Latest drivers (NVidia and ATI)

[chelski]

I think you'll have problems. The only time I got a 256Mb GFX card to work was on Windows XP 32. It was also running at 1280x1024 as the screen I had was only a 17" CRT at the time. The screen modes for Vista and Win 7 seem to require more memory, and not leave enough for the app to use...
Giz.

Thanks.  Guess there's nothing much that can be done about it, after playing around with various settings and Cuda version, it is a pity that the GPU have to remain idle

[TouchuvGrey]

Can anyone tell me if the newest NVidia driver ( 258.96 ) will
let my GT220 and GTS250 cards crunch faster ? Currently
using  197.45 

[ndblaikie]

From the post on the Seti forums at Berkeley, it seems that a lot of people have noticed a small decrease in crunch time for work units. So yes it should.
http://setiathome.berkeley.edu/forum_thread.php?id=60815

[Questor]

I have just upgraded one Windows XP 64 PC (2 * GTX470) to using 258.96 drivers and noticed an odd behaviour.

Some CUDA Fermi tasks start normally, high CPU while loading data until they reach about 87MBytes of Windows memory and then drop down to 0% CPU utilisation. Elapsed and completion time then starts clocking up but progress remains at 0. Normal tasks use about 100Mbytes.

I have found a couple of tasks after a few hours like this - proably as a result of increased rsc_fpops_bound to overcome -177 errors.

Suspending the tasks allows a new task to start OK and progress so not a GPU problem. Unsuspending the task allows it to reload and it still sits there doing nothing.

This might have happened before the driver upgrade but thought I'd ask before dropping back to a previous version to see if it still occurs. I mainly upgraded to this version as it was highly recommended in the 0.37 Unified Installer readme.

I want to try out the new installer to try out the new CUDA app - might do that first to see if it behaves the same way.


John.

Edit: Changed to task to run on CPU and it errored out after a few seconds with a -9 result.
Shortly after another GPU task stopped with just over 8MBytes RAM used and no progress. Have also changed to CPU but not run yet.

Edit2: Installed 0.37 UI and allowed previous WU to rerun as CUDA task. The Lunatcs CUDA app errored out after about 30 seconds with a -12 error rather than hanging like the stock cuda_fermi app.
Second hung task also completed sucessfully with no errors (both running x32f preview)

1   21ap10aa.3011.1703.10.10.0
2   21ap10ae.28533.18063.7.10.171

[Fredericx51]

I noticed an unexpected behavior from my GTX470! (QX9650+GTX470),  since memory-- and Cntr-- and GPU-LOAD, are low, I let  run 2 SETI MB (VLAR also), but now I'm seeing it runs a third tasks, a GPUGrid WU, also using the 470.
Running WIN XP64, BOINC 6.10.58, 'f' version of FERMI app. and SSSE3 optimized MB.
Will keep an eye, on this, OTOH, heard from several people, running 3 MB task's on 1 FERMI, isn't
 giving troubles and task's.

Hmm,  a second look reveils after 1 (or 2) second's,  both SETI MB tasks are pauzed, considering
necessary GPU--; Memory-- and MemCntr- Load ,  gets way too high, also too much difference between app.'s.
Anyway the Lunatics, made an excellent app, congratz to this great application.

So far no -9; or too much time passed, forgot the error-number ,  even at lower Angle Ranges
like 0.0106(AR), the 470 doesn't give problems.
I've the 257.21 drivers installed and CUDA 3.1. (3010)

[Questor]

I also had been running 2 tasks on the GTX470 using stcok app. After running the Lunatics app run for a day I have that alos happily running 2 tasks per GPU. (but not mixed with other project apps) Using the CUDA which came with the installer which says 3.0.14.

No repeat so far of the hung tasks and no -12s yet.

As you say, hat's off to the developers!


John.

[_heinz]

Hi,
its no problem to running 4 tasks parallel on GTX470,
with latest unified installer.
change 0.5 to 0.25 does it
mine GTX470 in the V8-Xeon runs it already sucessful.

heinz

[Questor]

Hi,
its no problem to running 4 tasks parallel on GTX470,
with latest unified installer.
change 0.5 to 0.25 does it
mine GTX470 in the V8-Xeon runs it already sucessful.

heinz

What O/S are you using? I am using Win XP 32 & 64. I tried 3 tasks but the more you run in parallel the longer each one takes so it was less productive.


John.

[_heinz]

Hi,
its no problem to running 4 tasks parallel on GTX470,
with latest unified installer.
change 0.5 to 0.25 does it
mine GTX470 in the V8-Xeon runs it already sucessful.

heinz

What O/S are you using? I am using Win XP 32 & 64. I tried 3 tasks but the more you run in parallel the longer each one takes so it was less productive.


John.

OS Vista 64Bit Ultimate.
<count>0.25</count>
4 need ~38 - 40 min, that is 9,5- 10 min per wu(depending on ar)
___________________
<count>0.33</count>
3 need ~30 min, thats 10 min per wu(depending on ar), give a gpu load of 95 to 99 %
___________________
<count>0.5</count>
2 need still 12-16, 20-21min, 33-34 min,that is 6-8, 10, 17 min per wu(depending on ar), give a gpu load of 91-98%, memory controller load=43-54%
___________________
<count>1.0</count>
1 need 8-12 min, gpu load of ~68% memory controller load=34%
___________________
2 gives the min throughput time( ~8, 10, 17 min per wu, 180, 144, 84 wu's per day)
I will work now with 2 in parallel again, that seems the max output per day.
But throughput time is depending on ar, what sort of wu you get.
heinz

[M_M]

Can someone compare how fast is GTX460 comparing to GTX275 and GTS250 for SETI crunching? Is it worth upgrading?

I'm thinking of upgrading GTX275 (power hungry monster, over 220W) to faster and more power efficient GTX460 (around 160W), which I also expect to be faster (in games, it should be around 50% faster, but how much is it in SETI does anyone know?)

As I understand, SETI application is currently poorly optimized for Fermi architecture, am I right?

Thanks.

[Ghost0210]

Can someone compare how fast is GTX460 comparing to GTX275 and GTS250 for SETI crunching? Is it worth upgrading?

Can't speak for the 275 or 250, but compared to my 260, the 465 crunches 2 VHARs at a time in around 5mins 30. Compared to my 260 crunching 1 at a time in around 3mins 30

I'm thinking of upgrading GTX275 (power hungry monster, over 220W) to faster and more power efficient GTX460 (around 160W), which I also expect to be faster (in games, it should be around 50% faster, but how much is it in SETI does anyone know?)

As I understand, SETI application is currently poorly optimized for Fermi architecture, am I right?

Thanks.

It's not just Seti that is poorly optimised for Fermi, it's brand new so all developers need time to get used to what the cards can do and where their limits are.
The new v0.37 installer contains the latest fermi compatible build from the Lunatics. It's cut out a lot of the -12 errors (almost non-existent now ) and has some minor optimisations included.
Jason has already posted that it is going to to take time to make full use of the new cards, but given that time it should just get better and better.

[Josef W. Segur]

Can someone compare how fast is GTX460 comparing to GTX275 and GTS250 for SETI crunching? Is it worth upgrading?

I'm thinking of upgrading GTX275 (power hungry monster, over 220W) to faster and more power efficient GTX460 (around 160W), which I also expect to be faster (in games, it should be around 50% faster, but how much is it in SETI does anyone know?)

As I understand, SETI application is currently poorly optimized for Fermi architecture, am I right?

Thanks.

At SETI Beta, the Application details are now showing values for "Average processing rate" which are GFLOPS derived from actual MB task times, with the definition of a flop being based on the rsc_fpops_est values produced by the splitters. I was looking through the top hosts list there last week and gathered some of those values for a similar comparison, but not GTX275. Both GTX260 and GTX285 averaged about 170 running 6.09 cuda23, GTX460 about 250 runnning 6.10 cuda_fermi. I only found 5 of the 460s, 12 260s, and 13 285s so those numbers aren't very reliable. I also didn't make any attempt to judge whether some of the cards were overclocked.

There are some 275s there if you want to check when Beta's data-driven web pages are next available.
                                                                                       Joe

[Ghost0210]

Here's a link to how nVidia rate their current cards
http://www.nvidia.co.uk/object/graphics_cards_buy_now_uk.html
Not sure how accurate it is but may give you an idea

[Jason G]

Here's a link to how nVidia rate their current cards
http://www.nvidia.co.uk/object/graphics_cards_buy_now_uk.html
Not sure how accurate it is but may give you an idea

I'd say going from the cards on that list that I've used (more than a few), that it's a pretty reasonable ranking in terms of usage/gaming performance.  The newer cards at this stage are penalised a bit for crunching due to immature software,  so come out roughly equal IMO to the best of the prior gen, bit will slowly pull ahead.

[Brodo]

Here's a link to how nVidia rate their current cards
http://www.nvidia.co.uk/object/graphics_cards_buy_now_uk.html
Not sure how accurate it is but may give you an idea

Interesting, I'm running a GTX285 and a GTX470 in the same box using V0.37. with no overclock I'm finding that the 285 is quicker by roughly a minute on a "standard" 0.4AR unit.
~11 mins for the 470 as against ~10m for the 285