because if you look at the images you will see that it is called python.
GPU grid runs python.
The specific task name is Python apps for GPU hosts.
I wish you would just stay with the conversation and not argue with me over names.

You said WCG and Folding, you didn't say GPUgrid, if you're going to give me incorrect information I can't help you.

It's all ok...we figured it out.
I was trying to make dinner and answer the two of you hitting me with questions at the same time and try to get images uploaded. All this after work. So excuse me if I was jumbling answers.

We know now what the issue is and I figured that was probably the case last night when I opened up the tab in task manager for BOINC. Was shocked at how many processes were running with GPU Python.

Anyway...I'll dump GPU Grid when it finishes and see how the system balances out after that.

WCG is back. But they keep coughing up those damn transient errors now and then. I would have thought they had that fixed by now the way they are hyping being back up and running.

No errors here. I'm on 2 million credit per day with zero failed tasks.

Tasks are running fine from WCG. What Greg's talking about is the downloading of tasks throwing up http transient errors by the bucketful.
I'm getting the same and it's driving me crackers.
They did solve it a week or two back, but it's returned with a vengeance in the last several days.
It's taking longer to get a successful download of all the task components than the OPN GPU tasks are taking to run here

---

Was shocked at how many processes were running with GPU Python.
Anyway...I'll dump GPU Grid when it finishes and see how the system balances out after that.

PythonGPU works good if run right. Do not try to run it on a CPU with less than 32 threads. I've tried 24 threads and it's very slow.
Run 2 PythonGPU WUs and nothing else is best. The 2 PythonGPU WUs play well together and can share those CPU threads. If you try to run a different project with a PythonGPU WU it'll have annoying quirks. Since most of the work is actually done on the CPU it can use less powerful GPUs, e.g. 1080, than the acemd4 WU needs. Here's the app_config I use on an i9-10980XE with a 3060 Ti:

<app_config>
<!-- i9-10980XE   18c36t   2x16=32 GB   L3 Cache 24.75 MB  3060 Ti -->
    <app>
        <name>PythonGPU</name>
        <plan_class>cuda1131</plan_class>
        <gpu_versions>
            <cpu_usage>32</cpu_usage>
            <gpu_usage>0.5</gpu_usage>
        </gpu_versions>
        <max_concurrent>2</max_concurrent>
        <fraction_done_exact/>
    </app>
</app_config>

---

I've given up on GPUgrid. Despite countless promises they refuse to produce tasks to run on AMD GPUs, which is what I have exclusively, due to them being faster for the cost than Nvidia, which severely cripples double precision, which is needed for a lot of projects, not just Milkyway. Some of the calculations for most projects are DP, a lot more than the 1/64 nonsense that Nvidia give us.

I've also given up on Folding@Home. Nice idea, because there's not much biology on Boinc. But if Folding aren't willing to join the rest of us on Boinc, I can no l0nger be bothered. It's too much effort to run a different program alongside Boinc, which has no clue what other projects are running on the processors and GPUs, so balancing loads is ridiculous.

I've also given up on Folding@Home. Nice idea, because there's not much biology on Boinc. But if Folding aren't willing to join the rest of us on Boinc, I can no l0nger be bothered. It's too much effort to run a different program alongside Boinc, which has no clue what other projects are running on the processors and GPUs, so balancing loads is ridiculous.

I have no problem with FAH on my system.
Seems to balance fine with BOINC

Is this guy done with 4.2 already? 0 tasks in queue!
FFS...this is nuts.
I guess 4.2 is small batches of stuff.
Where is all the rest of the stuff that Robetta has stocked up?
All for the in house systems?

Just divide 100 by number of cores and substract resulting number from 100 and put number in use at most x% of the cpus
Example 100/8=12.5
100-12.5=87.5

I've also given up on Folding@Home. Nice idea, because there's not much biology on Boinc. But if Folding aren't willing to join the rest of us on Boinc, I can no l0nger be bothered. It's too much effort to run a different program alongside Boinc, which has no clue what other projects are running on the processors and GPUs, so balancing loads is ridiculous.

I have no problem with FAH on my system.
Seems to balance fine with BOINC

I don't see how. If I run two projects in Boinc, it chooses to run one or the other. If I run Folding aswell, Folding doesn't know when Boinc managed to get WCG tasks.

Just divide 100 by number of cores and substract resulting number from 100 and put number in use at most x% of the cpus
Example 100/8=12.5
100-12.5=87.5

Until you have 24 cores and get recurring decimals.... I have to remember Boinc rounds down for CPUs and up for GPUs!

I've also given up on Folding@Home. Nice idea, because there's not much biology on Boinc. But if Folding aren't willing to join the rest of us on Boinc, I can no l0nger be bothered. It's too much effort to run a different program alongside Boinc, which has no clue what other projects are running on the processors and GPUs, so balancing loads is ridiculous.

I have no problem with FAH on my system.
Seems to balance fine with BOINC

I don't see how. If I run two projects in Boinc, it chooses to run one or the other. If I run Folding aswell, Folding doesn't know when Boinc managed to get WCG tasks.

I don't know...maybe because I am not using CPU for FAH. I have so many CPU projects there is no room for anything else. So I put FAH in GPU only mode.

Since this project doesn't have anything going on, I am going back to GPU and see what that does to the other projects I run in CPU. If it cuts them in half then I will terminate GPU and find something else.

But this all takes time since I run only 14 hrs a day.

But this all takes time since I run only 14 hrs a day.

Lightweight, I run 10 machines 24 hours a day :-)

Ooh, new batch of tasks again? No idea how big yet
We could get used to this

---

Ooh, new batch of tasks again? No idea how big yet
We could get used to this

I certainly hope so!

Looks like things are running okay so far. This is batch 2. ~5mil total for this experiment on the way.

Looks like things are running okay so far. This is batch 2. ~5mil total for this experiment on the way.

Great!!

---

Ooh, new batch of tasks again? No idea how big yet
We could get used to this

I certainly hope so!

Looks like things are running okay so far. This is batch 2. ~5mil total for this experiment on the way.

Because there was a slight gap between the previous batch ending and the new one being issued, my systems switched over to my backup project, so I haven't started running any yet.
But the previous batch all ran successfully - no reason to think this one will be any different.

Ideally, I don't want to run my backup project, so if there are any other researchers needing to have work done, send them our way and we'll get a flow going without interruptions.
Things work more reliably on both sides of the server divide that way
[hint][/hint]

---

Ooh, new batch of tasks again? No idea how big yet
We could get used to this

I certainly hope so!

Looks like things are running okay so far. This is batch 2. ~5mil total for this experiment on the way.

Nice to have lots of R4.2 work without problematic pythons eating my ssd :-)

Nice to have lots of R4.2 work without problematic pythons eating my ssd :-)

Those things last forever despite what people say. Anyway, SSDs suck, I use NVME.

[quote]... SSDs suck, I use NVME.

You might benefit from a good google search on NVMe. It's not what you seem to think it is.

[quote]... SSDs suck, I use NVME.

You might benefit from a good google search on NVMe. It's not what you seem to think it is.

I know exactly what they are and have several, they're much faster. An SSD over SATA is far too slow, because of the SATA.

Are you from the Android forum?