SuperPi on GPU, were going CUDA

[Charles Wirth]

The first steps were taken to port Superpi to CUDA so we should see working software in less than 30 days.

What is CUDA?
"The CUDA™ Toolkit is a C language development environment for CUDA-enabled GPUs."
http://www.nvidia.com/object/cuda_learn.html

CUDA will enable SuperPi to run on the GPU.

We will soon be able to bench on hundreds of cores inside our GPU's for an exact performance number. This will be very interesting to massive parallel versus massive cache.

With the success of Superpi exceeding 100 million downloads it will have an existing installed base of 70 million CUDA capable GPU's to run on and will grow by 100 million per year or so.

Current list of CUDA enabled GPU's
http://www.nvidia.com/object/cuda_learn_products.html

My thanks to everyone around the world for making it one of the most popular benching applications.

It will remain free of adverts and free to use, the program name will include CUDA GPU.

[Ashraf]

Nice! Will there be a validation, like this: http://www.xtremesystems.com/pi/?

[Warboy]

Nice! Can't wait.

[ownage]

It is possible to run CUDA on ATI right? Or will this be a nVidia bench only?

[Charles Wirth]

I will fix validation, not sure what broke it still. It should work but it doesn't.

I do know when it was served on IIS it does work so I may go that route with its own server.

[madcho]

cuda is nvidia's API

[Charles Wirth]

Not sure of current situation with ATI on GPGPU, they feel its best to exclude me.

[eva2000]

sweet holy goodness

[dinos22]

i look forward to doing a 20 second 32M SuperPi run eek

[[XC] gomeler]

This should be very interesting. I suspect though that GPGPUs won't have the single-threaded speed that a 5+GHz Core 2 would have. Look forward to this though

[ExodusC]

Aww, my 7800GTX isn't CUDA enabled... I bet it could beat a lot of the top of the line CPUs in SuperPi, though.

I can't wait to get a new video card... It's gonna be cool to bench GPUs like this!

[metro.cl]

while you are at it add 64m, 128m and more

Great news i always thought that NVIDIA would do this just to get on Intel's toes.

[lowfat]

Hopefully this is just the beginning of GPGPU

[BenchZowner]

Sort of comedian look & accent:
"And there goes the world's "cheapest" 2D benchmark"
SuperPi is now a..."3D" benchmark

Boy SuperPi records are about to get owned

[nfm]

Very nice! Make it open source please, I want to see the source code pr0n.

[DRCRAWFISH]

Woah, this is awesome.

We will be owning CPu records I hope!

[dinos22]

Sort of comedian look & accent:
"And there goes the world's "cheapest" 2D benchmark"
SuperPi is now a..."3D" benchmark

Boy SuperPi records are about to get owned

see this is the problem.....ppl conceive all sorts of ideas over a few beers
you can blame Vince for planting this idea in my head during dinner


Hey Charles how can you name this SuperPi benchmark CUDA GPU and not have any reference to SuperPi

Maybe reconsider calling it CUDA GPU SuperPi or something unless that's what you meant in first place

[Boogerlad]

maybe call it gpuPi?

[mrcape]

Wow CUDA sounds really cool. Great news!

[SKYMTL]

It will remain free of adverts and free to use, the program name will include CUDA GPU.

Naming it with the CUDA moniker is the same as advertising IMO.

Great news though.

[MuffinFlavored]

This inspired me to download the CUDA SDK and toolkit, so I did.
But, then it got blurry.

I hope this rocks.

Instead of multi-threading SuperPi for CPU, make the jump and make hundreds of threads on the GPU.

[dinos22]

what about using those parallel processing units to compute single threaded applications

i thought that should be doable and produce good results as well

anyways it will be interesting to see how it all turns out and whether superpi speed will scale with the amount of cores the CPU has

then we would have to start to tweak RAM latencies EEK
actually that would not be very smart in case you can't go back to defaults lol

[cadaveca]

Not sure of current situation with ATI on GPGPU, they feel its best to exclude me.

Really?

Of course it runs on ATI cards! Don't tell THEM that though!

[Luka_Aveiro]

then we would have to start to tweak RAM latencies EEK
actually that would not be very smart in case you can't go back to defaults lol

nvflash + vga pci card FTW?

headaches FTL?



^Jen-Hsun Huang is a very moody guy

[Cooper]

Hopefully this is just the beginning of GPGPU

It's not the beginning of GPGPU, not the first implementation of SPi on GPU and definitely CUDA was not made for benchmarks...it's just a byproduct.