Hi ,)

It is possible to make a Client for NVIDIA-CUDA?

NVIDIA-CUDA GPU ...

http://developer.nvidia.com/object/cuda.html


Seti source code´s

http://setiathome.berkeley.edu/sah_porting.php


Maxxx

---

Well

I tried to sign up to their program to get access to their FFT libraries.

You have to fill out a massive form and i did, i haven't even got any response from them yet (48 Hours) so it seems like that Nvidia doesn't want a couple of thousand people to rush out to buy a new graphicsboard just to run S@H on it..

That's really a shame because i would have dug deeper into that matter because the 8800 is arriving tomorrow and to try developing a Cudabased app to that board would've been much more inspiring than trying out games on it in my case..

Kind Regards Vyper

---

_________________________________________________________________________
Addicted to SETI crunching!
Founder of GPU Users Group

Do you think it´s possible to form a client to be produced? Lg Maxxx

---

Do you think it´s possible to form a client to be produced? Lg Maxxx

If the precision is high enough for the unified shader = yes

&

The main executable is rewritten so you can include Cuda based optimisations just like the IPP is working , so you don't need to be a rocketscientist to figure that one out = yes

If all above = yes then i see no reason to why not..

//Vyper

---

_________________________________________________________________________
Addicted to SETI crunching!
Founder of GPU Users Group

Already discussed here http://setiathome.berkeley.edu/forum_thread.php?id=35342

---

Already discussed here http://setiathome.berkeley.edu/forum_thread.php?id=35342

Yup, it has been discussed.. But some of the questions are still not answered and that is if the GPU math precision enough to let the gpu calculate S@H data.. That is one issue that hasn't been discussed thoroughly though..

//Vyper

---

_________________________________________________________________________
Addicted to SETI crunching!
Founder of GPU Users Group

Already discussed here http://setiathome.berkeley.edu/forum_thread.php?id=35342

Yup, it has been discussed.. But some of the questions are still not answered and that is if the GPU math precision enough to let the gpu calculate S@H data.. That is one issue that hasn't been discussed thoroughly though..

//Vyper

It is ;-)


Regards Hans

---

ATI is trying to outdo Nvidia -- turns a graphics card into a scientific accelerator.

AMD Puts Gaming Graphics Chip into High-Performance Computing Apps.

---

does anyone think that a unit like this could be worked aster, I mean most of the stuff seti does is based on fft. wouldnt it make much more sense to develop a seti specific hardware? Something like a fpga that receives data via rs232? if th core is developed just do this kind of work, i guess it would increase the speed, wouldn't it?

does anyone think that a unit like this could be worked aster, I mean most of the stuff seti does is based on fft. wouldnt it make much more sense to develop a seti specific hardware? Something like a fpga that receives data via rs232? if th core is developed just do this kind of work, i guess it would increase the speed, wouldn't it?

You'll need an FPGA with about 4MB of on-chip RAM to really make it take off.
I don't think FPGAs have grown this big by now.


Regards Hans

---

does anyone think that a unit like this could be worked aster, I mean most of the stuff seti does is based on fft. wouldnt it make much more sense to develop a seti specific hardware? Something like a fpga that receives data...

You'll need an FPGA with about 4MB of on-chip RAM to really make it take off.
I don't think FPGAs have grown this big by now.

However, a cheap FPGA chip + 4MByte or more of RAM chips is easily doable. Also, there are various development cards available that plug into the PCI bus.

And the FPGA could be programmed up via Handel-C.

Meanwhile, the GPU move away from deep pipelines to multiple discrete execution units as in the latest nVidia 8xxxx GPUs should make retasking the GPU for general number crunching much easier.

Looks interesting!

Regards,
Martin

---

See new freedom: Mageia Linux
Take a look for yourself: Linux Format
The Future is what We all make IT (GPLv3)

However, a cheap FPGA chip + 4MByte or more of RAM chips is easily doable. Also, there are various development cards available that plug into the PCI bus.

And the FPGA could be programmed up via Handel-C.

You'll have to compete with the L2 bandwith of a current CPU when doing a FPGA implementation.

There are plenty of inexpensive FPGA boards to be had with single-channel SDRAM at 100Mhz.

At 800MB/s memory speed they won't cut it for seti.

You'll need one helluva memory setup with lots of independent channels to get to the required speed.

This means huge FPGAs with lots of pins and a very expensive board.

Meanwhile, the GPU move away from deep pipelines to multiple discrete execution units as in the latest nVidia 8xxxx GPUs should make retasking the GPU for general number crunching much easier.

Looks interesting!

Regards,
Martin

I'm already drooling over the 8800 specs myself :o)


Regards Hans

P.S: This doesn't rule out crunching on an FPGA just for the fun of it :o)

---

[...]
I'm already drooling over the 8800 specs myself :o)


Regards Hans

P.S: This doesn't rule out crunching on an FPGA just for the fun of it :o)

So am I, Hans, so am I...

However, then I read the specs for the upcoming ATI card and forgot all about the 8800 again.

It will have an estimated 80% bandwidth advantage over the 8800 models (up to ~150 GB/sec vs. 88) due to the wider 512-bit memory bus (which is also arranged differently, as a ring-bus).

Also, from what I read, it will support general purpose processing even better because it uses SIMD processing units, albeit fewer than the scalar ones on the 8800s. It will have 64 of them and be able to do 4x the work on the SIMD parts (ATI R600) vs. having 128 and all of them scalar (nVidia 8800 GTX), so in effect, it'll be (at least) twice as powerful for general purpose stuff (it will also come in 1GB and 2GB variants, not 640MB and 768MB as the current nV models).

Anyway, since both of these cards now were designed with general purpose processing (GPGPU) in mind, adapting code to their architecture should be much more straightforward. No more writing shaders for FFTs, no more deep pipeline penalty for cache misses, much more complex calculations possible, better FP precision, the works.

I've already had about 10 people ask me or mail me about an app for the 8800s...it'd be a miracle if some enterprising fellow didn't have a go at it :)

The only thing holding back many developers will be the prohibitive cost (600+ dollars/euros) for a piece of hardware to play with.

Regards,
Simon.

---

Donate to SETI@Home via PayPal!

Optimized SETI@Home apps + Information

oki, i got my new 8800 GTX with a new QX6700 thanks to my friends at Nvidia and Intel. Please someone make a client for the 8800 so i can test it.

---

The only thing holding back many developers will be the prohibitive cost (600+ dollars/euros) for a piece of hardware to play with.

This is one of the biggest drawbacks of ROI for programming the GPU into a cruncher - not many of them will be had initially because of cost. But I'm sure someone will still want to make a cruncher just to prove that it can be done, and eventually these chips will become commonplace and affordable.


Vyper asked about the new graphics chips and their floating point capability; from what I've seen of these new chips' specs, the 8800 nVidia chip can do 128bit floating point math, which should be plenty strong enough to do crunching (I'm unsure about the new AMD/ATI chip, but I wouldn't doubt it if it was similar). Plus, as Simon said (Simon Says? ;) ), these new graphics chips are being designed as General Purpose Graphics Processing Units (or ATI has called some of theirs Visual Processing Units/VPUs), which means they can be programmed a lot easier than previous chips, making them much more palatable to do math calculations for science.

---

oki, i got my new 8800 GTX with a new QX6700 thanks to my friends at Nvidia and Intel. Please someone make a client for the 8800 so i can test it.

Just mail your hardware to me - I'll see what I can do :o)

Regards Hans

---

Hans -

check this thread at lunatics.at if you're seriously interested :o) (or if you can recommend someone suitable).

Regards,
Simon.

---

Donate to SETI@Home via PayPal!

Optimized SETI@Home apps + Information

check this thread at lunatics.at if you're seriously interested :o) (or if you can recommend someone suitable).

Just looked but:

"The topic or board you are looking for appears to be either missing or off limits to you."

OK, so I'm not Hans!

Cheers,
Martin

---

See new freedom: Mageia Linux
Take a look for yourself: Linux Format
The Future is what We all make IT (GPLv3)

Hans -

check this thread at lunatics.at if you're seriously interested :o) (or if you can recommend someone suitable).

Regards,
Simon.

Done ;-)


Regards Hans

---

check this thread at lunatics.at if you're seriously interested :o) (or if you can recommend someone suitable).

Just looked but:

"The topic or board you are looking for appears to be either missing or off limits to you."

OK, so I'm not Hans!

Cheers,
Martin

You're right, you're not ;o)

Seriously though, that's because that thread lives in a restricted area (the pre-release boards at lunatics.at).

If you're seriously interested in writing optimized code for SETI@Home or BOINC, whether for GPUs or for CPUs, tell me your username on lunatics.at (you need to be registered there) and I'll give you access. Serious offers only, please (that goes for everyone interested ;o)).

HTH,
Simon.

---

Donate to SETI@Home via PayPal!

Optimized SETI@Home apps + Information