Just installed GTX 480 with driver 197.41
All job cancelled after a few seconds.

Any ideas?

Ton van Born (ftpd)

---

Ton (ftpd) Netherlands

Have a look at this:

http://milkyway.cs.rpi.edu/milkyway/forum_thread.php?id=1620&nowrap=true#38664

Going to be a while before it will run, as code changes are needed, even then its performance will be poor due to the double precision limitation. You would be better off using a Fermi on a Single Precision Project

Regards
Zy

Unfortunate the cuda app has compute capability 1.3 hard coded into it, so the app will not find a valid card and exit. The behaviour is effectively the same as if you were running it on older G80 or G90 with no double precision support. The good news is that is it only 1 one line change to the code and a one line change to the Makefile to fix it. It will also require building against Cuda 3.0 and need 195 series drivers on linux or 196/197 series on Windows.

Normally if you do not have the right card Milkyway does not download any WU, so why now it downloads WU??

Ton

The milkyway application itself doesn't have any influence on when jobs are downloaded, the boinc client does that. I am going to guess that the boinc client contains a compute capability test like this:

if (CUDACapabilityMajorrevisionnumber >= 1) and (CUDACapabilityMinorrevisionnumber >= 3) then card is OK

whereas the milkway app has a test like this:

if (CUDACapabilityMajorrevisionnumber == 1) and (CUDACapabilityMinorrevisionnumber == 3) then card is OK

so that compute 2.0 cards are OK with the boinc client and not OK with the milkway app. The first first is a guess, because I haven't seen the boinc client code, but the second part is definitely right - you can see it in the code available for download here.

Correct, the code specifically asks for compute capability 1.3. I will have to look into the changes that compute capability 2.0 brought and determine whether or not they support the CUDA Framework 2.3.

Any idea when the change to CC 2.0 can take place?
My Fermi-cards are still waiting for WU.

Ton (ftpd)

I posted new applications on Milkyway@Home version 3, I only have a GTX 285 so I am not able to test whether it works or not.

I posted new applications on Milkyway@Home version 3, I only have a GTX 285 so I am not able to test whether it works or not.

hi testing no error so far:) 5.20min GTX480

Hmmm...slower than my 4850. What's it like at Collatz?

Hmmm...slower than my 4850. What's it like at Collatz?

hi:) 8.15min

Hmmm...slower than my 4850. What's it like at Collatz?

hi:) 8.15min

5 minutes faster than my 4870.

Hmmm...slower than my 4850. What's it like at Collatz?

hi:) 8.15min

5 minutes faster than my 4870.

1:05 slower than my 5830.

---

I posted new applications on Milkyway@Home version 3, I only have a GTX 285 so I am not able to test whether it works or not.

hi testing no error so far:) 5.20min GTX480

and:

hi:) 8.15min

To contrast I have a 5870 and the comparable times are >2Min at MW and ~5:44-6:00 at Collatz, and at a cost of a little over $400 a bit cheaper ...

Just food for comparison. I hate losing the ability to run other projects like GPU Grid, but the cost benefit ratio says ATI to me ... I can get a suite of 5870 cards to replace my older Nvidia cards as soon as I can ... and my production is going to go up a lot... and my power is staying the same or falling (depending on whose numbers you use) ...

In one system I am doing about 294K CS per day with a 5870 and GTX280 card... I had a system with a pair of GTX260 cards and it was doing about 72K a day ... it is too soon to know if it is going to go all the way up to 290K or not, would be nice, but even so, one card is likely to double that machine's production... my pair of 4870 cards do about 140K per day together which also is not bad at all ...

I will know better in a couple days when the stat sites start to crunch the data but here is hoping ... :)

The benchmarking I have done for the GTX470 shows it to be slightly better than twice as fast at double precision compared to a GTX275. The linear algebra benchmarks I use are generally memory bandwidth limited - a stock GTX275 hits about 77 Gflop/s double precision, and the GTX470 hits about 160 Gflop/s doing the same operation running identical code. There are new architectural features in Fermi which should allow that to improve further with some tuning.

On pure compute bound jobs (and Milkyway seems to be one of the few), Cypress has a considerable advantage. On memory bandwidth bound codes (or mixed single-double precision codes), the performance gap will be a lot smaller.

I hope to get a Telsa C2050 to test in the next week or so.

GTX480 - WU Milkyway 3 = OK - 5 min. 8 secs.
Is it possible just to download the new version 3 WU's, because all other WU
will cancel?
If possible also more WU and not just 12!!!

Ton (ftpd)

If you want to push the beast - there is a good review on o/c the 480 at Guru3d (they also did one for the 470, which was a little dubious in outcome).

They did not push further with the 480 or 470 due to the crazy heat and noise levels, but it gives an idea what it can do, albeit it was games orientated comparisons and therefore single precison as such.

3D Mark Vantage GTX480

Guru3d Concluding Remarks

Regards
Zy

I have a GTX 480 and also a GTX 470.

Not any problems with noise and heat. There are very much OK!

Ton (ftpd)

---

Ton (ftpd) Netherlands

At stock it will be "ok" - even NVidia are not going to field something that needs earplugs at stock. Although the 480 gets close to that in its guise as an egg-fryer.

They crippled the original design by killing of a shader cluster bringing it down from 512 to 480, because they had to shove more volts through the beast to make up for poor production quantities. So at current stock its "ok" at the price of some pretty lacklustre performance in comparison to the generation of card that its meant to be.

The point of the Guru3d review was o/c. The noise and heat generated at levels of o/c, where there should not be issues, was not acceptable to the point they did not go further.

In fact they explicitly stated they would not until some custom water cool solutions were available to bring heat and power under control.

Regards
Zy

The benchmarking I have done for the GTX470 shows it to be slightly better than twice as fast at double precision compared to a GTX275. The linear algebra benchmarks I use are generally memory bandwidth limited - a stock GTX275 hits about 77 Gflop/s double precision, and the GTX470 hits about 160 Gflop/s doing the same operation running identical code. There are new architectural features in Fermi which should allow that to improve further with some tuning.

And were they the only game in town I would be salivating over upgrading my 260/280/295 cards to the latest and greatest.

However, the numbers you post here for the 470 shows that it is twice as fast as my 4870 for Collatz (~16 min) it is more than twice as slow for MW (~3:15) and the 4870's are only about $150 US ...

In a way I regret my slow migration to the world of ATI because I really do like GPU Grid for example and there is no doubt that the CUDA applications are more common still ... but, we are up to 3 ATI projects with SaH looming on the horizon in having a viable 4th application "real soon now" to quote Jerry Pournell ...