I found a Free antispy (stand alone, no installation) tool for Windows 10, O&O ShutUp10 (chip.de recommendation): https://www.oo-software.com/en/shutup10

---

It is set for new tasks. I get new ones from the main site.

They are two separate projects, you can set BETA to NNT and still get tasks from main and vice versa.

---

For the last couple of days I've had opencl_nvidia_sah tasks running on both gtx980 on this PC (http://setiweb.ssl.berkeley.edu/beta/show_host_detail.php?hostid=61323) and it is suffering from significant stuttering, with on occasions a couple of seconds between moving the mouse and any change on the screen.

The CPU is mostly running stock v8, with the odd resend, six such tasks, one per available core.
All CPU cores are maxed out.

It looks as if the openCL Nvidia app is running mostly on the CPU as I see it is occupying the two "free" cores.

It looks as if the openCL Nvidia app is running mostly on the CPU as I see it is occupying the two "free" cores.

Does your CPU capable to finish MB task in 543.78 seconds?

---

News about SETI opt app releases: https://twitter.com/Raistmer

No, but I think you are missing the point.

My computer is VERY sluggish when running OpenCL/Nvidia tasks, one per GPU.

It has an eight core CPU, with 2 cores "spare", that is not dedicated to BOINC.
There are six v8 task running, as I would expect. These tasks are each occupying a single core, which again is what I would expect.
There are two OpenCL/Nvidia tasks running on the other two cores, and they are each using one complete core. BONIC reports that the default cores per task, which is currently 0.419, but task amanger reports ~1 core per task almost continuously.
Looking at the GPU utilisation, using GPU-Z. This is varying between 55% and 75% - which is more or less what I would expect for a single GPU task on these GTX980s.

A short time ago I was running three v7 MB tasks on each of these GPUs and the total CPU usage from all the tasks was about 10% of a core, and with no apparent lags or stutters.

It looks to me as if the OpenCL Nvidia application needs a lot of CPU assistance to achieve its performance.

Your host with NV GTX980 GPUs has AMD CPU. These AMD CPUs have one little aspect which is very relevant for these calculations : Eight cores have to share four FPUs.

So, if you run more than 4 CPU tasks in parallel you get a timewise penalty for overcommiting some of the CPUs resources. Two tasks will have to use one FPU.

My suggestion would be to not run more than 4 CPU tasks in parallel on such an AMD CPU. That way there should be enough resources left to feed the GPUs.

---

_\|/_
U r s

It looks to me as if the OpenCL Nvidia application needs a lot of CPU assistance to achieve its performance.

My experience, as well.

---

If I can help out by testing something, please let me know.
Available hardware and software is listed in my profile here.

Not sure of there this goes:

Power Mac
Power Mac G4 (4,2; AltiVec)
Operating Sys: Darwin (8.11.0) Sys (10.4.11)
Boinc: (6.12.35)


Name - 06ap11ag.24345.16018.13.47.148.vlar_1
Workunit - 7957457
Created - 15 Jan 2016, 15:23:22 UTC
Sent - 15 Jan 2016, 16:28:40 UTC
Report deadline - 8 Mar 2016, 21:28:22 UTC
Received - 15 Jan 2016, 16:28:59 UTC
Server state - Over
Outcome - Computation error
Client state - Compute error
Exit status - 10 (0xa) Unknown error number
Computer ID - 72412
Run time
CPU time
Validate state - Invalid
Credit - 0.00
Device peak FLOPS - 0.51 GFLOPS
Application version - SETI@home v8 v8.07 powerpc-apple-darwin

Stderr output
<core_client_version>6.12.35</core_client_version>
<![CDATA[
<message>
process got signal 10
</message>
<stderr_txt>

</stderr_txt>
]]>


This is my Intel iMac 14,1
Genuine Intel (i5-4570R CPU @ 2.7 ghz; x86 Family 6 Model 70 Stepping 1)
Intel Iris Pro; OpenCL:1.2
Operating Sys: Darwin 15.2.0 sys (10.11.2)
Boinc: 7.6.22

Name: 06ap11ag.24345.16018.13.47.161.vlar_2
Workunit: 7957470
Created: 15 Jan 2016, 15:23:22 UTC
Sent: 15 Jan 2016, 16:39:15 UTC
Report deadline: 8 Mar 2016, 21:38:57 UTC
Received: 15 Jan 2016, 20:21:49 UTC
Server state: Over
Outcome: Success
Client state: Done
Exit status: 0 (0x0)
Computer ID: 72520
Run time: 1 hours 40 min 11 sec
CPU time: 1 hours 39 min 53 sec
Validate state: Valid
Credit: 117.83
Device peak FLOPS: 4.25 GFLOPS
Application version: SETI@home v8 v8.05 x86_64-apple-darwin
Peak working set size: 124.59 MB
Peak swap size: 2,556.10 MB
Peak disk usage: 0.31 MB

Stderr output
<core_client_version>7.6.22</core_client_version>
<![CDATA[
<stderr_txt>
setiathome_v8 8.00 Revision: 3310XCode LLVM 4.2.1 Compatible Apple LLVM 6.0 (clang-600.0.57) x86_64

libboinc: 7.7.0
libboinc: 7.7.0

Work Unit Info:
...............
WU true angle range is : 0.011545
Optimal function choices:
--------------------------------------------------------
name timing error
--------------------------------------------------------
v_BaseLineSmooth (no other)
v_vGetPowerSpectrum 0.000053 0.00000
sse2_ChirpData_ak8 0.004318 0.00000
v_vTranspose4x16ntw 0.001087 0.00000
AK SSE folding 0.000741 0.00000

Flopcounter: 44697485164990.921875

Spike count: 3
Autocorr count: 1
Pulse count: 11
Triplet count: 2
Gaussian count: 0
14:21:38 (8736): called boinc_finish(0)

</stderr_txt>
]]>


As you can see my Mac PPC doesn’t like 8.0

Charles

Mini bug. The following files are marked executable:

Here:

bl_bkg_804.jpg
bl_bkg_805.jpg

Main:

bl_800.jpg
bl_bkg_800.jpg
seti_800.jpg
setiathome-8.00_AUTHORS
setiathome-8.00_COPYING
setiathome-8.00_COPYRIGHT
setiathome-8.00_README

Urs - I take your point, however I've not seen any similar screen lags previously when running 2xAP per AP + 6MB (v7) on the CPUs, or 3MB per GPU + 6MB (v7) on the CPU, or 2 x AP on each GPU plus 6AP on the CPU.

I have however previously observed that running 1xAP plus 1xMB (v7) on a GPU resulted in a doubling of the execution time for the MB, but no significant impact on the AP.

I have also noticed a big slowdown when an opencl unit is running sometimes borderline freeze-up and the exact same thing boinc is reporting .115 cpu but task manager is showing a full core in use even to the point of 1 of the cpu wu showing 0 cpu usage many times while opencl is running. My cpu is only a quad core so the slowdown is very apparent but the cuda 50 and 42 do just fine with minimal cpu usage.

---

ME AND MY BOY LOOKING FOR E.T.

It looks to me as if the OpenCL Nvidia application needs a lot of CPU assistance to achieve its performance.

No points missed, please re-read: https://setiweb.ssl.berkeley.edu/beta//forum_thread.php?id=2266&postid=55931

---

News about SETI opt app releases: https://twitter.com/Raistmer

Your host with NV GTX980 GPUs has AMD CPU. These AMD CPUs have one little aspect which is very relevant for these calculations : Eight cores have to share four FPUs.

So, if you run more than 4 CPU tasks in parallel you get a timewise penalty for overcommiting some of the CPUs resources. Two tasks will have to use one FPU.

My suggestion would be to not run more than 4 CPU tasks in parallel on such an AMD CPU. That way there should be enough resources left to feed the GPUs.

It's relevant to CPU app performance but unrelevant to logical CPU usage by NV drivers. Most probably that usage purely integer one (busy-wait loop hardly involves any FPU computations).

---

News about SETI opt app releases: https://twitter.com/Raistmer

Urs - I take your point, however I've not seen any similar screen lags previously when running 2xAP per AP + 6MB (v7) on the CPUs, or 3MB per GPU + 6MB (v7) on the CPU, or 2 x AP on each GPU plus 6AP on the CPU.

I have however previously observed that running 1xAP plus 1xMB (v7) on a GPU resulted in a doubling of the execution time for the MB, but no significant impact on the AP.

What tasks create "slugishness" ? VLAR? Midrange?
Did you try to tune available options already?
What CLinfo says about your GPU?

---

News about SETI opt app releases: https://twitter.com/Raistmer

I am also noticing a slowdown, in only one area: Windows Explorer. If I copy file(s), create a folder, delete file(s), the action occurs, but the folder window doesn't update. It will hang for perhaps 15-20 seconds. I have noticed it on this machine, which is the only one I actually use on a regular basis.

I know it's v8 OpenCL doing it because there's no CUDA running (haven't seen a one in a week or so) and if I do the folder window action above and then "Snooze GPU" quickly, the update occurs instantaneously as soon as it's suspended.

Running 100% stock with no app_info.xml; connected to the project and set a queue size and nothing else.

I have also noticed a big slowdown when an opencl unit is running sometimes borderline freeze-up and the exact same thing boinc is reporting .115 cpu but task manager is showing a full core in use even to the point of 1 of the cpu wu showing 0 cpu usage many times while opencl is running. My cpu is only a quad core so the slowdown is very apparent but the cuda 50 and 42 do just fine with minimal cpu usage.

That's exactly why i definitely prefer CUDA apps in any way. They are just not intrusive and the PC is usable as usual. This is not the case with the OpenCl apps. They are heavily intrusive and render the PC partially unusable and are therefor a NOGO - at least - for me.

---

Aloha, Uli

That's exactly why i definitely prefer CUDA apps in any way. They are just not intrusive and the PC is usable as usual. This is not the case with the OpenCl apps. They are heavily intrusive and render the PC partially unusable and are therefor a NOGO - at least - for me.

I may be wrong, but I'm guessing that the excess of OpenCL we've been seeing is because we'd best get used to it: apparently CUDA may be a no-go for the GBT data, as it's going to be mostly VLARs, which as you doubtlessly know choke CUDA for hours but don't bother OpenCL.

'Twould be really neat if the actual cause of this could be found and worked around without having to use OpenCL.



"I CUDA been a contender!"

If I had the cores to spare I would not care but right now I do not have them to spare(upgrades coming in Feb. after I get my refund) the beta opencl v8.06 does seem to need more cpu than previous versions.

---

ME AND MY BOY LOOKING FOR E.T.

I may be wrong, but I'm guessing that the excess of OpenCL we've been seeing is because we'd best get used to it: apparently CUDA may be a no-go for the GBT data, as it's going to be mostly VLARs, which as you doubtlessly know choke CUDA for hours but don't bother OpenCL. (...)

This may well be true, but if my PC is unusable with OpenCL app running, i would just prefer a CUDA app, even if it would be dead slow compared to OpenCL, just because i can REALLY USE my PC, when the app is running.

---

Aloha, Uli

You can add the -use_sleep in the command txt file in setiathome beta folder.

It will lower the CPU usage down to around 20% or lower.