Cuda100 vs Cuda 120 Wus

Author	Message
Sabroe_SMC Send message Joined: 10 Jan 15 Posts: 3 Credit: 521,264,195 RAC: 810,627	Message 9220 - Posted: 31 Oct 2023, 18:41:02 UTC
I dont want to have Cuda100 Wus on my GTX 2080ti. But how can i realize that? They are roughly 30 sec slower than the Cuda120 Wus. If I delete the Cuda100 Wus including the application and log file then of course all Cuda100 Wus abort with calculation errors. But there is not a single Cuda120 Wu, only Cuda100 Wus. Why is this the case and what should/can I do about it? ____________
ID: 9220 · Rating: 0 · rate: / Reply Quote

rebirther Volunteer moderator Project administrator Project developer Project tester Project scientist Send message Joined: 2 Jan 13 Posts: 7479 Credit: 43,657,867 RAC: 40,735	Message 9221 - Posted: 31 Oct 2023, 19:01:15 UTC - in response to Message 9220. Last modified: 31 Oct 2023, 19:01:38 UTC
I dont want to have Cuda100 Wus on my GTX 2080ti. But how can i realize that? They are roughly 30 sec slower than the Cuda120 Wus. If I delete the Cuda100 Wus including the application and log file then of course all Cuda100 Wus abort with calculation errors. But there is not a single Cuda120 Wu, only Cuda100 Wus. Why is this the case and what should/can I do about it? Its a plan_class thing which the server controlled. Perhaps driver 525.60 could help not to decide to switch or an app_config could help. Here is a snipped of the plan_class_spec: <plan_class> <name>cuda100</name> <gpu_type>nvidia</gpu_type> <cuda/> <min_cuda_version>10000</min_cuda_version> <min_nvidia_compcap>300</min_nvidia_compcap> <max_nvidia_compcap>705</max_nvidia_compcap> <min_driver_version>41048</min_driver_version> <min_gpu_ram_mb>384</min_gpu_ram_mb> <gpu_ram_used_mb>384</gpu_ram_used_mb> <gpu_peak_flops_scale>0.22</gpu_peak_flops_scale> <cpu_frac>0.01</cpu_frac> </plan_class> <plan_class> <name>cuda111</name> <gpu_type>nvidia</gpu_type> <cuda/> <min_cuda_version>11000</min_cuda_version> <max_cuda_version>11090</max_cuda_version> <min_driver_version>45532</min_driver_version> <min_gpu_ram_mb>384</min_gpu_ram_mb> <gpu_ram_used_mb>384</gpu_ram_used_mb> <gpu_peak_flops_scale>0.22</gpu_peak_flops_scale> <cpu_frac>0.01</cpu_frac> </plan_class> <plan_class> <name>cuda112</name> <gpu_type>nvidia</gpu_type> <cuda/> <min_cuda_version>11000</min_cuda_version> <max_cuda_version>11090</max_cuda_version> <min_nvidia_compcap>800</min_nvidia_compcap> <max_nvidia_compcap>808</max_nvidia_compcap> <min_driver_version>46027</min_driver_version> <min_gpu_ram_mb>384</min_gpu_ram_mb> <gpu_ram_used_mb>384</gpu_ram_used_mb> <gpu_peak_flops_scale>0.22</gpu_peak_flops_scale> <cpu_frac>0.01</cpu_frac> </plan_class> <plan_class> <name>cuda120</name> <gpu_type>nvidia</gpu_type> <cuda/> <min_cuda_version>12000</min_cuda_version> <max_cuda_version>13000</max_cuda_version> <min_nvidia_compcap>600</min_nvidia_compcap> <min_driver_version>52560</min_driver_version> <min_gpu_ram_mb>384</min_gpu_ram_mb> <gpu_ram_used_mb>384</gpu_ram_used_mb> <gpu_peak_flops_scale>0.22</gpu_peak_flops_scale> <cpu_frac>0.01</cpu_frac> </plan_class> Its a hard trial and get thing to get all cards under control, too many drivers and cuda versions. If we can make things better Iam always open for better optimisations.
ID: 9221 · Rating: 0 · rate: / Reply Quote

Sabroe_SMC Send message Joined: 10 Jan 15 Posts: 3 Credit: 521,264,195 RAC: 810,627	Message 9222 - Posted: 31 Oct 2023, 20:46:29 UTC
I am running Nvidia driver 545.84 under Win 10 with newest updates. Can you write me an app_config that only works with Cuda120 WUs on my computer? I dont think so. So i will wait and waist my electricity for your project. This is the host: https://srbase.my-firewall.org/sr5/show_host_detail.php?hostid=226942 I looked for other users with the same problem of mine. There are wastings of electricity for a whole bunch of computers.
ID: 9222 · Rating: 0 · rate: / Reply Quote

rebirther Volunteer moderator Project administrator Project developer Project tester Project scientist Send message Joined: 2 Jan 13 Posts: 7479 Credit: 43,657,867 RAC: 40,735	Message 9223 - Posted: 31 Oct 2023, 21:14:15 UTC - in response to Message 9222.
I am running Nvidia driver 545.84 under Win 10 with newest updates. Can you write me an app_config that only works with Cuda120 WUs on my computer? I dont think so. So i will wait and waist my electricity for your project. This is the host: https://srbase.my-firewall.org/sr5/show_host_detail.php?hostid=226942 I looked for other users with the same problem of mine. There are wastings of electricity for a whole bunch of computers. Sorry to hear that. I made a change on the plan_class_spec which cuda100 should only be for cards with max compute cap 6.0. 7.5+ is for your card but need cuda12. If this is not working I will write an app_config.
ID: 9223 · Rating: 0 · rate: / Reply Quote

rebirther Volunteer moderator Project administrator Project developer Project tester Project scientist Send message Joined: 2 Jan 13 Posts: 7479 Credit: 43,657,867 RAC: 40,735	Message 9224 - Posted: 31 Oct 2023, 21:34:49 UTC
<app_config> <app_version> <app_name>TF</app_name> <plan_class>cuda120</plan_class> <min_gpu_ram_mb>384</min_gpu_ram_mb> <gpu_ram_used_mb>384</gpu_ram_used_mb> <gpu_peak_flops_scale>0.22</gpu_peak_flops_scale> <cpu_frac>0.01</cpu_frac> </app_version> </app_config> untested
ID: 9224 · Rating: 0 · rate: / Reply Quote

Sabroe_SMC Send message Joined: 10 Jan 15 Posts: 3 Credit: 521,264,195 RAC: 810,627	Message 9227 - Posted: 1 Nov 2023, 9:45:32 UTC
Thx, rebirther I think its now ok. On the rig i get nowonly Cuda120 Wus. Thx again
ID: 9227 · Rating: 0 · rate: / Reply Quote

Author

Message

Sabroe_SMC
Send message
Joined: 10 Jan 15
Posts: 3
Credit: 521,264,195
RAC: 810,627
1M in Sierpinski / Riesel Base credit

5M in Sierpinski / Riesel Base - short credit

2M in Sierpinski / Riesel Base - long credit

2M in Sierpinski / Riesel Base - average credit

I dont want to have Cuda100 Wus on my GTX 2080ti. But how can i realize that?
They are roughly 30 sec slower than the Cuda120 Wus.
If I delete the Cuda100 Wus including the application and log file then of course all Cuda100 Wus abort with calculation errors. But there is not a single Cuda120 Wu, only Cuda100 Wus.
Why is this the case and what should/can I do about it?
____________

ID: 9220 · Rating: 0 · rate:

Reply Quote

rebirther
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist

Send message
Joined: 2 Jan 13
Posts: 7479
Credit: 43,657,867
RAC: 40,735
2M in Sierpinski / Riesel Base credit

1M in Sierpinski / Riesel Base - long credit

500k in Sierpinski / Riesel Base - average credit

I dont want to have Cuda100 Wus on my GTX 2080ti. But how can i realize that?
They are roughly 30 sec slower than the Cuda120 Wus.
If I delete the Cuda100 Wus including the application and log file then of course all Cuda100 Wus abort with calculation errors. But there is not a single Cuda120 Wu, only Cuda100 Wus.
Why is this the case and what should/can I do about it?

Its a plan_class thing which the server controlled. Perhaps driver 525.60 could help not to decide to switch or an app_config could help.

Here is a snipped of the plan_class_spec:

<plan_class> <name>cuda100</name> <gpu_type>nvidia</gpu_type> <cuda/> <min_cuda_version>10000</min_cuda_version> <min_nvidia_compcap>300</min_nvidia_compcap> <max_nvidia_compcap>705</max_nvidia_compcap> <min_driver_version>41048</min_driver_version> <min_gpu_ram_mb>384</min_gpu_ram_mb> <gpu_ram_used_mb>384</gpu_ram_used_mb> <gpu_peak_flops_scale>0.22</gpu_peak_flops_scale> <cpu_frac>0.01</cpu_frac> </plan_class> <plan_class> <name>cuda111</name> <gpu_type>nvidia</gpu_type> <cuda/> <min_cuda_version>11000</min_cuda_version> <max_cuda_version>11090</max_cuda_version> <min_driver_version>45532</min_driver_version> <min_gpu_ram_mb>384</min_gpu_ram_mb> <gpu_ram_used_mb>384</gpu_ram_used_mb> <gpu_peak_flops_scale>0.22</gpu_peak_flops_scale> <cpu_frac>0.01</cpu_frac> </plan_class> <plan_class> <name>cuda112</name> <gpu_type>nvidia</gpu_type> <cuda/> <min_cuda_version>11000</min_cuda_version> <max_cuda_version>11090</max_cuda_version> <min_nvidia_compcap>800</min_nvidia_compcap> <max_nvidia_compcap>808</max_nvidia_compcap> <min_driver_version>46027</min_driver_version> <min_gpu_ram_mb>384</min_gpu_ram_mb> <gpu_ram_used_mb>384</gpu_ram_used_mb> <gpu_peak_flops_scale>0.22</gpu_peak_flops_scale> <cpu_frac>0.01</cpu_frac> </plan_class> <plan_class> <name>cuda120</name> <gpu_type>nvidia</gpu_type> <cuda/> <min_cuda_version>12000</min_cuda_version> <max_cuda_version>13000</max_cuda_version> <min_nvidia_compcap>600</min_nvidia_compcap> <min_driver_version>52560</min_driver_version> <min_gpu_ram_mb>384</min_gpu_ram_mb> <gpu_ram_used_mb>384</gpu_ram_used_mb> <gpu_peak_flops_scale>0.22</gpu_peak_flops_scale> <cpu_frac>0.01</cpu_frac> </plan_class>

Its a hard trial and get thing to get all cards under control, too many drivers and cuda versions. If we can make things better Iam always open for better optimisations.

ID: 9221 · Rating: 0 · rate:

Reply Quote

Sabroe_SMC
Send message
Joined: 10 Jan 15
Posts: 3
Credit: 521,264,195
RAC: 810,627
1M in Sierpinski / Riesel Base credit

I am running Nvidia driver 545.84 under Win 10 with newest updates.
Can you write me an app_config that only works with Cuda120 WUs on my computer?
I dont think so. So i will wait and waist my electricity for your project.

This is the host: https://srbase.my-firewall.org/sr5/show_host_detail.php?hostid=226942

I looked for other users with the same problem of mine. There are wastings of electricity for a whole bunch of computers.

ID: 9222 · Rating: 0 · rate:

Reply Quote

rebirther
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist

Send message
Joined: 2 Jan 13
Posts: 7479
Credit: 43,657,867
RAC: 40,735
2M in Sierpinski / Riesel Base credit

I am running Nvidia driver 545.84 under Win 10 with newest updates.
Can you write me an app_config that only works with Cuda120 WUs on my computer?
I dont think so. So i will wait and waist my electricity for your project.

This is the host: https://srbase.my-firewall.org/sr5/show_host_detail.php?hostid=226942

I looked for other users with the same problem of mine. There are wastings of electricity for a whole bunch of computers.

Sorry to hear that. I made a change on the plan_class_spec which cuda100 should only be for cards with max compute cap 6.0. 7.5+ is for your card but need cuda12. If this is not working I will write an app_config.

ID: 9223 · Rating: 0 · rate:

Reply Quote

rebirther
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist

Send message
Joined: 2 Jan 13
Posts: 7479
Credit: 43,657,867
RAC: 40,735
2M in Sierpinski / Riesel Base credit

<app_config> <app_version> <app_name>TF</app_name> <plan_class>cuda120</plan_class> <min_gpu_ram_mb>384</min_gpu_ram_mb> <gpu_ram_used_mb>384</gpu_ram_used_mb> <gpu_peak_flops_scale>0.22</gpu_peak_flops_scale> <cpu_frac>0.01</cpu_frac> </app_version> </app_config>

untested

ID: 9224 · Rating: 0 · rate:

Reply Quote

Sabroe_SMC
Send message
Joined: 10 Jan 15
Posts: 3
Credit: 521,264,195
RAC: 810,627
1M in Sierpinski / Riesel Base credit

Thx, rebirther
I think its now ok. On the rig i get nowonly Cuda120 Wus.
Thx again

ID: 9227 · Rating: 0 · rate:

Reply Quote