As promised after the database work, three changes are live:
• Up to 8 GPU tasks per card (it was 2). You can run two at once on the same card (app_config.xml with gpu_usage 0.5) and still keep a small buffer.
• Choose your applications: in your Eternity II project preferences, you can now tick which applications you want, for example CPU only or GPU only. Important for GPUs: the server picks the GPU task size for your card model, and a card only runs its own size. So if you want GPU work, keep all the GPU applications ticked: unticking a size can leave your card with nothing to do. If you also answer yes to "If no work for selected applications is available, accept work from other applications?", your machine never runs dry.

• More GPU models accepted: TITAN Xp, Tesla T4, T500, Quadro RTX 6000, and the RTX 4070 / 4080 / 4090 / 5060 Laptop GPUs. If your NVIDIA card is still refused, please check your driver first: we need at least 528.33 on Windows or 525.60 on Linux (CUDA 12). With an older driver the card is refused whatever its model.

Thanks to mmonnin, Freewill and Magiceye04 for asking!

That's great! What's the app name to use in the app_config for GPU 2X?

It is interesting that we can choose now which GPU version to accept but that feature is pretty much useless if the scheduler decides the app version anyway. I guess mmonnin's proposal aimed more at the volunteer choosing the GPU app version to collect hours on WUProp for all versions on the same GPU. This may be less effective on the projects side but a good incentive for volunteers to continue crunching until a certain star colour is reached on all versions. It could be restricted upwards so that somebody with a GT730 doesn't get the really big tasks it cannot handle anyway.

hello,

GPU tasks remain in a suspended state even though the machine is available
https://i.postimg.cc/jqB6SB6J/Capture-d-ecran-2026-11-05-145002.png.

I don't understand why, CPU usage is low and GPU usage is at 0%.

I am glad that this project was shared on BOINC and that GPU calculations are available.

Regards


edit : It was enough to disable "suspend GPU when computer is in use"

@Freewill: the GPU application names are:
eternity_gpu_272_65536v
eternity_gpu_272_32768v
eternity_gpu_272_16384
eternity_gpu_272_8192
eternity_gpu_273_8192
eternity_gpu_273_4096
Your card only receives some of these sizes (it depends on the model), so the simplest is to list all six with 0.5 for 2 tasks per card, for example:

eternity_gpu_272_65536v0.5
(same line for the five other names)

The server allows up to 8 GPU tasks in progress per card.

@Dargo75: you were right, and it's now partly possible. By default each card still gets the size chosen for its model, but there is a new box in your Eternity II project preferences: "GPU: also accept the next bigger task size than the one chosen for my card". Tick it and your card also gets the next size up, so it runs (and collects WUProp hours on) a second GPU application. One step only, so a small card never gets tasks far too long for it. Details in News.

@Dargo75: you're right, and I should have been clearer. For GPUs, the choice mainly lets you turn GPU or CPU work on or off: the task size each card receives is still matched to its model, because a size that doesn't fit the card runs much less efficiently (too small for the big cards, much too long for the small ones). I understand the WUProp side of mmonnin's request: receiving every size a card can handle is noted, but I can't promise it yet.

@artik: glad you found it! For others: if GPU tasks stay suspended while the computer is idle, check "Suspend GPU computing when computer is in use" in the BOINC computing preferences.

Ok there is an app selection, but you still have your limits. I still cannot get any of the 4k or 8k tasks. The vmem usage goes up from 16-32-65k, so why can I not get the lower ticket apps with lower vmem usage?

In reply to mmonnin's message of 6 Nov 2026:

Ok there is an app selection, but you still have your limits. I still cannot get any of the 4k or 8k tasks. The vmem usage goes up from 16-32-65k, so why can I not get the lower ticket apps with lower vmem usage?

@mmonnin: the limit isn't video memory, it's how many tickets the card works on at once. The GPU application runs as many tickets in parallel as the card has compute units × 640, so your cards are sized for 16k, 32k and 65k tasks. A 4k or 8k task would keep only a small part of the card busy, and it would do much less work per hour. Those two sizes are made for the smallest and oldest cards (the 273 ones use a different kernel for older generations). The "next bigger size" box in the preferences gives you one more application per card. Receiving the smaller sizes as well is on my list for WUProp, but I can't promise it yet.

In reply to benjamin's message of 6 Nov 2026:

In reply to mmonnin's message of 6 Nov 2026:

Ok there is an app selection, but you still have your limits. I still cannot get any of the 4k or 8k tasks. The vmem usage goes up from 16-32-65k, so why can I not get the lower ticket apps with lower vmem usage?

@mmonnin: the limit isn't video memory, it's how many tickets the card works on at once. The GPU application runs as many tickets in parallel as the card has compute units × 640, so your cards are sized for 16k, 32k and 65k tasks. A 4k or 8k task would keep only a small part of the card busy, and it would do much less work per hour. Those two sizes are made for the smallest and oldest cards (the 273 ones use a different kernel for older generations). The "next bigger size" box in the preferences gives you one more application per card. Receiving the smaller sizes as well is on my list for WUProp, but I can't promise it yet.

I can run multiple tasks at once. I don't care what you think I should run. Your app selection is pointless.

Hello,

Is there a real gain to running multiple GPU tasks on the same card?
Tasks 65536v seem to occupy my 4080 at 100% for 2 hours.(the last % is very long)

(7.5 go / 16 Vram used)

I think an RTX 4080 is capable of handling the tasks of 131,072 tickets. (15 go vram 4 hours )

Could you please share the exact content of the app_config.xml file? thx

it doesn't work for me

<app_config>
    <app>
        <name>eternity</name>
        <app_version>
            <gpu_usage>0.5</gpu_usage>
        </app_version>
    </app>
</app_config>

This is what it should look like to run 2 GPU tasks at once, each with 2 CPU threads and half of your GPU. You can adjust those numbers and see what works best for you. There probably is no more app version called "eternity" since they all are taylored options for your GPU but leave it for now, read the config files and check your Boinc messages. It will most likely say that the app "eternity" is unknown, known options are "....." and then you replace it by the version name it tells you there.

<app_config>
<app>
<name>eternity</name>
<gpu_versions>
<cpu_usage>2.0</cpu_usage>
<gpu_usage>0.5</gpu_usage>
</gpu_versions>
</app>
</app_config>

be careful on name, from my latest test

<name>eternity_gpu_272_16384</name>

@mmonnin: fair point, running several tasks at once on one card is perfectly fine, and many of you do it. The server still sends each card the sizes made for it, and receiving the smaller sizes as well is on my list.

@artik: yes, running 2 tasks at once usually helps: the end of a task (the last %) uses the card much less, and a second task fills that gap. Thanks to Dargo75 and rebirther for the help! The two fixes are:
• the name must be the application's name, not "eternity" (rebirther is right);
• the settings go in gpu_versions, not app_version.

For your 4080 (65,536-ticket tasks):

<app_config>
<app>
<name>eternity_gpu_272_65536v</name>
<gpu_versions>
<gpu_usage>0.5</gpu_usage>
<cpu_usage>2</cpu_usage>
</gpu_versions>
</app>
</app_config>

Then use "Options → Read config files" in BOINC Manager. Watch the video memory: two 65,536 tasks at 7.5 GB each almost fill your 16 GB.

For the 131,072-ticket tasks: tick the "next bigger size" box in your project preferences. On a 5070 Ti we measured about 21% more work per hour with them than with the 65,536 ones. At about 15 GB of video memory, run them one at a time.