Jake Weiss wrote:

Peciak,

That's weird. The version info doesn't say its an OpenCL app. Let me see if I forgot to set the flag when I recompiled it.

Sorry about this, I haven't slept much over the last couple days trying to get this update done.

Jake

Jake Weiss wrote:

Hey Everyone,

I'm still working on a fix. Having issues with other the Windows OpenCL version but not Linux OpenCL. Need to see if my drivers are back on the machine I'm testing with or if there is actually an issue with the client.

Might take a couple hours for a fix.


Sorry,

Jake

Jake, what you think about to cancel the release of the new (Windows) >1.42 apps?

Just go back to former 1.39 and let it over the weekend like before.

You could relax and 'recharge your batteries' over the weekend - and at Monday a new working day begin.

:-)

---

I am getting a 0% load on my GPU as measured by TechPowerUp GPU-Z that can be downloaded from https://www.techpowerup.com/downloads/. I think that your application is wasting a significant amount of time in between the time each work unit is being processed.

I've been sent tasks that have estimates of over 6 hours:-( Taasks that used to take 32 seconds under 1.39 now take 9 mins plus..
The estimate for 1.42 bundle 5 taks was apparently 10 mins, but took 38mins to do 50% then went back to 0% and carried on with an updated estimate of 6 hours plus..
I am NOT a happy bunny:-(

Therefore I have aborted ALL bundle 5 tasks which for a cache set to 1 day I got several days worth of bundle 5's hence the abort, I'm prepared to accept a 9 min completion under 1.42 for those modfit's that used to take 32 sec under 1.39. but not for longer than to complete those tasks sent me.

Thereafter unless MW@H reverts to 1.39 for those tasks I will suspend MW@H until such time as the situation has been resolved. Given that the credit return on tasks taking 32 sec was about 26cr I'm far from pleased if those same tasks now take 9 mins plus for probably the same return of 26cr. Let alone for tasks that apparently will take many hours to complete.

I run my computers at cheap[ish] rate for only 7 hours per day, and cannot afford an extra £256 per month to run 24/7.

So I'm hoping this situation can be resolved as soon as Jake is fully rested up:-)

---

Regards,
Cliff.
--
Been there Done That, still no Damn T-Shirt

The 1.42s are working for me. However, even though they are taking up to 3 times as long to complete, (both GPU and CPU) the Credit applied is the same as the previous version - 1.39?

At least on my dual Xeon CPU PC with Win8.1 x64 Pro the 1.42 CPU app run like it should - I guess:

de_modfit_fast_19_3s_136_ModfitConstraints3_3_1478718902_2055111_2
Run Time - CPU Time - gr.Cr. - app
796.49 - 633.84 - 26.73 - MilkyWay@Home v1.42 (1 WU)

de_modfit_fast_19_3s_136_bundle5_ModfitConstraints3_4_1478900148_40245_0
Run Time - CPU Time - gr.Cr. - app
4,331.97 - 3,166.94 - 133.66 - MilkyWay@Home v1.42 (5 WUs in 1 WU)

CPU Time and granted Credits are *5.

Run Time is higher than CPU Time because of currently SETI GPU app support. Because the currently Milkyway 1.42 ATI OpenCL GPU app don't work... ;-) - like I read here in this thread.
Milkyway 1.40 ATI OpenCL GPU app made computation errors (I sent Jake a PM because of the OS error message which appeared in my Win8.1 x64 Pro).

BTW.
The 1.39 CPU app was slower on the above mentioned PC.

de_modfit_fast_19_3s_136_ModfitConstraints2_3_1478718902_2595000_2
Run Time - CPU Time - gr.Cr. - app
809.55 - 668.44 - 26.73 - MilkyWay@Home v1.39

---

Hey Everyone,

Going to go ahead and throw a pause on the windows GPU WUs for the night. CPU work units seem to be working as intended (Taking 5x longer or so and giving 5x credit).

I'll try to get a new working GPU client by morning.

Thank you for your patience,

Jake

Thanks. Another speculative possibility that I should have thought of is that the Windows Nvidia GPU client might be running the work units in the CPU instead of the GPU because it is using as much CPU as a single core CPU client.

Hi Jake,
Get some rest in 1st..

Thanks for getting onto this so fast, but you also need some kip to work properly:-)

---

Regards,
Cliff.
--
Been there Done That, still no Damn T-Shirt

Hi Jesse,

Thanks. Another speculative possibility that I should have thought of is that the Windows Nvidia GPU client might be running the work units in the CPU instead of the GPU because it is using as much CPU as a single core CPU client.

On my rig BOINC shows 0.719 cpu + 1 GPU.
running on a GTX980ti.. That's the same cpu usage as before using 1.39..
But the 1.42 app takes 9 min 55 sec for the same type of task..

I think its taking a rest between munching on a slice of WU:-)

So I don't think its an increase in cpu use.
The CPU is AMD 9590 at 4.7 Ghz...

---

Regards,
Cliff.
--
Been there Done That, still no Damn T-Shirt

Hi Jesse,

Thanks. Another speculative possibility that I should have thought of is that the Windows Nvidia GPU client might be running the work units in the CPU instead of the GPU because it is using as much CPU as a single core CPU client.

On my rig BOINC shows 0.719 cpu + 1 GPU.
running on a GTX980ti.. That's the same cpu usage as before using 1.39..
But the 1.42 app takes 9 min 55 sec for the same type of task..

I think its taking a rest between munching on a slice of WU:-)

So I don't think its an increase in cpu use.
The CPU is AMD 9590 at 4.7 Ghz...

That's why many MW users run several WUs at once. There's some CPU crunch time where the GPU is not used. The 980Ti is much better suited for other projects. My 280x does 4 tasks in 31seconds at stock speeds.

I have some invalids and errors. I only had errors on the old version. All units are validated immediately. No more 10k+ WUs waiting for validation and I always have 80 WUs. As soon as WUs are reported I get replacements.

I just had a whole bunch WU's end in "Computation Error". I guess I need to download a new application file. . . or something.

---

Runtime on 290X went from 15 seconds to 27,066 seconds and I still get same 26.73 credits, no thanks : http://milkyway.cs.rpi.edu/milkyway/result.php?resultid=1886485669

Are the individual units in the new bundle of 5 units bigger than the old units?

The 1.42s are working for me. However, even though they are taking up to 3 times as long to complete, (both GPU and CPU) the Credit applied is the same as the previous version - 1.39?

Runtime on 290X went from 15 seconds to 27,066 seconds and I still get same 26.73 credits, no thanks : http://milkyway.cs.rpi.edu/milkyway/result.php?resultid=1886485669

As I said above, please check your GPU core and RAM clocks:

Engaging on v1.42 tasks results in immediate GPU core clock reduction to 300 Mhz and GPU RAM clock to 150 MHz on both 280X and 290X AMD graphics boards.

I will have to suspend MW until this issue is resolved.

Michael.

---

President of Rechenkraft.net e.V. - This planet's first and largest distributed computing organization.

I think Jesse hit the nail on the head in message 65703.

The "GPU" units aren't using any (or much) GPU processing. It is all CPU.

I have exactly the same situation. GPU-Z shows no GPU processing being used at all.


EDIT: And instead of one unit taking 8 minutes (I have a slow card) a bundle of 5 looks like it is going to take about 9 hours (which also indicates they are really CPU tasks)

EDIT: I am going to Suspend Milkyway until this issue resolved. By the way, this is on a Windows 10 machine. No evidence yet this is also a LINUX issue.

I think Jesse hit the nail on the head in message 65703.

I think I have already stated that in message 65688, but sure.
I have said that there's no GPU usage in opencl_ati tasks.

1.42 NVIDIA tasks are running fine on my system.
I am not as powerful as some of you guys so it is about 3:52 hrs to process a bundle5 task.
I see the modfitconstraints2_2 and 3_3 showing dual CPU and GPU usage with a estimated time of 46:35 minutes mins.

1.42 NVIDIA tasks are running fine on my system. (...)

No, they run not fine, because they don't use the GPU, they only use a full CPU core, thereby slowing down your other WUs, running on the CPU.

---

Aloha, Uli

I can confirm that GPU load for both AMD and Nvidia is zero, on Windows 7. As others have said they are only running on the CPU.

---

Team USA forum | Team USA page
Always crunching / Always recruiting

I am so drunk last night so I test it again.
TechPowerUp GPU-Z and CPUID HWMonitor shows absolutely no GPU usage but full core CPU usage. This means task not using opencl on gpu.

I suspend the milkyway till all goes good solved applications. I hope Jake take a good sleep and compile the applications.

---