1) Message boards : Number crunching : Windows 10 GPU meter can't see GPUGrid (Message 48329)
Posted 2 days ago by Profile Retvari Zoltan
The new GPU item(s) in Task Manager, lists usages for each "engine" that the GPU supports.

For my GTX 980 Ti GPUs, when I select one of them in Task Manager, there are 12 engines of activity to choose from. The CUDA compute activity actually DOES show up whenever I select the "Compute_0" engine for each GPU.
Thanks for that!

I'm not sure how the "Overall usage" value gets calculated for the left pane, but I suspect it excludes compute intentionally, in order to show that the GPU is otherwise available for non-compute-use.

Here is a link to some Windows 10 Feedback that I just submitted.
If you want Microsoft to change this, then please click the link and then click Upvote.



I've upvoted your feedback.
2) Message boards : Number crunching : Windows 10 GPU meter can't see GPUGrid (Message 48316)
Posted 3 days ago by Profile Retvari Zoltan
Turns out that Windows 10 now has (from a recent update, we think) a GPU usage meter in Task Manager.
This feature was rolled out to the genral public in the so called "creators fall update" (v1709).
Earlier versions do not have this feature in Task Manager.

But it can't see GPUGrid - and possibly general CUDA - activity.
True. It can't see Einstein@home OpenCL tasks too. It can only see the GPU memory usage.
If I recall it correctly it can see the iGPU usage though.
The other shortcoming of Task Manager is that it does not recognize DDR4 memory type (and some DDR3 memory type too), so it displays only the size of memory (or shows "other" as its type).

Just a heads-up in case anyone else gets confused - it's Microsoft, not you.
Slightly connected to this issue that Microsoft is working on a solution to eliminate the WDDM overhead of GPGPU applications by allowing GPGPU APIs to directly run kernels (that would make GPUGrid tasks 5-10-15% faster on Windows 7-10).
3) Message boards : Number crunching : Work Units???? (Message 48310)
Posted 4 days ago by Profile Retvari Zoltan
I use my PC as a heater during the winter and just started running again since it will be 45 tonight in Phoenix.
Me too. It is wise then to set up backup projects (=zero resource share set on the given project's computing preferences).

Are there no work units to be downloaded? It keeps saying "communications deferred" and it says there is one work unit ready to download. Is there not enough work???
It is intentional that there's no work in the queue, as this project needs very short turnaround times because the workunits are generated from the result of a previous workunit, so the queue is kept as short as possible (=0). There are always around 2200-2500 workunits in progress, which means that the staff continuously fills the queue with work, but it is instantly grabbed by some hosts.
4) Message boards : Graphics cards (GPUs) : Inside Volta (Message 48308)
Posted 4 days ago by Profile Retvari Zoltan
I still think that there's no point in buying TITAN series cards for crunching GPUGrid, as this series is overpriced so much that you can buy two gaming (GTX series) cards instead, and have 1.9 times crunching power for the same money.
5) Message boards : GPUGRID CAFE : The milestone thread (Message 48300)
Posted 5 days ago by Profile Retvari Zoltan
Just past 150M a couple of days ago!


Congrats to Zalster for passing over 1 Billion credits!
6) Message boards : Server and website : Problem with upload server? (Message 48299)
Posted 5 days ago by Profile Retvari Zoltan
Since last night, the upload server again is extremely slow. The upload speed of finished tasks is a fraction of what it normally is, sometimes it even comes to a halt. Uploads take forever.
I don't have any upload problems since last night. Usually it's only you reporting this slowness, so perhaps the error is on your ISP's side.
Try http://www.speedtest.net/; Open it with Microsoft Edge (or any flash enabled browser), and then you can select the location of the server you want to test your connection with. Practically you should select one in Barcelona. If the upload speed to those servers is slow too, then it's sure that GPUGrid has nothing to do about this.
7) Message boards : Number crunching : Unable to load module .mshake_kernel.cu. (702) (Message 48227)
Posted 19 days ago by Profile Retvari Zoltan
In case this hasn't been resolved, I've also run into GPUGRID tasks erring and found my solution to be increasing the virtual memory size. See here.
I don't think that increasing virtual memory size could fix such problems (perhaps indirectly by accident). Your PC has 32GB RAM. I can't imagine that even if you run 12 CPU tasks simultaneously it will run out of 32GB (+1GB virtual). If it does, then some of the apps you run have a memory leak, and it will run out even if you set +4GB or +8GB virtual memory.
SWAN : FATAL : Cuda driver error 999 in file 'swanlibnv2.cpp' in line 1965.
SWAN : FATAL Unable to load module .mshake_kernel.cu. (719)
errors are the result of the GPUGrid task gets suspended too frequently and / or too many times (or a failing GPU).
EDIT: SLI is not recommended for GPU crunching. You should try to turn it off for a test period (even remove the SLI bridge).
8) Message boards : Number crunching : Problem with PABLO tasks (Message 48224)
Posted 21 days ago by Profile Retvari Zoltan
Thanks for reply.
The above procedure is a workaround not a solution for me. I have sometimes high cpu-usage by video-playing, so I need the pause-option.
Then you should put those games and applications to the settings -> exclusive applications list.
9) Message boards : Number crunching : Pablo WU erroring out at a high rate - (Message 48217)
Posted 23 days ago by Profile Retvari Zoltan
This method supposedly blocks only your GPU from automatic Windows updates, but I have not tried it.



Well, there are 5 methods in the blog post you've linked.
1. Hiding a driver update is irrelevant regarding this issue, as you can use it after an automatic driver update happened.
2. Disabling Windows update completely to avoid driver updates makes your OS vulnerable, so this is forbidden.
3. Blocking driver installation for a particular device ID (through group policy) is good, but it will prevent you to update the driver manually without first removing the blocking group policy object (then you should disable the given device ID again). This method is linked in the post I've linked.
4. This method is the same I've posted.
5. The Windows update preferences does not have the option of driver updates in the latest release (Creators Fall Update, v1709), it has been removed from the system properties dialog box also.
10) Message boards : Number crunching : Pablo WU erroring out at a high rate - (Message 48212)
Posted 24 days ago by Profile Retvari Zoltan
Anyone else experince this over the past 12 hours ? I have multiple machines that errored out on multiple WUs consecutively.
I've experienced the same on one of my Windows 10 PCs. The source of this error is that the latest NVidia driver (388.13) has been released through Windows update, and it will update your GPU driver without stopping BOINC GPU tasks first.
There are two workarounds:
1. Update the drivers manually before it rolls out on Windows update (the tricky part is that not all NVidia driver versions get rolled out on Windows update, and I don't have a source to tell which version actually does)
2. Disable Windows update to update drivers. Here's how. The drawback of this method is that it will stop all driver updates, not just the NVidia drivers.

Next 10