391 isn't that old.
Sometimes drivers get slower as they get bigger.

A GTX 690 is compute capability 3.0. Driver versions 391.xx are cuda 9 capable. Upgrading the driver to 4xx.xx will be cuda 10 capable.

Maybe I'm wrong, but I think Cuda 8,9 and 10 will work on CC 3.0 cards like the GTX 690 as along as you enable CC 3.x support in the makefile.

https://en.wikipedia.org/wiki/CUDA

CUDA SDK 6.5 support for compute capability 1.1 – 5.x (Tesla, Fermi, Kepler, Maxwell). Last version with support for compute capability 1.x (Tesla)
CUDA SDK 7.0 – 7.5 support for compute capability 2.0 – 5.x (Fermi, Kepler, Maxwell)
CUDA SDK 8.0 support for compute capability 2.0 – 6.x (Fermi, Kepler, Maxwell, Pascal). Last version with support for compute capability 2.x (Fermi)
CUDA SDK 9.0 – 9.2 support for compute capability 3.0 – 7.2 (Kepler, Maxwell, Pascal, Volta)
CUDA SDK 10.0 – 10.1 support for compute capability 3.0 – 7.5 (Kepler, Maxwell, Pascal, Volta, Turing)

Now I was getting this this on windows 10 machine with Maxwell cards:


06/08/2019 8:51:33 AM | GPUGRID | Requesting new tasks for CPU and NVIDIA GPU
06/08/2019 8:51:34 AM | GPUGRID | Scheduler request completed: got 0 new tasks
06/08/2019 8:51:34 AM | GPUGRID | No tasks sent
06/08/2019 8:51:34 AM | GPUGRID | No tasks are available for Long runs (8-12 hours on fastest card)

When long runs were still available.

Long runs (8-12 hours on fastest card) 24 2,468 2.78 (0.44 - 19.86) 663

But my Kelper card windows 7 machine, just received cuda 6.5 WUs.

Now I was getting this this on windows 10 machine with Maxwell cards:


06/08/2019 8:51:33 AM | GPUGRID | Requesting new tasks for CPU and NVIDIA GPU
06/08/2019 8:51:34 AM | GPUGRID | Scheduler request completed: got 0 new tasks
06/08/2019 8:51:34 AM | GPUGRID | No tasks sent
06/08/2019 8:51:34 AM | GPUGRID | No tasks are available for Long runs (8-12 hours on fastest card)

When long runs were still available.

Long runs (8-12 hours on fastest card) 24 2,468 2.78 (0.44 - 19.86) 663

But my Kelper card windows 7 machine, just received cuda 6.5 WUs.

I had raised the minimum driver version required for cuda80 in order to avoid the previous problem. Now I reverted the change. Still, this level of ad-hocness is not sustainable; the next app may need updated drivers.

Now I was getting this this on windows 10 machine with Maxwell cards:


06/08/2019 8:51:33 AM | GPUGRID | Requesting new tasks for CPU and NVIDIA GPU
06/08/2019 8:51:34 AM | GPUGRID | Scheduler request completed: got 0 new tasks
06/08/2019 8:51:34 AM | GPUGRID | No tasks sent
06/08/2019 8:51:34 AM | GPUGRID | No tasks are available for Long runs (8-12 hours on fastest card)

When long runs were still available.

Long runs (8-12 hours on fastest card) 24 2,468 2.78 (0.44 - 19.86) 663

But my Kelper card windows 7 machine, just received cuda 6.5 WUs.

I had raised the minimum driver version required for cuda80 in order to avoid the previous problem. Now I reverted the change. Still, this level of ad-hocness is not sustainable; the next app may need updated drivers.

It seems to be working well, right now. Both my machines received WUs.

Just, tell us what we need to upgrade, and which cards will be supported and not supported.

.Just, tell us what we need to upgrade, and which cards will be supported and not supported.

Yes please. I have 2 machines I converted back to windows after the linux issue. But I haven't updated the drivers as they were working. If there is a minimum driver version that you would require please let us know and we can make that happen. On my linux machines I run the latest drivers as my other project requires cuda 10.1 to run on them.

---

Now I was getting this this on windows 10 machine with Maxwell cards:


06/08/2019 8:51:33 AM | GPUGRID | Requesting new tasks for CPU and NVIDIA GPU
06/08/2019 8:51:34 AM | GPUGRID | Scheduler request completed: got 0 new tasks
06/08/2019 8:51:34 AM | GPUGRID | No tasks sent
06/08/2019 8:51:34 AM | GPUGRID | No tasks are available for Long runs (8-12 hours on fastest card)

When long runs were still available.

Long runs (8-12 hours on fastest card) 24 2,468 2.78 (0.44 - 19.86) 663

But my Kelper card windows 7 machine, just received cuda 6.5 WUs.

I had raised the minimum driver version required for cuda80 in order to avoid the previous problem. Now I reverted the change. Still, this level of ad-hocness is not sustainable; the next app may need updated drivers.

What is special with 4xx drivers?

418 has cuda 10.1 as well as support for Turing cards

---

Thats nice but that doesn't change anything for a CUDA80 app or older cards like the 690 mentioned above.

Unless there is a specific CUDA10 command in the new version then drivers that are older should still be supported. It's like calling something an AVX512 app and limiting to only CPUs with AVX512 even though only SSE2 is used.

It's not like people are asking for pre Kepler drivers to be supported. Just last years.

The 390 series drivers are Legacy drivers now. Minimal support. But the 390 series is still the default repository version offered in many distros.

This Nvidia document shows the changes from CUDA backwards compatibility to CUDA forwards compatibility. It also shows CUDA level compatibility with the various generations of GeForce cards.

https://docs.nvidia.com/deploy/cuda-compatibility/

I agree it may be difficult to maintain support for the older hardware in the future.

Hi:

My laptop is eagerly awaiting for some more GPUGRID work units.

COME ON DOWN!!!

Clive

Hi:

My laptop is eagerly awaiting for some more GPUGRID work units.

COME ON DOWN!!!

Clive

Hi,
as a lot of us, you need an alternative project (plane B) to satisfy your Nvidia gpu cards.
Be patient.
K.

---

Dreams do not always come true. But not because they are too big or impossible. Why did we stop believing.
(Martin Luther King)

Hi:
Thanks for your reply.
GPUGRID.net is the only Bonic project that runs on GPU cards that I am aware of. Are there other BONIC GPU projects that I can use as a plan B?
Clive

Are there other BONIC GPU projects that I can use as a plan B?

yes, there are.
Seti@Home, MooWrapper, Einstein, and several more.
Just look up the BOINC projects list, there is the pertinent information for each of the projects.

Here are the ones I like, somewhat in order of preference and if they have work.

Milkyway@home Physical Science Astronomy Rensselaer Polytechnic Institute

Einstein@home Physical Science Astrophysics University of Wisconsin - Milwaukee (USA), Max Planck Institute for Gravitational Physics - Hanover (Germany)

SETI@home Physical Science Astrophysics, astrobiology University of California, Berkeley

Enigma@Home Mathematics, computing, and games Cryptography Independent

in a way it's too bad that GPUGRID doesn't offer enough tasks for crunching by us volunteers.
The computational power interested in and available for GPUGRID is much higher than they make use of.
Which is a pitty.

in a way it's too bad that GPUGRID doesn't offer enough tasks for crunching by us volunteers.
The computational power interested in and available for GPUGRID is much higher than they make use of.
Which is a pitty.

I used to have one and occasionally two waiting work units for every long or short gpugrid WU. Now all I have are Einstein.

Just a matter of time before they are in the gridcoin graylist. I could start a poll to see if they should be kicked out but it would cost me GRC to start it. I will let someone else initiate it.

Just a matter of time before they are in the gridcoin graylist. I could start a poll to see if they should be kicked out but it would cost me GRC to start it. I will let someone else initiate it.

This is quite futile to threaten this project with putting it on the graylist at GridCoin. This wouldn't result in plenty of workunits. As of yet I would vote against that, though it would cost me some GRC. Please be patient.

The scientists at GPUGrid don't care about Gridcoin whitelist/greylist. They are operating on their schedule that produces science in a timely manner. Just because there are limited WUs doesn't mean they aren't pumping out studies. Changing GPUGrid to the greylist would accomplish nothing.