cuda app windows is up + 500 more WUs

I already have Microsoft Visual C++ 2010 Redistributable installed. (both x86 and x64)
I also have these other versions installed -> 2005, 2008, 2012, 2013 and 2017

We could test this standalone.
download http://srbase.my-firewall.org/sr5/download/mfakto-win-v1.zip, copy this
Factor=3310968337,68,69
into the worktodo file and start the exe file.
At the moment I have less nvidia tester and it looks like only this version has only invalid results.

I am happy to help if GTX950 is sufficient.

I will add the cuda linux app later, there is one unclarified thing. The GPU WUs limit is 3 if its working.

More tests WUs added, one thing, enable nvidia and AMD in your prefs, if AMD is only enable you will not get work (maybe a bug in the older server software)

conclusion:

The error rate is very low, only see:

Linux cuda and mac opencl apps are up, 2000 test WUs will follow.

All my WU's finishes (validated) in ~ 6sec on 1660Ti and all have:
no factor for M3310968337 from 2^68 to 2^69 [mfaktc 0.21 barrett76_mul32_gs]
tf(): total time spent: 1.748s

ERROR: get_next_assignment(): no valid assignment found in "worktodo.txt"
09:38:29 (8096): mfaktc-win-64.exe exited; CPU time 0.015625
09:38:29 (8096): called boinc_finish(0)

All my WU's finishes (validated) in ~ 6sec on 1660Ti and all have:
no factor for M3310968337 from 2^68 to 2^69 [mfaktc 0.21 barrett76_mul32_gs]
tf(): total time spent: 1.748s

ERROR: get_next_assignment(): no valid assignment found in "worktodo.txt"
09:38:29 (8096): mfaktc-win-64.exe exited; CPU time 0.015625
09:38:29 (8096): called boinc_finish(0)

The current test looks good so far, this error at the end is only looking for more work and I cant get rid of this, the CUDA apps are much faster than AMD opencl.

Update:
The linux cuda app cant find the app itself, need more investigations.

What if you loaded worktodo.txt file with more than one entry? They are very short at the moment.

Linux app updated. I have changed a parameter.

conclusion:

100% success Windows CUDA / opencl
100% failed linux v1
linux v2 and Mac doesnt have an entry yet


Next test: 10 WUs for linux v2 only

10 WUs for Linux / Mac are ready to catch.

Linux v2 also failed, maybe a missing cuda lib, the rest looks ok in setup

Next test is prepared:

- checkpoint test (every 300s)
- runtime ~30min@RX5500XT
- 1500 credits
- 100 test WUs
- 1 GPU WU per host limit

If you are able to run only this project in BOINC, restart the client after you see a ckp file.

Note:

The wrapper is the same which SRBase is using so the runtime is not saved and will start from 0. Check the stderr.txt in slotfolder if you see a last restart point.

After a restart you will find this in your stderr.txt:

Found a valid checkpoint file.
last finished class was: 647

In my case its working.