Way back late July there was word of a new app for Pascal to take advantage of the new GTX 10x0 GPUs once Cuda 8 was released.

Afaik Cuda 8.0 came out September 28, 2016 (14 days ago) but I can't seem to find anything clear to say that the original app tested with Cuda 8.0 RC is going to be rolled out, or if a new app is being developed to replace it altogether. Given we're mid October and the year is rolling to a close soon, it'd be handy to know what's happening or planned.

It'd be great if someone could have a sticky - Moderator only GPUGRID App Status thread that lets everyone know what's current version, what's in testing, blockers (from stopping rollout) etc. as this would greatly assist in diseminating info out to all those who are treading water with Pascal GPUs (I've 2x Titan X Pascal GPUs myself sitting idle) waiting for a suitable Pascal compatible app release.

Here's the Admin and Moderator messages stating that they're waiting for NVidia and CUDA v8.5:

https://www.gpugrid.net/forum_thread.php?id=4341&nowrap=true#44342

http://www.gpugrid.net/forum_thread.php?id=4341&nowrap=true#44652

Seriously though the project can't even provide enough work to keep the current GPUs busy...

Seriously though the project can't even provide enough work to keep the current GPUs busy...

That's not true. There was no significant workunit shortage in the past months. However there was just enough to feed the active hosts.
Ironically, since your post Stefan announced the CASP batch, which will provide plenty of short and long workunits as well.

Seriously though the project can't even provide enough work to keep the current GPUs busy...

That's not true. There was no significant workunit shortage in the past months. However there was just enough to feed the active hosts.
Ironically, since your post Stefan announced the CASP batch, which will provide plenty of short and long workunits as well.

Actually there was. It's good to have the CASP WUs. I've had 3 so far. On a modest GTX 670, the short 1ns took only 7 minutes. The longest type (50ns) took 6 hours.

7 minutes on a gtx 670.. Wow that's pretty impressive! How long do the long tasks take you on the gtx 670?

By long tasks I mean long runs.

---

Cruncher/Learner in progress.

I'm having a hard time trying to understand. Unfortunately programming isn't in my CV, I simply work in Freight and Distribution.

I thought with the release of Cuda 8, that work previously carried out on a pascal compliant version would hit production and that through the evaluation of Cuda 8, Pascal and the Titan X (P) / GTX 1080 / 1070 would lead to ease of programming for the lower end (1060 / 1050) when they came out.

Given there's no news whatsoever on Cuda 8.5 (Beta / RC / Release), what is it that's in Cuda 8.5 that's not in Cuda 8? Is it because of Visual Studio / Cuda bugs? I'm assuming there's still massive underlying issues in the code that don't allow a Pascal GRID App to be released / tested.

Forgive my ignorance, just trying to get a feel for things as a whole.

Just feel kinda helpless having 22TFLOPS of GPU grunt sitting here idle (as I'm sure many others do too). I was hoping to upgrade my setup to incorporate several externally linked GTX1080 / Titan GPUs to assist with this project (and also double as an Iray rendering server for me) definitely going ahead with that plan if we can get a steady flow of Pascal WUs happening at some stage.

7 minutes on a gtx 670.. Wow that's pretty impressive! How long do the long tasks take you on the gtx 670?

By long tasks I mean long runs.

SDOERR_CASP times:
GTX 670: 1ns about 7 minutes, 5ns about 35 minutes, 20ns about 2.5 hours, 50 ns about 6 hours.
GTX 750Ti: 1ns about 9.5 minutes, 5ns about 47 minutes, 20ns about 3 hours, 50ns about 8.5 hours.

Given there's no news whatsoever on Cuda 8.5 (Beta / RC / Release), what is it that's in Cuda 8.5 that's not in Cuda 8? Is it because of Visual Studio / Cuda bugs? I'm assuming there's still massive underlying issues in the code that don't allow a Pascal GRID App to be released / tested.

Apparently one has to join the Accelerated Computing Developer Program of Nvidia to get access to the next CUDA Toolkit version.
https://developer.nvidia.com/accelerated-computing-developer

That is why we don't find any information in the web so far. Upcoming specs, features and hotfixes of Version 8.5 are provided to registered members only. That would be my guess.

Just feel kinda helpless having 22TFLOPS of GPU grunt sitting here idle (as I'm sure many others do too). I was hoping to upgrade my setup to incorporate several externally linked GTX1080 / Titan GPUs to assist with this project (and also double as an Iray rendering server for me) definitely going ahead with that plan if we can get a steady flow of Pascal WUs happening at some stage.

+1 Same here. One 1070 and one 1080 sitting idle. But it can't be helped.

---

I would love to see HCF1 protein folding and interaction simulations to help my little boy... someday.

Just feel kinda helpless having 22TFLOPS of GPU grunt sitting here idle (as I'm sure many others do too).

Unfortunately I can't answer any of your questions, but would be interested in the answers too. My GTX1060 was doing POEM, which I consider a good substitute for GPU-Grid. However, they just quit because they've calculated everything they needed BOINC for by now.. so I wnet to Folding@Home. It doesn't count towards BOINc credits, but it works very well and is (IMO) good science, which might appeal to anyone interested in GPU-Grid too.

BTW: this thread is about the Python Runtime. The "GPU-Grid on Pascal" thread is currently right next to it in the news forum.. please continue the discussion there.

MrS

---

Scanning for our furry friends since Jan 2002

Would love to continue there, but the thread is locked......

Sorry to steer the conversation away from it's subject.

Unlocked.

MrS

---

Scanning for our furry friends since Jan 2002

Three months are gone and I never saw a WU.
Any news?

Three months are gone and I never saw a WU.
Any news?

This project is CUDA based, so you can use NVidia cards here.
There is an AMD app, but the work for these cards is scarce.

Three months are gone and I never saw a WU.
Any news?

All of your 100m credit is from the 680s which are nvidia cards. The AMD portion of this project very limited

Hello,
I'm aware that gpugrid runs on CUDA. And I know that there's no WU for AMD cards.
But, this thread is about a cpu application that *maybe* can evolve in a OpenCL gpu version.

The application is linux-only and I no longer use nvidia cards: so, why not try to help with the cpu of my linux box?

Cheers.

After 7 months without any WU, my processor is tired of waiting.
Project suspended, see you.

I also have some AMD cards that I would love to throw on this project.

And once again...........The project is wasting my time and money. 100% failure rate is not acceptable.

Just wanted to start the weekend early?

---

And once again...........The project is wasting my time and money. 100% failure rate is not acceptable.

I have negligible failure rate, and I think others have same experiences, in the other case this forum would be flooded by complaints like yours (just as it happened earlier). It is easier to help you if you specify your failures more precisely (especially when you have multiple hosts). Perhaps while you are doing it you could even find the clue yourself.
Now I've checked your tasks on your hosts.
I think you are referring to your host #189416 (i7-970, 3x GTX 770). All of the failed tasks come from GPU 1 (the middle card). This card reaches 80°C (176°F), but there are tasks which fail on that card even at 74°C (165°F). It could be that this card has some kind of failure. I would separate this card, and test it with different tools like furmark. I would also check the all power connectors (PCIe, CPU, 24-pin MB) for burn marks in this PC. The topmost card in this host reaches 87°C (189°F), which is too high.
Your GTX 780Ti's are also quite hot (and have some failures) in your other hosts.
If you see

# The simulation has become unstable. Terminating to avoid lock-up (1)

messages in the stderr output of your tasks after a couple minutes of crunching, these are most probably the sign of a too hot, too much overclocked, or other way failing card (stripes or other artifacts on the screen); in some cases this is a sign of a failing PSU.
If you see this message in the output of a workunit after 5000 steps (= it fails immediately or in a very short time after start), this is a misconfigured workunit, but in this case not much of your resources are wasted.

After 7 months without any WU, my processor is tired of waiting.
Project suspended, see you.

I hate being right all the time...