Author |
Message |
|
Is there a list of GPUs with a corresponding compute capability? I've been looking for one and can't find any. I had a 9800gx2 that was compute capable of 1.1. Thought the 280 was highter. |
|
|
skgivenVolunteer moderator Volunteer tester
Send message
Joined: 23 Apr 09 Posts: 3968 Credit: 1,995,359,260 RAC: 0 Level
Scientific publications
|
The GTX 280 is CC1.3.
I see you are using driver driver: 19745 and Boinc version 6.10.43.
In Boinc Manager (Advanced View) under Messages, the 13th line down should say CC1.3.
If it does not, report back after you restart your computer. |
|
|
|
I restarted my computer and it still shows the task running at 1.0, but in line 13 it shows cc1.3 as you said it should |
|
|
|
I updated the driver to the latest one from nVidia to 257.21, rebooted and it is still computing at 1.0 capability. What else can I try? |
|
|
|
I updated to the latest version of BOINC, it did benchmarks and I'm still only computing at 1.0... I don't know what else to do. |
|
|
|
I restarted my computer and it still shows the task running at 1.0, but in line 13 it shows cc1.3 as you said it should
Where are you reading that figure from? It's obviously bugging you, but I'd like to hear your description before I post a silly theory.... |
|
|
skgivenVolunteer moderator Volunteer tester
Send message
Joined: 23 Apr 09 Posts: 3968 Credit: 1,995,359,260 RAC: 0 Level
Scientific publications
|
Where does it say Compute Capable 1.0? |
|
|
|
It shows it in the Status... (0.33 CPUs + 1.00 NVIDIA GPUs) with my 9800gx2 it showed 1.1 where it now has 1.00. Is this not where it shows how it's computing? I understand line 13, just thought it would also show in the status. |
|
|
skgivenVolunteer moderator Volunteer tester
Send message
Joined: 23 Apr 09 Posts: 3968 Credit: 1,995,359,260 RAC: 0 Level
Scientific publications
|
That just means it is using one GPU (graphics card)!
Perhaps your memory is shady WRT 1.1, it happens :)
The 9800GT was CC1.1. Only G80 based GPUs were CC1.0 and those cards no longer work here.
So don't worry, everything is fine. |
|
|
|
So the 9800gx2 was a 1.1 card, and it showed as such in the "status" column. So theoretically it should be showing as 1.30 and not 1.00, but just be wrong?
Let's say that it's showing wrong. Does that mean it is computing the way it's suppose to be? What would be the effect of running a 1.3 card at 1.0? That's all I'm wondering. |
|
|
|
No, the status column does not display the compute capability.
The status column does indicate the proportion of the GPU's time that BOINC has allocated to the task. 1.0 = 100%. BOINC's knobs don't go up to 11.
To say that BOINC has allocated 100% of the card to a task doesn't imply that the task uses the card at 100% efficiency, either. You'd need a GPU monitoring tool like GPU-Z (or EVGA Precision?) to monitor that.
Have you tried looking at GPUGrid's stderr_txt for one of your completed tasks on this website? There's some information in there (though not as much as I remember). That tells you that the card is being detected and used as planned. |
|
|
|
Curious question then... What did it mean when it said (... + 1.1 NVIDIA GPUs) if 1.00 is 100% That's what it did say with my 9800gx2
I have a handy sidebar gadget that shows the gpu temp, pcb temp, VPU and GPU usage, ram, speed... all of the graphics card. http://blog.orbmu2k.de/ The one I'm using is GPU Observer.
My GPU never goes much over 70%
I'll look to see if I can find the stderr_txt file... |
|
|
|
Curious question then... What did it mean when it said (... + 1.1 NVIDIA GPUs) if 1.00 is 100% That's what it did say with my 9800gx2
I have a handy sidebar gadget that shows the gpu temp, pcb temp, VPU and GPU usage, ram, speed... all of the graphics card. http://blog.orbmu2k.de/ The one I'm using is GPU Observer.
My GPU never goes much over 70%
I'll look to see if I can find the stderr_txt file... |
|
|
skgivenVolunteer moderator Volunteer tester
Send message
Joined: 23 Apr 09 Posts: 3968 Credit: 1,995,359,260 RAC: 0 Level
Scientific publications
|
You probably just dont remember it like it was!
Under Boinc Manager Tasks, it would have said “Running (0.30GPUs + 1.00 NVIDIA GPUs (device 1)”
and under Messages it would describe the 9800GX2 as being CC 1.1.
PS. That was a dual card.
Something similar to this,
23/12/2009 18:08:49 NVIDIA GPU 0: GeForce 9800GTX (driver version 19621, CUDA version 2000, compute capability 1.1, 512GB, 407 GFLOPS peak)
23/12/2009 18:08:49 NVIDIA GPU 1: GeForce 9800GTX (driver version 19621, CUDA version 2000, compute capability 1.1, 512GB, 407 GFLOPS peak)
On this web site you can also see a bit of info, but as Richard said, this has changed; there is less than there use to be.
Coprocessors NVIDIA GeForce GTX 280 (1005MB) driver: 25721
<core_client_version>6.10.56</core_client_version>
<![CDATA[
<stderr_txt>
# Using device 0
# There is 1 device supporting CUDA
# Device 0: "GeForce GTX 280"
# Clock rate: 1.30 GHz
# Total amount of global memory: 1054539776 bytes
# Number of multiprocessors: 30
# Number of cores: 240
Anyway, your CC1.3 GTX280 card is working well.
Nice NVIDIA Inspector Tool – Version 1.90 link by the way.
http://blog.orbmu2k.de/ |
|
|
Beyond Send message
Joined: 23 Nov 08 Posts: 1112 Credit: 6,162,416,256 RAC: 0 Level
Scientific publications
|
It shows it in the Status... (0.33 CPUs + 1.00 NVIDIA GPUs)
What's really troubling is that your CPU is only compute capability .33 ----- JK :-)
Seriously, everything's fine... |
|
|
|
haha! my cpu is slow... It's OC'd from 2.6 to 3.15. It does pretty well.
Something I cannot figure out is... What are the main factors to consider when looking for a great card for crunching numbers? If you compare the 9800gx2 to the GTX 295 co-op for example. The gx2 has a higher gpu clock (600vs576), higher shader clock (1500vs1242), equal memory clocks. The gtx 295 has more memory interface (448bit/gpu vs 256bit/gpu), more processor cores (480vs256), more memory bandwidth (224 GB/s vs 128), more texture filling units (160 vs 128)
The 295 is slower, but seems to have more stuff doing the work. It seems I've seen some who had a lower RAC than me with their 295 wehn I had my 9800gx2, unless they didn't have the co-op edition...I just realized that. So what's the conclusion? I'm just fooling around and don't know whether to get a 295 co-op, or another 280 to pair with what I have.... |
|
|
skgivenVolunteer moderator Volunteer tester
Send message
Joined: 23 Apr 09 Posts: 3968 Credit: 1,995,359,260 RAC: 0 Level
Scientific publications
|
The Compute Capability is very important as are the number of shaders and their speed. CC1.2 is much better than CC1.1 and CC1.3 is slightly better than CC1.2
As for Fermi cards GPUGRID is in the process of moving from CUDA 3000 to CUDA 3010 (3.1). The present Fermis are Compute Capable 2.0.
I would suggest you keep your existing GTX280 and think about adding another CC1.3 card (GTX260-sp216, GTX275, GTX280, GTX285 or GTX295), for compatibility reasons; it better not to mix and match cards with different Compute Capabilities. However, the power requirements of having two large GPU's is an important issue. Can your PSU support two large cards?
You might also consider selling the card you have and getting one or two GTX460's. These are due for release in a week or two, so hold off until somone here has tried one and reported its performance findings. |
|
|