AMD does reverse GPGPU, announces OpenCL SDK for x86

Printable View

Show 100 post(s) from this thread on one page

08-30-2009, 02:05 PM
David F

Quote:

Originally Posted by Drwho?

David, in Post http://www.xtremesystems.org/forums/...6&postcount=39, I pointed this page out already, I am very aware of this methode ;-) ... This is the next point I wanted to make ...

Sorry (not entirely sure why you coded the other method then).

See also http://gmplib.org/cgi-bin/gmp_calc.pl?expr=1000000!

Well under a second. (I didn't think a Java implementation could be the fastest going).
08-30-2009, 02:28 PM
[XC] riptide

Quote:

Originally Posted by Chumbucket843

i posted that because no one did a cpu to gpu comparison and i was curious to see what the results would be.

Ya. Remember as the shader count goes up, things get interesting. Also the shader clocks plus the Vram memory speed and latency is going to be critical.
08-30-2009, 02:30 PM
[XC] riptide

Quote:

Originally Posted by David F

Sorry (not entirely sure why you coded the other method then).

See also http://gmplib.org/cgi-bin/gmp_calc.pl?expr=1000000!

Well under a second. (I didn't think a Java implementation could be the fastest going).

Ya.... http://gmplib.org/cgi-bin/gmp_calc.pl?expr=500000000000

500000000000! is also under a second. :shrug:

Infact 50000000000000000000000000000000000000000000000000 0000000! is also under a second.

Its a broken site. ;)
08-30-2009, 02:34 PM
Drwho?

Well, I am not doing what many other algorithm are doing, I do not Store the previous known n! in a table file ... I beleive http://gmplib.org/cgi-bin/gmp_calc.pl?expr=500000000000 does ;)
I am writing a program that will really do all the calculation.
08-30-2009, 02:42 PM
David F

Quote:

Originally Posted by [XC] riptide

Ya.... http://gmplib.org/cgi-bin/gmp_calc.pl?expr=500000000000

500000000000! is also under a second. :shrug:

You're missing the !. With the ! it says "expression required too much memory to evaluate". (For me, at any rate).
08-30-2009, 02:52 PM
[XC] riptide

Quote:

Originally Posted by David F

You're missing the !. With the ! it says "expression required too much memory to evaluate". (For me, at any rate).

Ah right. In your link the "!" was outside.... ;)

http://gmplib.org/cgi-bin/gmp_calc.pl?expr=2000000!

Quote:

The result of executing 2000000! is:

computation took 1526 ms
result is about 11733474 digits, not printing it

:cool:
08-30-2009, 02:56 PM
David F

Quote:

Originally Posted by [XC] riptide

Ah right. In your link the "!" was outside.... ;)

Cursed vbulletin URL parsers...

But it does kind of make all the "3 minutes on CPU CUDA v.s. 1 minute on GPU" arguments look a little silly, doesn't it?
08-30-2009, 02:59 PM
[XC] riptide

Quote:

Originally Posted by David F

But it does kind of make all the "3 minutes on CPU CUDA v.s. 1 minute on GPU" arguments look a little silly, doesn't it?

I know.... sssshhhh. Don't tell Dr. Who though. :D:p:
08-30-2009, 03:04 PM
Drwho?

well, it is all about the funny GPGPU claims ...
http://2.bp.blogspot.com/_srnPcN5adC...0/NVBSrayT.jpg
This is nVidia 1000X faster than CPU raytracing ... lol ... do i need to add something ?

but ssssshhhhh, don't say anybody ... Raytracing is suppose to look like this:
http://2.bp.blogspot.com/_srnPcN5adC...3_334111_n.jpg

lol
08-30-2009, 03:22 PM
[XC] riptide

Quote:

Originally Posted by Drwho?

well, it is all about the funny GPGPU claims ...

lol

Ya it is very funny....

http://dl.getdropbox.com/u/600721/ElcGPU.JPG
http://dl.getdropbox.com/u/600721/ElcNoGPU.JPG
08-30-2009, 03:29 PM
Mechromancer

I need to give Elcomosoft Wireless Security Advisor a try on my 4870. Seems to work well with ATI hardware.

http://www.elcomsoft.com/images/ewsa.png
08-30-2009, 03:32 PM
Chumbucket843

Quote:

Originally Posted by Drwho?

well, it is all about the funny GPGPU claims ...
This is nVidia 1000X faster than CPU raytracing ... lol ... do i need to add something ?

it is funny but what else can nvidia say? our gpu are 10-20 times faster and you have to have mad skills to achieve this. by the way you might want to check out renderants.every raytracing demo on cuda zone sucks.
08-30-2009, 03:33 PM
[XC] riptide

Quote:

Originally Posted by Mechromancer

I need to give Elcomosoft Wireless Security Advisor a try on my 4870. Seems to work well with ATI hardware.

Works very well on GPU my friend. You see. the lesson we are learning here today is that to speak in absolute terms is wrong regarding CUDA, ATi Stream or X86 based CPU's. The thing is... and I know Dr. Who will agree, is that CUDA/GPU based vid. encoding might be bad for high end stuff... and plenty of things that GPU can't do... BUT... there are plenty of things a GPU can do.... and can do very fast in the field of computing, besides rendering games etc.
08-30-2009, 03:46 PM
Mechromancer

Quote:

Originally Posted by [XC] riptide

Works very well on GPU my friend. You see. the lesson we are learning here today is that to speak in absolute terms is wrong regarding CUDA, ATi Stream or X86 based CPU's. The thing is... and I know Dr. Who will agree, is that CUDA/GPU based vid. encoding might be bad for high end stuff... and plenty of things that GPU can't do... BUT... there are plenty of things a GPU can do.... and can do very fast in the field of computing, besides rendering games etc.

I think we find that CPUs and GPUs work well together to handle complex tasks that take advantage of each of their strengths. OpenCL will use both GPUs and CPUs at the same time and that is something to look forward to. Also a company called Caustic is using both CPUs and GPUs to simultaneously accelerate ray-tracing.

To further illustrate my point, take out your GPU and try to run Crysis. Then reinstall your GPU and take out your CPU and try to run Crysis. Seems we've been divvying up workloads in applications since the GPU hit the scene. Why should future apps that take advantage of computer shaders be any different?
08-30-2009, 03:55 PM
Drwho?

The most funny of all, an SSD increases more the performance the score on the big industry benchmarks than upgrading to a discret graphic card ... hehehe ... something to think about ...
Then, when you plug the SSD, the increase is mostly CPU related ...
Francois
PS: I hope you noted the humours tone.
08-30-2009, 04:02 PM
[XC] riptide

Quote:

Originally Posted by Drwho?

The most funny of all

... is your blog link in your sig. You say here it took 3 hours to do the factorial code, but in your blog you say it took 6 hours. :shrug:

Quote:

well, I putted this program together in 6 hours,
08-30-2009, 04:14 PM
Drwho?

Quote:

Originally Posted by [XC] riptide

... is your blog link in your sig. You say here it took 3 hours to do the factorial code, but in your blog you say it took 6 hours. :shrug:

yes ... 6 hours was with Threading.
08-30-2009, 10:52 PM
Calmatory

Quote:

Originally Posted by Drwho?

The most funny of all, an SSD increases more the performance the score on the big industry benchmarks than upgrading to a discret graphic card ... hehehe ... something to think about ...
Then, when you plug the SSD, the increase is mostly CPU related ...
Francois
PS: I hope you noted the humours tone.

Indeed. Credits go to OCZ for that. :)
09-01-2009, 11:41 AM
Drwho?

Quote:

Originally Posted by Calmatory

Indeed. Credits go to OCZ for that. :)

http://anandtech.com/storage/showdoc.aspx?i=3631
09-01-2009, 12:14 PM
Xoulz

Quote:

Originally Posted by Drwho?

well, it is all about the funny GPGPU claims ...
http://2.bp.blogspot.com/_srnPcN5adC...0/NVBSrayT.jpg
This is nVidia 1000X faster than CPU raytracing ... lol ... do i need to add something ?

but ssssshhhhh, don't say anybody ... Raytracing is suppose to look like this:
http://2.bp.blogspot.com/_srnPcN5adC...3_334111_n.jpg

lol

Exactly... :clap:
09-01-2009, 01:31 PM
trinibwoy

Quote:

Originally Posted by Xoulz

Exactly... :clap:

So you're impressed by a comparison between a finished CPU screenshot and a shot of a GPU doing it in almost real-time (something a CPU can't even dream of)? Don't let other people think for you my friend. There's no difference in quality between CPU and GPU ray-tracing, just speed. Anybody who says different is intentionally talking bull or simply doesn't know what they're talking bout.
09-01-2009, 01:35 PM
Manicdan

Quote:

Originally Posted by trinibwoy

So you're impressed by a comparison between a finished CPU screenshot and a shot of a GPU doing it in almost real-time (something a CPU can't even dream of)? Don't let other people think for you my friend. There's no difference in quality between CPU and GPU ray-tracing, just speed. Anybody who says different is intentionally talking bull or simply doesn't know what they're talking bout.

yeah, why cant people be happy that we have SOME kind of real time raytracing. its new and will be expanding very rapidly in the next few years. one screenshot took minutes to hours, the other was done in milliseconds. there were demos of raytracing were the final results looked identical to cpu renders, and it looked like it took 3 seconds to get a "perfect" frame.
09-02-2009, 12:42 PM
Chumbucket843

Quote:

Originally Posted by trinibwoy

So you're impressed by a comparison between a finished CPU screenshot and a shot of a GPU doing it in almost real-time (something a CPU can't even dream of)? Don't let other people think for you my friend. There's no difference in quality between CPU and GPU ray-tracing, just speed. Anybody who says different is intentionally talking bull or simply doesn't know what they're talking bout.

exactly.

its just a bunch of number crunching.
09-03-2009, 03:50 AM
gosh

OpenCL tutorial (video)
http://www.macresearch.org/files/opencl/Episode_2.m4v

http://www.macresearch.org/opencl_episode2
09-14-2009, 05:25 PM
Chumbucket843

these are some pretty awesome gpgpu numbers:
http://fah-web.stanford.edu/cgi-bin/...?qtype=osstats
its a pity my i7 only gets 24 gigaflops.:shrug:

Show 100 post(s) from this thread on one page