stderr.txt in /slots/

slot 1 : Cuda error 'cudaMalloc((void*) &dev_cx_DataArray' in file 'cuda/cudaAcceleration.cu' in line 360 : out of memory
setiathome_CUDA : CUDA runtime ERROR in device memory allocation, attempt 1 of 6

....
on attempt 6
.....
v8 task detected
setiathome_CUDA : FOund 1 CUDA Device(s):
Device 1 : GeFOrce 320M, 252 MiB, regsPerBlock 16384
computeCap 1.2, multiProcs 6
pciBusID =0, pciSlotID=0
clockRate = 950 Mhz
In cudaAcc_initializeDevice(): Boinc passed DevPref 1
...
Detected Autocorrelation as enabled, size 128k elements
Work Uni info
WU true angle range is : 0.0009111
re-using dev_GaussFitResults array for dev_AutoCorrIn, 4194304 bytes
re-using dev_GaussFitsResults+524288x8 array for dev_AutoCorrOut, 4194304 bytes
a cuFFT plan FAILED, Initiating Boinc temporary exit (180 secs)

---

I was looking for a certain 256mb Windows GPU I thought I saw the other day that was working with CUDA50, but found this instead; cuda/cudaAcceleration.cu' in line 360 : out of memory. Seems that is the Error you get with CUDA42, the other CUDAs on that host are giving slightly different Errors. It would be helpful if another Mac with a 256mb card gave the App a try, just to see if it's universal.

I tried a lower driver version to see if it uses less memory in the Event Log startup screen but couldn't go lower than 5.0.36 without CUDA-Z saying it can't find CUDA. 5.0.36 works on my machine and the Event Log shows a little more memory available. I don't know if that would help or not, http://www.nvidia.com/object/macosx-cuda-5.0.36-driver.html I also ran across some notes on the earlier drivers, if this applies to you or someone else you could give it a try;

• There is a known issue in this release where forcing or allowing the system to go to sleep while running CUDA applications on 2012 MacBook Pro models with automatic graphics switching will cause a system crash (kernel panic). You can prevent the computer from automatically going to sleep by setting the Computer Sleep option slider to Never in the Energy Saver pane of the System Preferences.

• There is a known issue in this release where CUDA applications will not automatically engage the discrete GPU on 2012 MacBook Pro models with automatic graphics switching. To run CUDA applications, it is necessary to uncheck the Automatic Graphics Switching checkbox in the Energy Saver pane of the System Preferences.

I suppose if none of that works the App probably won't work with that GPU.

I was looking for a certain 256mb Windows GPU I thought I saw the other day that was working with CUDA50, but found this instead; cuda/cudaAcceleration.cu' in line 360 : out of memory. Seems that is the Error you get with CUDA42, the other CUDAs on that host are giving slightly different Errors. It would be helpful if another Mac with a 256mb card gave the App a try, just to see if it's universal.

Yes same error !

I tried a lower driver version to see if it uses less memory in the Event Log startup screen but couldn't go lower than 5.0.36 without CUDA-Z saying it can't find CUDA. 5.0.36 works on my machine and the Event Log shows a little more memory available. I don't know if that would help or not, http://www.nvidia.com/object/macosx-cuda-5.0.36-driver.html I also ran across some notes on the earlier drivers, if this applies to you or someone else you could give it a try;

• There is a known issue in this release where forcing or allowing the system to go to sleep while running CUDA applications on 2012 MacBook Pro models with automatic graphics switching will cause a system crash (kernel panic). You can prevent the computer from automatically going to sleep by setting the Computer Sleep option slider to Never in the Energy Saver pane of the System Preferences.

• There is a known issue in this release where CUDA applications will not automatically engage the discrete GPU on 2012 MacBook Pro models with automatic graphics switching. To run CUDA applications, it is necessary to uncheck the Automatic Graphics Switching checkbox in the Energy Saver pane of the System Preferences.

I suppose if none of that works the App probably won't work with that GPU.

I don't want to spend too much time as it's a quite old computer (macbook pro 2010), I will soon have a new one (waiting the new 2016).
I think this GPU will compute more slowly than one skylake CPU core !

Was just a test for you to check if it works with your binaries. I think
you can drop support for such old cards ;)

---

I looked at 3.2, and decided not to go there. My 8800 GT & GTS 250 is working as usual in Darwin 11.4.2, so, it must be something to do with that low memory mobile GPU. The screen lag is horrible running 2 BLC7s, and the GTS 250 finished the Test BLC2 about a minute faster than the CPU. Not much advantage there, might as well just run the CPUs. Probably not a bad thing the lower end Low end cards won't run it. I'm going to finish the tasks it has and then go back to the newer cards.

Yeah, well in my mind NV still wrote off that generation too early. The memory amount considerations are definitely part of the situation (especially Fermi's 64 bit capability). IMO The real pain will come with the plans to similarly deprecate anything Pre-Kepler. That'll leave a lot of actively crunching hardware cold.

Seems there's another mobile GPU giving the Out of Memory Error, and this one has the correct Driver for Darwin 16.0;
NVIDIA GeForce GT 640M (511MB) driver: 5243.34 OpenCL: 1.2
I'd still like to know if these GPUs giving the out of memory error works in Standalone mode or Anonymous platform.
To run in Standalone mode download the KWSN OSX MB Bench v2.01.07 and v8 reference workunit from here, http://lunatics.kwsn.info/index.php?action=downloads;cat=45
Place KWSN OSX MB Bench v2.01.07 in your Home folder, Place the App in KWSN OSX MB Bench v2.01.07/APPS, the Libraries in KWSN OSX MB Bench v2.01.07, and v8 reference workunit in KWSN OSX MB Bench v2.01.07/testWUs. Then open a terminal window, cd to KWSN OSX MB Bench v2.01.07, and run ./Benchmark. It probably wouldn't hurt to place a CPU App in KWSN OSX MB Bench v2.01.07/REF_APPs, but, then you'd have to wait for the CPU App to finish before the GPU App will run.

The CUDA42 App works fine on my 512mb 8800 GT.
You can download the Apps & Libraries from Beta;
CUDA42 http://boinc2.ssl.berkeley.edu/beta/download/setiathome_8.11_x86_64-apple-darwin__cuda42_mac
libcudart.4.2.dylib http://boinc2.ssl.berkeley.edu/beta/download/libcudart.4.2.dylib
libcufft.4.2.dylib http://boinc2.ssl.berkeley.edu/beta/download/libcufft.4.2.dylib
libtlshook.dylib http://boinc2.ssl.berkeley.edu/beta/download/libtlshook.dylib

CUDA75 http://boinc2.ssl.berkeley.edu/beta/download/setiathome_8.11_x86_64-apple-darwin__cuda75_mac
libcudart.6.5.dylib http://boinc2.ssl.berkeley.edu/beta/download/libcudart.6.5.dylib
libcufft.6.5.dylib http://boinc2.ssl.berkeley.edu/beta/download/libcufft.6.5.dylib

Thanks...I'm in vacation...you'll have to wait until November.
Sorry

---

Soon there will hopefully be a new Mac OpenCL App on Main that does work with the 320M, even though most people wouldn't care to use it, https://setiweb.ssl.berkeley.edu/beta/results.php?hostid=59702
Fortunately the App works just as well on the Newer nVidia GPUs.

Thanks...I will test it.
My new Macbook Pro 2016 will arrive before christmas :)

---

First wu computed with my old macbook pro. Seems to works fine :)

https://setiweb.ssl.berkeley.edu/beta/workunit.php?wuid=9087792

---

I have now my new Macbook Pro with a Radeon 450 GPU running stock apps.

https://setiathome.berkeley.edu/results.php?hostid=8177300&offset=0&show_names=0&state=4&appid=

All wu using opencl_intel_gpu_sah are inconclusive.

Do you want I run beta apps ?

---

All wu using opencl_intel_gpu_sah are inconclusive.

Do you want I run beta apps ?

Beta would be good too, but even better if you would run alpha builds from Urs.
We need alpha testers with iGPU-capable Macs.

---

SETI apps news
We're not gonna fight them. We're gonna transcend them.

All wu using opencl_intel_gpu_sah are inconclusive.

Do you want I run beta apps ?

Beta would be good too, but even better if you would run alpha builds from Urs.
We need alpha testers with iGPU-capable Macs.

Is this the location to download alpha binaries ?
http://lunatics.kwsn.info/index.php?action=downloads;cat=30

---

All wu using opencl_intel_gpu_sah are inconclusive.

Do you want I run beta apps ?

Beta would be good too, but even better if you would run alpha builds from Urs.
We need alpha testers with iGPU-capable Macs.

Is this the location to download alpha binaries ?
http://lunatics.kwsn.info/index.php?action=downloads;cat=30

I'm not sure Urs published it anywhere already.
Better contact with him directly: Urs Echternacht

---

SETI apps news
We're not gonna fight them. We're gonna transcend them.

If you need a couple new Apps to try I have a New ATI SoG 3587 which runs about the same as the just released r3556, and a New Intel iGPU 3586. The ATI App works fine on my ATI card and the iGPU App works fine on my nVidia cards. I don't have an actual Intel iGPU to test on. Oh, the new ATI App appears to have fixed the Progress bar anomaly, it doesn't jump around any more....nice.

If you need a couple new Apps to try I have a New ATI SoG 3587 which runs about the same as the just released r3556, and a New Intel iGPU 3586. The ATI App works fine on my ATI card and the iGPU App works fine on my nVidia cards. I don't have an actual Intel iGPU to test on. Oh, the new ATI App appears to have fixed the Progress bar anomaly, it doesn't jump around any more....nice.

Is it available from here ?
http://www.arkayn.us/forum/index.php?topic=191.0

---

I just posted it here, http://www.arkayn.us/forum/index.php?topic=191.msg4501#msg4501
If you have any of the older SETI@home v8 v8.00 (opencl_ati5_mac) x86_64-apple-darwin tasks they will not be saved. It would be best to suspend the SETI@home v8 v8.20 (opencl_ati5_SoG_mac)
x86_64-apple-darwin tasks and finish the older opencl_ati5_mac tasks before making the change. The other tasks Should be carried over.
If the Intel App still gives Inconclusive results you can just plug in any newer versions into the same app_info.xml.

I just posted it here, http://www.arkayn.us/forum/index.php?topic=191.msg4501#msg4501
If you have any of the older SETI@home v8 v8.00 (opencl_ati5_mac) x86_64-apple-darwin tasks they will not be saved. It would be best to suspend the SETI@home v8 v8.20 (opencl_ati5_SoG_mac)
x86_64-apple-darwin tasks and finish the older opencl_ati5_mac tasks before making the change. The other tasks Should be carried over.
If the Intel App still gives Inconclusive results you can just plug in any newer versions into the same app_info.xml.

Thanks. I've selected 'No new taks' and will wait until they all completed.
I don't run CPU tasks....too slow compared to GPU (CPU : 4 hours / Intel GPU : 1 hour / ATI GPU : 30 min) on a Macbook Pro 2016 with i7 core.

---

I don't run CPU tasks....too slow compared to GPU (CPU : 4 hours / Intel GPU : 1 hour / ATI GPU : 30 min) on a Macbook Pro 2016 with i7 core.

Is that CPU time while running the iGPU? If so, it would be much faster with the iGPU not running tasks- shared caches & thermal/power limits result in significant reduction in CPU performance when using the iGPU to crunch.
And 4 hours on the CPU would still outperform the iGPU at 1 hour- 4 Cores with HyperThreading? The CPU would produce significantly more work per hour than the iGPU, even with 1 core dedicated to the GPU.

EDIT-
Although that does assume no thermal throttling.
Many of the current super slim laptops & notebooks can't handle extended crunching without throttling as their cooling ability is very much limited by their slim design.

---

Grant
Darwin NT

Thanks. I've selected 'No new taks' and will wait until they all completed.
I don't run CPU tasks....too slow compared to GPU (CPU : 4 hours / Intel GPU : 1 hour / ATI GPU : 30 min) on a Macbook Pro 2016 with i7 core.

The AVX App is much faster than the Stock OSX CPU App. With the Newer i7 CPUs the run time for a BLC5 is just over an Hour. Even my Old Core2s will finish a BLC5 in around 2 hours using the SSE41 App. With the Boost function the times are shorter when running fewer CPU tasks. I would recommend running about 50% CPU tasks, that would be 4 CPU tasks with a 8 core CPU. I think you'll be surprised how fast the newer CPUs run with the correct App. It might be slower when using the Intel iGPU, we'll just have to see. Yep, 4 hours is pretty bad.

Is that CPU time while running the iGPU? If so, it would be much faster with the iGPU not running tasks- shared caches & thermal/power limits result in significant reduction in CPU performance when using the iGPU to crunch.
And 4 hours on the CPU would still outperform the iGPU at 1 hour- 4 Cores with HyperThreading? The CPU would produce significantly more work per hour than the iGPU, even with 1 core dedicated to the GPU.

EDIT-
Although that does assume no thermal throttling.
Many of the current super slim laptops & notebooks can't handle extended crunching without throttling as their cooling ability is very much limited by their slim design.

yes CPU time while running iGPU.
I have suspended iGPU tasks to check....hope the laptop will cool a little because it's pretty hot ! (no problem with such freezing weather here ;)

---

yes CPU time while running iGPU.
I have suspended iGPU tasks to check....hope the laptop will cool a little because it's pretty hot ! (no problem with such freezing weather here ;)

Running the optimised AVX application will bump the heat up even more- it really pushes the CPU.
If you do as TBar suggested and just run 4 CPU instances with no iGPU processing, you should get plenty of throughput, and help keep the temperatures down (at least compared to what they would be like crunching 8 WUs at a time).

---

Grant
Darwin NT