Will CUDA 2.0 Support 9600GT(G94)?

Will CUDA 2.0 Support 9600GT(G94)?
I want to purchase a piece of 9600GT(G94) Video Card.

All CUDA versions will support the GPU that are G80+ the question is does it support the special features that CUDA 2.0 has. But I think it does, there is not much difference between 1.1 and 2.0 except for the 3D texture support and VISTA support.

I’m doing quite well with my 9600GT and Cuda 1.1 :D

To make it work with cuda 2.0b you have to install beta drivers (which have lower version number than current release drivers for 9600gt ).

I f u have any question just ask :D

thx. :D
What about the performance?
Is it has better performance in CUDA than the Geforce 8800GS?
I know 9600GT has the better GAME performance than 8800GS.

hard to say because i run it on quite slow cpu (AMD64 3000+ @2333mhz) so results could be different on current CPUs like Core2…

As example you can take my “GPU convolution time” which is = 11.991748 msec with release build of convolutionSeparable CUDA SDK example, so you should compare to results with different cards… hmmm :D

Hmm, what card are you using?

I did the convolutionSeparable in 20,… msec

4096 x 4096

Initializing data...

Warm up...

GPU convolution...

GPU convolution time : 20.962000 msec //800.363328 Mpixels/sec

Reading back GPU results...

Checking the results...

...running convolutionRowCPU()

...running convolutionColumnCPU()

...comparing the results

L1 norm: 8.621137E-08

TEST PASSED

Shutting down...

I’m using a 8800 GTS 320MB on a Quad Xeon 5355 @ 2.66GHz

Oh wait, when I run the 1.1 version it does the job in

GPU convolution time : 11.781000 msec //801.051175 Mpixels/sec

this is report from convolution separable example with 2.0b sdk:

Using device 0: GeForce 9600 GT

3072 x 3072

Initializing data...

Warm up...

GPU convolution...

GPU convolution time : 11.981411 msec //787.652140 Mpixels/sec

Reading back GPU results...

Checking the results...

...running convolutionRowCPU()

...running convolutionColumnCPU()

...comparing the results

L1 norm: 9.449628E-008

TEST PASSED

Shutting down...

Press ENTER to exit...

but i can see different texture size… hmmm

For what it’s worth, on a 9800 GTX (2.0beta, WinXP-32):

GPU convolution time : 6.415076 msec //1471.094703 Mpixels/sec

Can you run the 2.0 Beta on that card? It says in the install that the hardware (9800 GTX) is not supported yet. Not true?

WOW nice scores on those Geforces. I have an older AMD x2 64 fx 2.6 ghz sys with a quadro 570:

3072 x 3072
Initializing data…
Warm up…
GPU convolution…
GPU convolution time : 220.162338 msec //42.864661 Mpixels/sec
Reading back GPU results…
Checking the results…
…running convolutionRowCPU()
…running convolutionColumnCPU()
…comparing the results
L1 norm: 9.502583E-008
TEST PASSED
Shutting down…

Press ENTER to exit…

EEK

and on my AMD Quad AM2+ 64 3.2 ghz with a Geforce9800GTX:

3072 x 3072
Initializing data…
Warm up…
GPU convolution…
GPU convolution time : 6.648331 msec //1419.481636 Mpixels/sec
Reading back GPU results…
Checking the results…
…running convolutionRowCPU()
…running convolutionColumnCPU()
…comparing the results
L1 norm: 9.502583E-008
TEST PASSED
Shutting down…

Press ENTER to exit…

It’d be cool to start seeing more bench like stuff somewhere on these forums

I am running a SuperMicro SuperServer with Intel Xeon Quad Core @ 2.77 GHz, (2) Tesla C1060, (1) Quadro FX 370LP.

[codebox]3072 x 1536

Allocating and intializing host arrays…

Allocating and initializing CUDA arrays…

Running GPU convolution (10 identical iterations)…

Average GPU convolution time : 1.735300 msec //2719.179292 Mpixels/sec

Reading back GPU results…

Checking the results…

…running convolutionRowCPU()

…running convolutionColumnCPU()

…comparing the results

Relative L2 norm: 0.000000E+00

TEST PASSED

Shutting down…

Press ENTER to exit…[/codebox]

This result shouldn’t be surprising since the Tesla is a repurposed Quadro Chipset.