Is there a minimum driver version or CUDA version required?

Is there a minimum driver version or CUDA version required?

yes, all the new ACEMD tasks here are CUDA 10.0 on Linux, and CUDA 10.1 on Windows. so you need the appropriate drivers for that CUDA version.

Linux, CUDA 10.0 - >=410.48
Windows, CUDA 10.1 - >=418.96

---

My GTX 750ti appears to have been disqualified by the server. Despite over 1K WUs waiting to be sent my log says "Scheduler request completed: Got 0 new tasks" and "No tasks sent".

The previous WU was completed in 408,214 seconds, with just 23,786 seconds to spare. Is that why?

Edit;
I see the server gave it a grace period of ~35 min to finish. That might explain it.

My GTX 750ti appears to have been disqualified by the server. Despite over 1K WUs waiting to be sent my log says "Scheduler request completed: Got 0 new tasks" and "No tasks sent".

The previous WU was completed in 408,214 seconds, with just 23,786 seconds to spare. Is that why?

It's not just your GTX 750Ti.
My GTX 1080Ti/Linux didn't receive work:

2021. febr. 16., Tuesday, 00:19:34 CET | GPUGRID | checking NVIDIA GPU
2021. febr. 16., Tuesday, 00:19:34 CET | GPUGRID | [work_fetch] set_request() for NVIDIA GPU: ninst 1 nused_total 0.00 nidle_now 1.00 fetch share 1.00 req_inst 1.00 req_secs 25920.00
2021. febr. 16., Tuesday, 00:19:34 CET | GPUGRID | NVIDIA GPU set_request: 25920.000000
2021. febr. 16., Tuesday, 00:19:34 CET | GPUGRID | [work_fetch] request: CPU (0.00 sec, 0.00 inst) NVIDIA GPU (25920.00 sec, 1.00 inst)
2021. febr. 16., Tuesday, 00:19:34 CET | GPUGRID | Sending scheduler request: To fetch work.
2021. febr. 16., Tuesday, 00:19:34 CET | GPUGRID | Requesting new tasks for NVIDIA GPU
2021. febr. 16., Tuesday, 00:19:35 CET | GPUGRID | work fetch suspended by user
2021. febr. 16., Tuesday, 00:19:36 CET | GPUGRID | Scheduler request completed: got 0 new tasks
2021. febr. 16., Tuesday, 00:19:36 CET | GPUGRID | No tasks sent
2021. febr. 16., Tuesday, 00:19:36 CET | GPUGRID | No tasks are available for New version of ACEMD
2021. febr. 16., Tuesday, 00:19:36 CET | GPUGRID | Project requested delay of 31 seconds

also my RTX 2080Ti/Windows didn't receive work:

2021. 02. 16. 0:23:06 | GPUGRID | checking NVIDIA GPU
2021. 02. 16. 0:23:06 | GPUGRID | [work_fetch] set_request() for NVIDIA GPU: ninst 1 nused_total 1.00 nidle_now 0.00 fetch share 1.00 req_inst 0.00 req_secs 23728.26
2021. 02. 16. 0:23:06 | GPUGRID | NVIDIA GPU set_request: 23728.255416
2021. 02. 16. 0:23:06 | GPUGRID | [work_fetch] request: CPU (0.00 sec, 0.00 inst) NVIDIA GPU (23728.26 sec, 0.00 inst)
2021. 02. 16. 0:23:06 | GPUGRID | Sending scheduler request: To fetch work.
2021. 02. 16. 0:23:06 | GPUGRID | Requesting new tasks for NVIDIA GPU
2021. 02. 16. 0:23:08 | GPUGRID | Scheduler request completed: got 0 new tasks
2021. 02. 16. 0:23:08 | GPUGRID | No tasks sent
2021. 02. 16. 0:23:08 | GPUGRID | No tasks are available for New version of ACEMD
2021. 02. 16. 0:23:08 | GPUGRID | Project requested delay of 31 seconds

Something broke in the scheduler, as the tasks in progress is decreased by about 1400.

I've managed my other host to get a new task by updating manually a couple of times, but the others still didn't get one.
It looks like the scheduler thinks that the majority of the unsent tasks aren't for the "new version of ACEMD" app despite they are shown next to that label.

Thanks Zoltan, I gave up for now and switched that GPU to FAH for now.
It seems to be doing more FLOPS/hr when running FAHcore CUDA vs ACEMD, but that may be just the difference in scoring procedures.

I have a 1660 super that takes about 34-38 hours on these new units, still seeing temps in the 60s, only uses 15% of gpu in task manager tho

I've managed my other host to get a new task by updating manually a couple of times, but the others still didn't get one.
It looks like the scheduler thinks that the majority of the unsent tasks aren't for the "new version of ACEMD" app despite they are shown next to that label.

I haven't seen my systems have any issue with getting new tasks.

but I do wonder what's going on with the massive shift of tasks from out in the field to waiting to be sent. are they erroring en masse somehow? today is about 5 days since these new tasks started showing up, so perhaps that's why. thousands of tasks hitting their deadlines from systems not fast enough to process, or systems that are fast enough if they run 24/7, but aren't processing 24/7, or systems that downloaded but were shut off for the past 5 days. or some combination of all 3.

---

我发现我只接收到一个任务，完成后不再有任务，我是GTX1650,这正常吗

I have a dual GPU system, GTX 980 and GTX 1080 TI, and all of these work units have failed. Drivers are current as of Decemeber. I had to roll back January update as it didn't play nice with Milkyway@Home while you folks were on Holiday. Suddenly I can't complete a work unit without error.

I'm not seeing many resends, mostly _0 and _1 original tasks.

No issues getting work or returning valid results.

Does the scheduler know the correct size estimate for these WUs?

A WU on my GTX 1060-6GB should take ~34 hours (~ 4 calendar days) yet it keeps sending me 2. My queue's set to 0.4 days of work, and I believe it should know that my computer is only 30% active, so I would expect it to only send 1 WU.

but I do wonder what's going on with the massive shift of tasks from out in the field to waiting to be sent. are they erroring en masse somehow? today is about 5 days since these new tasks started showing up, so perhaps that's why. thousands of tasks hitting their deadlines from systems not fast enough to process, or systems that are fast enough if they run 24/7, but aren't processing 24/7, or systems that downloaded but were shut off for the past 5 days. or some combination of all 3.

In fact, giving a certain time offset before overdue tasks are resent, would effectively act as extending the deadline by this offset for them to be reported by slower GPUs
One example: WU #27023500 has been reported by a GTX 750 Ti at one of my hosts in 446,845.54 seconds, more than 4 hours after deadline.
This task has been rewarded with 348,750.00 credits, and hasn't been resent to any other host, with "Didn't need" legend.
May be the project managers are hiddenly attending this way the request of many Gpugrid users in this regard (?)

Do these work with Ampere?

Do these work with Ampere?

No.

Ah damn ill keep waiting then :)

Re: New WU's and tuning GPU's
Interested in OverClocking to reduce WU duration to hit target 'due date/times'?
Even if your GPU is locked down you can improve by using a curve with Frequency and Voltage of your GPU auto managed.
Message me if you have questions or need help.

I'm crunching the new WU's with RTX2080 mobile with MSI Afterburner and MSI Kombustor linked (that auto overclocks the GPU with a good curve).
I am getting 2 days and 2 hours for e20s2_e11s14p0f75-ADRIA_D3RBandit_batch1-0-1-RND2090. 1.08% per hour = 92.59 hours or 3.58 days.


0.998 CPUs + 1 NVIDIA GPU
Estimated computation size
5,000,000 GFLOPs
CPU time
00:49:23
CPU time since checkpoint
00:08:14
Elapsed time
00:49:40
Estimated time remaining
2d 01:07:38
Fraction done
1.333%
Virtual memory size
668.21 MB
Working set size
343.72 MB
Progress rate
1.080% per hour
Executable
wrapper_6.1_windows_x86_64.exe

Re: New WU's and tuning GPU's
Interested in OverClocking to reduce WU duration to hit target 'due date/times'?
Even if your GPU is locked down you can improve by using a curve with Frequency and Voltage of your GPU auto managed.
Message me if you have questions or need help.

I'm crunching the new WU's with RTX2080 mobile with MSI Afterburner and MSI Kombustor linked (that auto overclocks the GPU with a good curve).
I am getting 2 days and 2 hours for e20s2_e11s14p0f75-ADRIA_D3RBandit_batch1-0-1-RND2090. 1.08% per hour = 92.59 hours or 3.58 days.


0.998 CPUs + 1 NVIDIA GPU
Estimated computation size
5,000,000 GFLOPs
CPU time
00:49:23
CPU time since checkpoint
00:08:14
Elapsed time
00:49:40
Estimated time remaining
2d 01:07:38
Fraction done
1.333%
Virtual memory size
668.21 MB
Working set size
343.72 MB
Progress rate
1.080% per hour
Executable
wrapper_6.1_windows_x86_64.exe

---

Dears,

as you may have noticed, we sent a new batch of WUs for a new experiment. This time the WUs are rather large and require relatively new cards. For reference, should be ~18h on a 1080 Ti.

Thanks!

T

Re: New WU's and tuning GPU's
Interested in OverClocking to reduce WU duration to hit target 'due date/times'?
Even if your GPU is locked down you can improve by using a curve with Frequency and Voltage of your GPU auto managed.
Message me if you have questions or need help.

I'm crunching the new WU's with RTX2080 mobile with MSI Afterburner and MSI Kombustor linked (that auto overclocks the GPU with a good curve).
I am getting 2 days and 2 hours for e20s2_e11s14p0f75-ADRIA_D3RBandit_batch1-0-1-RND2090. 1.08% per hour = 92.59 hours or 3.58 days.

If you get despondent or cannot meet deadlines but want GPU points, check out Moo Wrapper. (28 minutes per WU on an RTX 2080 mobile)

Re: New WU's and tuning GPU's
Interested in OverClocking to reduce WU duration to hit target 'due date/times'?
Even if your GPU is locked down you can improve by using a curve with Frequency and Voltage of your GPU auto managed.

If your GPU is crunching a single workunit for days, it's not worth the risk of a computing error caused by overclocking.
You'll receive 0 credits for a failed workunit (after many hours, even days of crunhing it's very frustrating).
Therefore I do not recommend overclocking and especially overvolting a GPU, especially a mobile GPU.
GPUGrid workunits are very power hungry compared to games or other projects (except for FAH).
The cooling of an average GPU is made for general use, not for crunching 24/7.
Laptops with mobile GPUs can't have that big coolers as discrete GPUs have in desktop PCs.
If you have a GPU with decent cooling, then it's usually overclocked by the factory. In this case you don't have to overclock it more.

Power dissipation is a product of two key factors:
· It's in direct ratio with GPU frequency.
· It's in direct ratio with GPU voltage squared.
Say you raise the frequency and the voltage by 10% (it's a bit of an exaggeration, as you can't raise the GPU voltage by 10%).
In this case the power dissipation of your GPU is raised by 33.1% (1.1 by the frequency, and 1.1*1.1=1.21 by the voltage, 1.1*1.21=1.331).
Luckily you can't raise your GPU's power consumption above it's limits set by the factory.
You can check these limits from an administrative command prompt by

nvidia-smi -q -d power

Raising the GPU's power dissipation raises its temperature, as its cooling stays the same, while it should be better to achieve the same temperatures (and life expentancy). Usually you improve the cooling of your GPU only by raising the RPM of it's fans, which could be very annoying (especially if it's a laptop), it also reduces the lifespan of the fans.