Message boards :
Number crunching :
App_info help for OSX apps?
Message board moderation
Author | Message |
---|---|
Gecko Send message Joined: 17 Nov 99 Posts: 454 Credit: 6,946,910 RAC: 47 |
Looking for some help if any fellow OSX gents / developers / testers can spare a few minutes? I've been out of the OSX scene for a lil while and lots has changed the past few years. Got an older laptop with gforce320m running Mavericks that want to run project. Played around with Cuda 5.5 & 6 w/ Cuda42 ap but WUs pause or get run errors. The 256MB video memory may not be enough. I read some threads @ 2016 with a couple others that pointed out same. Not sure if anyone ever got Cuda42 ap working on the older Tesla CC 1.2 cards? Right now, just running CPU app below and it works like a charm on Macbook Air w/ Intel SU9400 Core 2 Duo 1.40. CPU supports sse41 but not AVX. MBv8_8.22r3711_sse41_x86_64-apple-darwin This OpenCL ap from main appears to run also. SETI@home v8 v8.19 (opencl_nvidia_mac_old) x86_64-apple-darwin Any newer MB ap builds avail for CPU or Open CL? Can anyone share a recent OSX app_info w/ a matching Open CL section for main ap above? Much Thanks! -Gecko |
TimeLord04 Send message Joined: 9 Mar 06 Posts: 21140 Credit: 33,933,039 RAC: 23 |
Look into TBar's Apps: CUDA90 for Mac - (Sierra and High Sierra...) The OS you mention, (Mavericks), is just too old to do anything with for the Anonymous Platform that TBar programs for... As for CUDA42, that's ANCIENT now... With our newer task loads on Non-Arecibo work, even CUDA50 has been reported to have Errors. (Windows/Linux based Apps...) TBar currently is working on new OpenCL Applications in addition to his newest v0.98b1 CUDA90 App for MacOS. In addition to Mavericks being too old now to work with, (outside of the CPU Apps), your GeForce 320m I believe would be unusable for the Anonymous Platform CUDAxx Apps. MacVidCards sells Pascal NVIDIA Cards IF you are interested in staying with NVIDIA. Pascal would be the GTX-10x0 Line. Pascal Cards require MacOS Sierra 10.12.6 or High Sierra 10.13.x to run TBar's CUDA90 Apps. IF you are interested in MacOS Mojave, then NVIDIA is out of the question, (pretty much), since Web Drivers DON'T exist for the Maxwell and Pascal Cards. Kepler Cards STILL run in Mojave and Catalina, but your best bet for Mojave or Catalina would be to go AMD/ATI and go with the Sapphire Radeon Pulse RX-580 8GB Card by MacVidCards. I recommend MacVidCards, because they Pre-Flash the GPU for MacOS for you. This gives the GPU "Mac EFI Boot ROM" and gives you the ability to have the Apple Boot Screen. Stock or "Off the Shelf" GPUs that are NOT flashed will NOT give you the Apple Boot Screen. (Classic Mac Pro Systems come to mind for this issue, as they are the easiest to Upgrade for employing Discrete GPUs.) One can also build a HIGH POWER Hackintosh instead of investing in an older Classic Mac Pro System... Keeping in mind what I've mentioned above. I'm sure TBar will chime in here. TimeLord04 TimeLord04 Have TARDIS, will travel... Come along K-9! Join Calm Chaos |
TimeLord04 Send message Joined: 9 Mar 06 Posts: 21140 Credit: 33,933,039 RAC: 23 |
An example of a High Power Hackintosh: See my Computer List and look at Computer ID: 7952666 This is my iMac 18,3 - Profiled Hackintosh i7 7700K 4.2GHz System with MacVidCards' GTX-1070 8GB Card and a Secondary EVGA GTX-1050 2GB Card. I'm on High Sierra 10.13.6 and run TBar's v0.97 CUDA90 App. Due to how I run my Systems, I CANNOT make good use of TBar's v0.98b1 CUDA90 App. TimeLord04 TimeLord04 Have TARDIS, will travel... Come along K-9! Join Calm Chaos |
Gecko Send message Joined: 17 Nov 99 Posts: 454 Credit: 6,946,910 RAC: 47 |
TimeLord, Thank you for the excellent tips! This laptop is old but like new condition so I'll let it crunch with the best it can handle. Figured CUDA was off the table, but OpenCL maybe? However, I'm specing out a new iMac which will allow to play with the newer Radeon apps. |
TimeLord04 Send message Joined: 9 Mar 06 Posts: 21140 Credit: 33,933,039 RAC: 23 |
Glad to be of help, Gecko. It took me just about a year to get all the parts for my Hackintosh. Got her built in late December of 2018. The March 2016 Creation Date for the BOINC Computer ID for this Hackintosh actually goes back to my first Hackintosh, test System. The first Hackintosh ran El Capitan and had two EVGA GTX-750TI SC 2GB Cards. Now that the war between NVIDIA and Apple is over, and NVIDIA has stated that NO Web Drivers will be released for Mojave, or later, AND CUDA Driver 418.163 IS THE LAST MacOS CUDA Driver, I will be saving my coins to replace the NVIDIA cards with the MacVidCards' Sapphire Radeon Pulse RX-580 Cards. This means that I will be employing TBar's OpenCL App, when I make this Card change. It may take me a couple years to make this happen, though. (Budget constraints...) To be honest, most people here in Number Crunching have been pushing less expensive base PC Systems with HIGH End NVIDIA RTX-20x0 Cards. THEN, install your favorite flavor of Linux, (Ubuntu or Mint seem to be popular here), and run TBar's CUDA90 or CUDA101 Apps for Linux. With the Linux CUDA90 and CUDA101 Apps on the RTX Cards, Crunch Times are dropped to LESS than 2 Min per Unit. Unfortunately, I CANNOT make sense, nor full use, of Linux as a daily driver for me. I CAN get Ubuntu installed on a PC, BUT for the life of me, I CANNOT make sense, nor utilize to capacity the Command Line functions in Terminal. I'm a GUI guy, and am sticking to MacOS. Hackintosh is my thing. IF you can muster your way through Linux Command Line, I do recommend getting hold of an inexpensive base PC and a High End RTX-20x0 Card and go Linux with TBar's CUDA90 or CUDA101 AIO Package. Good Luck with whatever you decide to do. TimeLord04 TimeLord04 Have TARDIS, will travel... Come along K-9! Join Calm Chaos |
Wiggo Send message Joined: 24 Jan 00 Posts: 36828 Credit: 261,360,520 RAC: 489 |
Just having a quick look at the details of that GeForce 320m GPU (and it's MCP89 core) which is just an enhanced core structure that dates back to a pre-supported time (aka: GeForce 6xxx days), but nothing I can find says anything about what type of enhancements it carried over earlier versions and it also requires shared system memory so I'd very much doubt that you'd get it to work anyway (maybe it just has a greater shared memory capacity as I doubt that it would've got a die shrink at that late stage of life). You should be able to run 1 CPU task at a time though while using it, maybe 2 if not, but do check for any large discrepancies between the actual run time and the CPU time of completed tasks if running 2. Cheers. |
TimeLord04 Send message Joined: 9 Mar 06 Posts: 21140 Credit: 33,933,039 RAC: 23 |
Just having a quick look at the details of that GeForce 320m GPU (and it's MCP89 core) which is just an enhanced core structure that dates back to a pre-supported time (aka: GeForce 6xxx days), but nothing I can find says anything about what type of enhancements it carried over earlier versions and it also requires shared system memory so I'd very much doubt that you'd get it to work anyway (maybe it just has a greater shared memory capacity as I doubt that it would've got a die shrink at that late stage of life). Wiggo, I had a friend look up the "300"-Series. It's Tesla based, which is Pre-Fermi. I don't think it's usable for Crunching. [EDIT:] NOT with 252MB of System RAM diverted to it... I think only his CPU is usable for Crunching. TL TimeLord04 Have TARDIS, will travel... Come along K-9! Join Calm Chaos |
Wiggo Send message Joined: 24 Jan 00 Posts: 36828 Credit: 261,360,520 RAC: 489 |
I'm almost sure that that was what I was stating TL as a GeForce 8xxx was the minimum required for GPU processing. ;-)Just having a quick look at the details of that GeForce 320m GPU (and it's MCP89 core) which is just an enhanced core structure that dates back to a pre-supported time (aka: GeForce 6xxx days), but nothing I can find says anything about what type of enhancements it carried over earlier versions and it also requires shared system memory so I'd very much doubt that you'd get it to work anyway (maybe it just has a greater shared memory capacity as I doubt that it would've got a die shrink at that late stage of life).Wiggo, [edit] In fact you needed at least a GeForce 8300GS so a G8x or greater core for GPU processing. Cheers. |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
Looking for some help if any fellow OSX gents / developers / testers can spare a few minutes?If you ever do get it to work, the 320m is much slower than the CPU. It's even slower than the 9400 with 256mb, and it uses the CPU to be slower than the CPU...not good. It would be best to just use the CPU. It's really dependent on the OS and even the point version of the OS as to rather it will work or not, and those older cards will work better with cuda over opencl. Just make sure you have this driver installed in Mavericks, https://www.nvidia.com/en-us/drivers/cuda/macosx-cuda-6-0-51-driver/ The Control panel will try to update it to 6.5.xx, don't use the 6.5 driver. The error you are receiving is "Library not loaded: @rpath/libcudart.dylib", which could be a problem with the driver, or, you didn't copy the cuda libraries to the setiathome.berkeley.edu folder. Make sure the three libraries are in the setiathome.berkeley.edu folder; libcudart.4.2.dylib libcufft.4.2.dylib libtlshook.dylib You should be getting something like 'out of memory' instead of 'library not loaded'. Or, it might even work. |
Gecko Send message Joined: 17 Nov 99 Posts: 454 Credit: 6,946,910 RAC: 47 |
Thank you Tbar, TimeLord and Wiggo. The320m is Tesla compute capability 1.2 so the arch + Mavericks 10.9.5 is compatible up to Cuda 6 driver I've played around with drivers 5.5 and 6 but get same errors. The libraries were in the seti project folder within Boinc folder . Should I have put them in Boinc folder instead? I was also getting "task pause" errors, Would run 18 secs, then pause with error, then retry in 2-3 mins...then pause again. Endless cycle. Any thoughts about downgrading OS to Snow Leopard and running Cuda 4.2 driver? : / |
Gecko Send message Joined: 17 Nov 99 Posts: 454 Credit: 6,946,910 RAC: 47 |
Bump. Tbar, please see question in my post below. Thx! |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
I would just give it up if i were you. Why would you run both the CPU & GPU to finish a task a slower than it would be if you just used the CPU? It's a complete waste of time and energy. The OpenCL App would probably work, but would be Slightly slower than CUDA, and use more CPU time. It would be faster and more efficient to just use the CPU. Look at these times from El Capitan and compare them to how fast your CPU would complete the Same task by itself running the CPU App, https://setiweb.ssl.berkeley.edu/beta/results.php?hostid=71749 Name: 18no10ab.8437.6611.13.47.35_2 WU true angle range is : 0.438398 App: SETI@home v8 v8.19 (opencl_nvidia_mac_old) Run time: 16 hours 8 min 40 sec CPU time: 9 hours 12 min 8 sec Name: 19fe20aa.28891.24198.10.37.143_0 WU true angle range is : 0.441102 App: SSE4.1xj OS X 64bit Build 3711 Run time: 4 hours 0 min 2 sec CPU time: 3 hours 58 min 40 sec I would just leave the Machine running Mavericks and use the CPU App as Old & Weak GPUs are a waste of time. |
Gecko Send message Joined: 17 Nov 99 Posts: 454 Credit: 6,946,910 RAC: 47 |
I would just give it up if i were you. Wow thanks for this. I see what you mean, lol. |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
I looked into why you were getting the Library Error instead of the Out of Memory Error with the CUDA App. Seems a while back the Server at C.A was changed and all the Downloads had to be reloaded, an older copy of the CUDA42 App was loaded, and that copy didn't have the Execute path set. It has been replaced with a copy that has the correct path now. I still think you will get the Out of Memory Error with that low power 256mb GPU, but, it will work for everyone else. I setup my Mac Pro 3,1 with the App just to see how it would do with a GTS250. Seems the CPU is still faster than a GTS250 on the VLARs, but the non-VLARs are faster on the GPU. I'd forgotten about the screen lag with the VLARs on the Baseline CUDA Apps, it's still there and quite annoying. Huge difference between the LapTop 320m and a desktop 250. I'll try the OpenCL App in a while and see how it runs, https://setiathome.berkeley.edu/results.php?hostid=7199204 |
Gecko Send message Joined: 17 Nov 99 Posts: 454 Credit: 6,946,910 RAC: 47 |
I looked into why you were getting the Library Error instead of the Out of Memory Error with the CUDA App. Seems a while back the Server at C.A was changed and all the Downloads had to be reloaded, an older copy of the CUDA42 App was loaded, and that copy didn't have the Execute path set. It has been replaced with a copy that has the correct path now. I'll try the OpenCL App in a while and see how it runs. Sorry for the late reply. Was out of town a few days. I'm getting a 404 error when trying to DL the corrected file. http://www.arkayn.us/lunatics/setiathome_x41zi_x86_64-apple-darwin_cuda42.zip I don't mind giving CUDA42 another try just for heck of it. The r3711 SSE4.1 CPU app sure works good though on this lil 1.4ghz core 2 duo Did you ever re-try the OpenCL app on your GTS250? |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
The new file is a 7zip file, try, http://www.arkayn.us/lunatics/setiathome_x41zi_x86_64-apple-darwin_cuda42.7z The lower tasks listed as (opencl_nvidia_mac_old) are the ones using opencl, quite a bit slower than with cuda 4.2. The files at the top were run using the CUDA Special App on a 960...much faster, https://setiathome.berkeley.edu/results.php?hostid=7199204 |
Gecko Send message Joined: 17 Nov 99 Posts: 454 Credit: 6,946,910 RAC: 47 |
The new file is a 7zip file, try, http://www.arkayn.us/lunatics/setiathome_x41zi_x86_64-apple-darwin_cuda42.7zThank you Tbar. I installed CUDA 6.0.51 & the app. Unfortunately, getting same error as before. Thu Feb 27 21:03:23 2020 | SETI@home | Task 26fe20ac.13161.4566.10.37.159.vlar_0 postponed for 180 seconds: CuFFT Plan Failure, temporary exit I freed-up 1 of the 2 physical cores also, but still same problem. Any ideas?... maybe cconfig_xml flags that might help? |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
CuFFT Plan Failure, is the Error you get when you don't have enough vram. People get the same error when trying to run the Special App on a 1 GB card, or run too many webpages with a 2 GB card. You might get it to work running it as a headless machine, and networking in to start boinc, but, I don't remember anyone ever trying that. Best thing to do is get an old Classic Mac Pro that can run High Sierra, install a 1080Ti, and run the Special App, like this machine, https://setiathome.berkeley.edu/results.php?hostid=6796479&offset=1140 It's not really running the Stock CUDA App, I set it up that way back when Anonymous platform wasn't working and decided to just leave it. I plan on leaving it that way until the GPUs meltdown, or it's just not economical to run the Pascal cards. |
Gecko Send message Joined: 17 Nov 99 Posts: 454 Credit: 6,946,910 RAC: 47 |
CuFFT Plan Failure, is the Error you get when you don't have enough vram. You might get it to work running it as a headless machine, and networking in to start boinc, but, I don't remember anyone ever trying that.Thank you for confirming my suspicions. Oh well, I had low expectations of this working on this 2010 model. The laptop is like-new, but ancient, and I try to let old rigs do what they can. Used to have a 2003 G4 MDD PPC 1.25ghz that crunched one of Alex's old AKv8 sse3 apps up until @ 2013. I'll look into an older Mac Pro just for fun. Thanks again for trying to help. |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.