Windows 11 problem
27 May 2023 0:54:59 UTC
Topic 229589
(moderation:
This is to provide a heads-up to a potential problem that might affect others. Some months ago my computer started to experience a lot of freezing up which actually required a manual power off. I suspected a virus and ran many different scans. As part of the problem search I suspended my E@H tasks and saw the problem cease. Should I blame E@H? It had been running since 2008 with no such problem. After many changes to my preferences and no improvement I actually (and shamefully?) aborted 2 E@H tasks which seemed bloated were estimated to overrun their deadlines. No improvement! Finally I took it to a repair shop and heard the scuttlebutt that windows 11 was memory hog. I increased the memory from the original 8 gigabyte to 40 gig and added a new 1 terabyte solid state hard drive with another windows 11 on it. Now I can run E@H peacefully & happily. Looking back on it, I might been able to pre-empt it all by just buying another 32 gig and installing it myself. It's something most of us can do and the i5 processors in my Dell G5 can address that much memory. I hope this info helps others avoid these headaches.
Davissimo wrote: This is to
)
Yes it would be very helpful if the Developers stickied a post saying how much memory a task would take for each of the kinds of tasks, both cup and gpu, to help people run the tasks more efficiently and not just give up and go away and never come back instead of trouble shooting the problem as you did.
I'll add something I talked
)
I'll add something I talked to Mickey about. Windows 11 together with Radeon RX 6900 XT gives a very short CPU Time for FGRP (16-30 sec.). I went back to Windows 10 and CPU Time was back to normal (~80 sec.). Run Time remained unchanged. So Win 11 is either great or buggy :).
K_PL wrote:... So Win 11 is
)
That's probably too simplistic a view :-).
I had a look at both of your setups - Win 11 and Win 10 - assuming that the hardware was not changed when you went back to Win 10. Both instances (with their completed results) are visible so I looked at both sets of validated results - just eyeballing the last page of results in each case. My estimate for Run/CPU times were 518/19 for the Win 11 instance and 516/83 for the newer Win 10 results.
I say 'newer' to remind readers that there has been a data file change. The Win 11 results were for the file LATeah4021L12.dat whilst the Win 10 results were largely for LATeah4021L13.dat. It's always possible that results can change slightly when there is a change in the data file being analysed.
However, there's a more interesting change in how the hardware is being displayed by BOINC. For the Win 11 setup (BOINC version 7.20.2) the GPU is shown as "AMD AMD Radeon(TM) Graphics (12269MB)" whilst for the Win 10 setup (BOINC version 7.22.2) it's listed as "AMD AMD Radeon RX 6900 XT (16368MB)". I have no idea if this is due to the change in BOINC version, the potentially quite different drivers, or even a change in the GPU itself. Whilst I assume it's the same GPU, the change in the detected VRAM seems puzzling.
So there are other explanations for the CPU time difference rather than just saying there's something wrong with Win 11. The GPU does the bulk of the calculations. The CPU time component is for 'supporting' the GPU. Some of that support is for transferring data to/from the GPU and some is for handling the last 10% when the 'toplist' of potential candidates is created. Perhaps the newer Win version has a more efficient driver (provided by AMD I assume) that doesn't need as many CPU cycles to provide that support.
Another possible explanation for the shorter CPU time might be a greater CPU frequency. Could it be that the Win 10 setup has the CPU frequency throttled down to save power whereas the Win 11 setup might be running at the full frequency? Did you check the CPU speed being used in each case?
With my machines, I tend to have a setup that most efficiently uses the available resources. Since I don't use the CPU for crunching (GPU tasks only), I do tend to run the CPU at idle frequency as much as possible. That tends to minimise the amount of total power consumed without significantly affecting task run times. Of course, the CPU time reported will tend to increase but the total time to crunch a task tends to stay the same.
Cheers,
Gary.
Gary Roberts. My hardware
)
Gary Roberts.
My hardware remained unchanged. I also observed a lower CPU Time result with CPU primegrid calculations (Win 11). And now I give an example of another user with similar equipment: https://einsteinathome.org/pl/account/310074/computers I don't have a good theory. I'm leaning towards the fact that Windows 11 may use a better driver. However, if Time CPU data is erroneous, it should not be included in statistical research. The samples, of course, are counted correctly. Perhaps a test of the power consumption of both systems would clarify the matter.
Regards.