max. work group size = 512
OpenCL Device 1: Type = TYPE_CPU
OpenCL Device 1: Name = Intel(R) Core(TM)2 Duo CPU     P8700  @ 2.53GHz
OpenCL Device 1: Compute units = 2
OpenCL Device 1: Max. work group size = 1
[SELECTED] OpenCL Device 0: Type = TYPE_CPU
[SELECTED] OpenCL Device 0: Name = Intel(R) Core(TM)2 Duo CPU     P8700  @ 2.53GHz
[SELECTED] OpenCL Device 0: Compute units = 2
[SELECTED] OpenCL Device 0: Max. work group size = 1
Reading file 'preprocessed_rendering_kernel.cl' (size 16931 bytes)
OpenCL Device 0: kernel work group size = 1


It sits around 6000K samples a sec. for CPU

Code: Select all
Usage: ./smallptGPU <use CPU/GPU device (0=CPU or 1=GPU)> <workgroup size (0=default value or anything > 0 and power of 2)> <kernel file name> <window width> <window height> <scene file>
Reading scene: scenes/caustic.scn
Scene size: 3
OpenCL Platform 0: Apple
OpenCL Device 0: Type = TYPE_GPU
OpenCL Device 0: Name = GeForce 9400
OpenCL Device 0: Compute units = 2
OpenCL Device 0: Max. work group size = 512
OpenCL Device 1: Type = TYPE_CPU
OpenCL Device 1: Name = Intel(R) Core(TM)2 Duo CPU     P8700  @ 2.53GHz
OpenCL Device 1: Compute units = 2
OpenCL Device 1: Max. work group size = 1
[SELECTED] OpenCL Device 0: Type = TYPE_GPU
[SELECTED] OpenCL Device 0: Name = GeForce 9400
[SELECTED] OpenCL Device 0: Compute units = 2
[SELECTED] OpenCL Device 0: Max. work group size = 512
Reading file 'preprocessed_rendering_kernel.cl' (size 16931 bytes)
OpenCL Device 0: kernel work group size = 192
OpenCL Device 0: forced kernel work group size = 32


and around 4500K for GPU :cry:
Plus the system noticeable slows down in gpu mode

EDIT
With cornel scene in GPU mode 450K and around 600k in CPU mode... quite a drop off in performance :roll:

2nd EDIT
Jens just pointed something out to me: workgroup size set to max changes the GPU results in cornel scene quite dramatically ( for a nv 9400 that is :) )
GPU mode 840K , CPU mode stayed pretty much the same...


Sheltem


Great work guys!!
Last edited by Sheltem on Sat Jan 09, 2010 2:04 am, edited 1 time in total.
Sheltem
 
Posts: 122
Joined: Sun Jun 15, 2008 11:06 am

Re: ATI released OpenCL SDK with hardware support

Postby Dade » Fri Jan 08, 2010 6:20 pm

MikeJ wrote:Dade's Q6600 should be roughly 2x the processing speed of my E6600. And his 4870 is almost 10x faster than his Q6600. So this means that a 4870 would be ~20x faster than my E6600?


SmallptGPU would be 20x faster, MandelGPU would be 124x faster and Luxrender ? No one really know yet.

First of all we have to modify Luxrender to be able to use GPUs. It is not going to be a trivial or short task even if we have now a good idea of how to modify Lux (and the idea is supported by some field experiment).

Do you want me to make an hypothesis of how fast will be the very first rendering done by Lux with the help of a GPU ? I would place my bet on 1.5 - 2 times faster (but it is really hard to say and should become a lot of faster with more time spent on optimizing the software).

MikeJ wrote:Does this mean that if I were to put a 5870 in my little E6600 box that it would render at 60 - 80x what my CPU can do?

This can't be right... Where is my mistake? These numbers are too huge.


The 5870 is probably more than 500-600 times faster than your E6600 on raw floating point computation but it is not like we can just recompile Lux and run on the GPU; the problem is more complex.

An E6600+HD5870 combo would be anyway too unbalanced. The CPU would be too slow to keep the GPU busy. However, in the future, you could buy a 5770 and cut your rendering times in half or 1/3 (by buying a 120euro GPU). This is exactly why we are all excited about this GPU thing: it is high floating point performances at low cost for everyone.

@Sheltem: your result isn't really surprising: your card has 16 streaming processor Vs the 1600 of a HD5870 or the 800 of my HD4870. You can expect a good speed only with the high end of the old generation GPU or the medium/high-end of the new one.

Anyway the point isn't really CPU Vs GPU but CPU Vs CPU+GPU so you will get a speed up on rendering times even if your GPU is slow.

P.S. I uploaded the SmallptGPU 1.6 to http://davibu.interfree.it/opencl/small ... ptGPU.html
It should include all the fix for the Mac and a new optional kernel that can do direct lighting integrator like rendering:

Image

It is quite funny to use, indeed it is really fast.
User avatar
Dade
Developer
 
Posts: 8311
Joined: Sat Apr 19, 2008 6:04 pm
Location: Italy

Re: ATI released OpenCL SDK with hardware support

Postby MikeJ » Fri Jan 08, 2010 6:50 pm

Dade wrote:An E6600+HD5870 combo would be anyway too unbalanced. The CPU would be too slow to keep the GPU busy. However, in the future, you could buy a 5770 and cut your rendering times in half or 1/3 (by buying a 120euro GPU). This is exactly why we are all excited about this GPU thing: it is high floating point performances at low cost for everyone.


Thanks for the thoughtful reply. I had not even considered the CPU/GPU balance issue. I guess it will take some experimentation to know what the minimum CPU requirement will be for each GPU type. Looks like it will be considerable time before the GPU tab will appear in Lux anyway, so I will fade back into the shadows on this topic and let you guys continue to have your fun. :)

Mike
Lux: 0.6.1
Blender: 2.49b
WinXP(32) - E6600 - 4GB
MikeJ
 
Posts: 221
Joined: Sun Nov 22, 2009 11:37 am

Re: ATI released OpenCL SDK with hardware support

Postby Sheltem » Fri Jan 08, 2010 8:39 pm

@ Dade just out of curiosity: is the cornel scene available as a blender / lux file? I would like to find out how many samples a sec the 'standard' lux is achieving...
I know that isn't exact science and will not be an indication on how much quicker a 'gpuLux' finally might be but hey I'm curios

Sheltem ;)
Sheltem
 
Posts: 122
Joined: Sun Jun 15, 2008 11:06 am

Re: ATI released OpenCL SDK with hardware support

Postby jensverwiebe » Sat Jan 09, 2010 12:33 am

@ Sheltem: iám pretty sure , you measured with simple.scn ;)

Jens
User avatar
jensverwiebe
Developer
 
Posts: 3383
Joined: Wed Apr 02, 2008 4:34 pm

Re: ATI released OpenCL SDK with hardware support

Postby Dade » Sat Jan 09, 2010 4:59 am

Sheltem wrote:@ Dade just out of curiosity: is the cornel scene available as a blender / lux file? I would like to find out how many samples a sec the 'standard' lux is achieving...
I know that isn't exact science and will not be an indication on how much quicker a 'gpuLux' finally might be but hey I'm curios


It wouldn't be a fair comparison, the samples produced by Luxrender have far less noise so you can not do a Samples/Sec Vs Samples/Sec comparison ... who care of being fair :D let's do a test.

Luxrender rendering after 60 seconds (Metropolis+Path):

test-lux.jpg


SmallptGPU after 60 seconds:

test-smallptgpu.jpg


Is "Brute Force" going to win over "Being Smart" ? This is not the right question: "Brute Force" and "Being Smart" is going to win over "Brute Force" or "Being Smart".

The number you see above (103K samples/sec Vs 6500K samples/sec), should give you a good idea why I can not keep my hand out OpenCL ;)

P.S. the Lux scene is attached to this message.
Attachments
cornell.lxs
(4.06 KiB) Downloaded 47 times
User avatar
Dade
Developer
 
Posts: 8311
Joined: Sat Apr 19, 2008 6:04 pm
Location: Italy

PreviousNext

Return to GPU Acceleration

Who is online

Users browsing this forum: No registered users and 2 guests