Latest drivers (NVidia and ATI)

[Jason G]

LoL, maybe they shifted the bug department into the web server division & now the SDK & drivers will be bug free 

[Mike]

AMD APP SDK 2.5 RC2 is available for download

http://developer.amd.com/sdks/AMDAPPSDK/downloads/Pages/default.aspx

Claggy

Its full version now.

Still RC2. Release link gives not found 404 error

If your get the URL for the 32bit release, and edit it to the 64bit filename, then the download works,

Claggy

ROFL, if such tricks required only to get this SDK I could imagine how it will work after that.... better stay with older one for now

I´ve downloaded full version without any problems last night.

[arkayn]

I have the 64-bit version downloading now.

It will not do any good at the moment as I am running only Nvidia currently.

[Jason G]

I have the 64-bit version downloading now.

It will not do any good at the moment as I am running only Nvidia currently.

LoL

[Raistmer]

How important to know not only marketing hypes and hardware specifications but SDK capabilities too:
http://forums.amd.com/devforum/messageview.cfm?catid=390&threadid=149524&enterthread=y

concurrent kernel execution is not supported in SDK 2.4 and not planned for 2.5 either.

-------------------------
Micah Villmow
Advanced Micro Devices Inc.
--------------------------------
The information presented in this document is for informational purposes only and may contain technical inaccuracies, omissions and typographical errors. Links to third party sites are for convenience only, and no endorsement is implied.

So, forget about FERMI-like capabilities of HD6xxx at least for next half year. BTW, it would be nice to get direct confirmation that so discussed FERMI capabilities per se are supported in NV SDK...
Running few tasks at once on FERMI GPU doesn't means anything in this context, kernel overlapping behavior is required.

[Raistmer]

Aha, and for nVidia...:

I also have a NVIDIA Fermi card, and there is no way to get concurrent kernels on that using their OpenCL (driver v280.19 w/ OpenCL 1.1), with either single out-of-order queue or multiple queues.  You can get concurrent kernels using CUDA Runtime API, but then it's not in OpenCL.

Ken

[Jason G]

280.26 WHQL is out.  If 280.19 is anything to go by it might me worth a look for many.  Will test it myself here soon.

[Later:] Similar performance with experimental optimised code to 280.19 beta driver here, Temps reached 107 degrees C.

Device: GeForce GTX 480, 1640 MHz clock, 1503 MB memory.
Compute capability 2.0
Compiled with CUDA 3020.

                Chirp Unit test #3d (double Chirp)

Stock:
  Processing...
 Done!
  GFlops Avg(   68.94) Peak(   68.95) Min(   68.90)
  GB/s   Avg(   91.91) Peak(   91.93) Min(   91.87)


Opt: Elems/thrd 1, thrds/blk 64 - pk thru 156.4984 GB/s, 195.6230 GFlops
Opt: Elems/thrd 2, thrds/blk 64 - pk thru 149.9762 GB/s, 187.4702 GFlops
Opt: Elems/thrd 4, thrds/blk 64 - pk thru 159.4488 GB/s, 199.3110 GFlops
Opt: Elems/thrd 8, thrds/blk 64 - pk thru 164.2478 GB/s, 205.3098 GFlops

Opt: Elems/thrd 1, thrds/blk 128 - pk thru 154.4557 GB/s, 193.0696 GFlops
Opt: Elems/thrd 2, thrds/blk 128 - pk thru 170.7901 GB/s, 213.4876 GFlops
Opt: Elems/thrd 4, thrds/blk 128 - pk thru 163.2162 GB/s, 204.0203 GFlops
Opt: Elems/thrd 8, thrds/blk 128 - pk thru 162.6486 GB/s, 203.3107 GFlops

Opt: Elems/thrd 1, thrds/blk 192 - pk thru 161.0965 GB/s, 201.3706 GFlops
Opt: Elems/thrd 2, thrds/blk 192 - pk thru 160.4949 GB/s, 200.6186 GFlops
Opt: Elems/thrd 4, thrds/blk 192 - pk thru 164.0530 GB/s, 205.0662 GFlops
Opt: Elems/thrd 8, thrds/blk 192 - pk thru 160.4640 GB/s, 200.5800 GFlops

Opt: Elems/thrd 1, thrds/blk 256 - pk thru 162.5058 GB/s, 203.1322 GFlops
Opt: Elems/thrd 2, thrds/blk 256 - pk thru 148.2003 GB/s, 185.2503 GFlops
Opt: Elems/thrd 4, thrds/blk 256 - pk thru 161.9992 GB/s, 202.4989 GFlops
Opt: Elems/thrd 8, thrds/blk 256 -  N/A

Too hot.  I'll invent some way to throttle optimisations back.... 

[Richard Haselgrove]

Let me eat lunch first, then I'll download it and run chirp test on the 420M (we have temps for that, too). Just 3d, or would any of the others be useful?

[Jason G]

Let me eat lunch first, then I'll download it and run chirp test on the 420M (we have temps for that, too). Just 3d, or would any of the others be useful?

Anything you can throw into the mix might turn out useful.  The current suspicion (of mine)  is that some of my kernels are simply 'too hot', so that throws some factory OC'd models over the edge.  Probably treading new ground then, so some effective way to handle things might be in order (like advanced user flags + failsafes).  We'll See.

[Ghost0210]

From my 465 with the 280.26 drivers:

Device: GeForce GTX 465, 1450 MHz clock, 993 MB memory.
Compute capability 2.0
Compiled with CUDA 3020.

                Chirp Unit test #3d (double Chirp)

Stock:
  Processing...
 Done!
  GFlops Avg(   43.42) Peak(   43.49) Min(   43.07)
  GB/s   Avg(   57.90) Peak(   57.99) Min(   57.42)


Opt: Elems/thrd 1, thrds/blk 64 - pk thru  98.4503 GB/s, 123.0628 GFlops
Opt: Elems/thrd 2, thrds/blk 64 - pk thru  95.3779 GB/s, 119.2224 GFlops
Opt: Elems/thrd 4, thrds/blk 64 - pk thru  97.0317 GB/s, 121.2896 GFlops
Opt: Elems/thrd 8, thrds/blk 64 - pk thru  94.1059 GB/s, 117.6324 GFlops

Opt: Elems/thrd 1, thrds/blk 128 - pk thru  99.5902 GB/s, 124.4878 GFlops
Opt: Elems/thrd 2, thrds/blk 128 - pk thru  96.2985 GB/s, 120.3731 GFlops
Opt: Elems/thrd 4, thrds/blk 128 - pk thru  94.9835 GB/s, 118.7293 GFlops
Opt: Elems/thrd 8, thrds/blk 128 - pk thru  96.1054 GB/s, 120.1318 GFlops

Opt: Elems/thrd 1, thrds/blk 192 - pk thru  98.2473 GB/s, 122.8092 GFlops
Opt: Elems/thrd 2, thrds/blk 192 - pk thru  95.6510 GB/s, 119.5637 GFlops
Opt: Elems/thrd 4, thrds/blk 192 - pk thru  96.5704 GB/s, 120.7129 GFlops
Opt: Elems/thrd 8, thrds/blk 192 - pk thru  95.4054 GB/s, 119.2568 GFlops

Opt: Elems/thrd 1, thrds/blk 256 - pk thru  98.5406 GB/s, 123.1758 GFlops
Opt: Elems/thrd 2, thrds/blk 256 - pk thru  95.8551 GB/s, 119.8189 GFlops
Opt: Elems/thrd 4, thrds/blk 256 - pk thru  96.5229 GB/s, 120.6536 GFlops
Opt: Elems/thrd 8, thrds/blk 256 -  N/A

Card peaked @ 68c whilst running the last block
I'll run the rest and see if I get any different results to last time

[Richard Haselgrove]

Hmmm. Yes, the notebook 280.26 WHQL driver is available too. And there's a special note:

For the best experience, NVIDIA strongly recommends installing the following DELL SBIOS update before installing the NVIDIA notebook reference driver.

Installing this driver without the latest Dell SBIOS may cause the system fan noise to become more noticeable.

Since my fan is running full-blast all day, every day (it's positively distracting when an MB task finishes, and it cools down and reduces the fan speed for 10 seconds or so), that doesn't worry me - but it's interesting that they mention it.

What's an SBIOS?

Oh, OK - it takes me to a "System BIOS" page. For mine, version A08, released August 1st 2011.  That's interesting, too. Might be something they discovered in Beta or pre-beta.

Later - updated the BIOS, installed 280.26 Here's the full set of ChirpTest results. Conditions are slightly different from last time - I ran these immediately after the reboot, before starting BOINC, so the CPU was effectively unloaded: last time, BOINC was running, and I just suspended GPU crunching. But each set is internally consistent. Temperatures seemed lower during the ChirpTest, perhaps because of that. Now I'm back to normal running, the GPU temperature is stable at 71C, much as it was before the upgrade.

[Jason G]

...Temperatures seemed lower during the ChirpTest, perhaps because of that. Now I'm back to normal running, the GPU temperature is stable at 71C, much as it was before the upgrade.

Thanks.  Still seems pretty warm to me for 16 GFlops, but I suppose it is a mobile chipset & those tend to be rated for higher temps IIRC.

In the meantime, my mother's new i5 w/560ti has come together well, and has been passing all Pre-Boinc/Lunatics stress & burn-in tests etc.  When I can eventually get Mum off it, we'll be able to compare a few of the issues...  [Have made sure it has the 280.26 WHQL driver just now]

...Only problem so far is that she chose all good gear with the intent of making it last for at least a few years, so now it completely hoses my wolfdale/480 machine, is much smaller, quieter, cooler & more efficient , so getting her off of it for long testing periods might be tricky (at least for a while).

Jason

[Richard Haselgrove]

...Only problem so far is that she chose all good gear...

She didn't have any, er, assistance in making that choice, did she - by any chance?

[Jason G]

She didn't have any, er, assistance in making that choice, did she - by any chance?

LoL, 'course she did... options above & below given.  She has expensive tastes & the Aussie dollar is high (or everywhere else is in the toilet)    ( The mistake was letting her play games on my machine beforehand, so naturally she wanted better, but quieter, smaller & prettier  )

[arkayn]

I just installed the new driver on my GTX560 machine, will see how it does with it.

I am coming back up from 270.48 that I installed trying to find the desktop bug I am still seeing.