Message boards :
Number crunching :
Linux CUDA 'Special' App finally available, featuring Low CPU use
Message board moderation
Previous · 1 . . . 39 · 40 · 41 · 42 · 43 · 44 · 45 . . . 83 · Next
Author | Message |
---|---|
Brent Norman Send message Joined: 1 Dec 99 Posts: 2786 Credit: 685,657,289 RAC: 835 |
Speaking approval as a main stock app, I think the ability to read command lines from mbcuda.cfg or cmdline_x41p.txt will need to be addressed since there is no app_info.xml file as a stock app to do any changes. The unroll is not implemented in any of the file interfaces. As for 1Mb cards, couldn't that be coded? If Vram<1200mb -> set unroll=1 |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
As for 1Mb cards, couldn't that be coded? If Vram<1200mb -> set unroll=1 Conceptually things like this can be done as things are made more generalised, though it is complicated by a number of factors. These include that since VRAM was virtualised after original XP driver models, on all the platforms, the memory reporting virtually doesn't really match physical things much anymore. It was much easier when GPUs could only really do one thing at a time. "Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions. |
Jeff Buck Send message Joined: 11 Feb 00 Posts: 1441 Credit: 148,764,870 RAC: 0 |
Ah, that's very enlightening. I had somehow gotten the impression that he was still (slowly) working out the bugs at his end and it was just a Windows version that he wasn't addressing. So now I know better!Problems with restarts have existed forever with Petri's App. Since his tasks finish in under a few minutes I don't think he uses checkpoints.But he's not just coding for himself. Or is he? Fortunately that area is my kindof speciality, and that becomes more feasible now the pulse thing is addressed. Time to work on this has been a problem for me, though am expecting things to get better soon, especially since work is drying up, and I have better internet.While I'm certainly eagerly awaiting a Windows version of the Special App (I view my Linux excursion as merely a temporary visit to a carnival funhouse), I surely hope you don't have to end up in the unemployment line to make it happen. ;^) |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
While I'm certainly eagerly awaiting a Windows version of the Special App (I view my Linux excursion as merely a temporary visit to a carnival funhouse), I surely hope you don't have to end up in the unemployment line to make it happen. ;^) Oh that's on the cards anyway, partially intentionally, partially the local economy not doing so well. Quite sick of working for 'the man', I've elected to study updated web development, and freelance tech help around the neighbourhood like I used to as a kid, adding the web dev stuff to the portfolio as I go with that. There'll be some short term pain, but likely a lot more time to dedicate to what I want to do over time. "Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions. |
Jeff Buck Send message Joined: 11 Feb 00 Posts: 1441 Credit: 148,764,870 RAC: 0 |
Sounds like a great plan, if you can carry it through. Good luck!While I'm certainly eagerly awaiting a Windows version of the Special App (I view my Linux excursion as merely a temporary visit to a carnival funhouse), I surely hope you don't have to end up in the unemployment line to make it happen. ;^) |
Stephen "Heretic" Send message Joined: 20 Sep 12 Posts: 5557 Credit: 192,787,363 RAC: 628 |
. . Hi Jeff, . . For me the revision below is more suitable ...
. . But while I am also waiting breathlessly for a windows version I certainly don't want any developers to suffer in their careers over the matter. Stephen :) |
tazzduke Send message Joined: 15 Sep 07 Posts: 190 Credit: 28,269,068 RAC: 5 |
Greetings All Well here's hoping things fall your way Jason. Well I must be one of the many that have crossed over to Linux and have had a excellent run, either running normal Cuda (a few years ago), the SOG and now the Cuda Special App. Seeing how I am not phased if the cache runs dry or mucking around with reschedulers. I have a happy medium and I like to keep that. Now if I only could ween the rest of the family off the Windows, I would be happy. Back to switching my main linux cruncher back onto the Beta, as I see we are stilling testing, forgot to change back when I switched out GPU's. Cheers. |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
Back to switching my main linux cruncher back onto the Beta, as I see we are stilling testing, forgot to change back when I switched out GPU's.Since there are at least Two known problems with the Best signals, any attempt at Beta has been pushed back. So, running the App at Beta is probably not needed at this time if you'd rather do something else, thanks for the help though. Between the known problems and jabs at the developers OS, who knows when the App will ever make it to Main. |
tazzduke Send message Joined: 15 Sep 07 Posts: 190 Credit: 28,269,068 RAC: 5 |
Evening TBar Okay, switching back to Main. Regards |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
BTW, anyone know the specifics on how many machines M$ allows you to install Window 8? I think I read where you could install it on Two machines as long as you only used One machine at a time. I can't remember where I saw that though. The "How to Run a Portable Version of Windows from a USB Drive" looks interesting, but I'd rather just install it on the HDD of the machine running the ATI cards where I might actually use it. |
Tom M Send message Joined: 28 Nov 02 Posts: 5124 Credit: 276,046,078 RAC: 462 |
BTW, anyone know the specifics on how many machines M$ allows you to install Window 8? My impression from installing several different versions of windows in the last 6 months is they are using your motherboard, cpu, bios, hard disk etc to "fingerprint" your computer and associate it with the product code you have registered. I think this is true across Win7 to Win10. It is true, on say Microsoft Office, you can have a single copy on the desktop and under the same product code have one on your laptop. I have been seeing Windows 7 for low prices and actually ran across someone selling Win 10 at a price competitive with Win 7. If you have two Win 8 product codes, you can, of course, install two copies off one media. HTH, but I doubt it. :( Tom A proud member of the OFA (Old Farts Association). |
Mike Send message Joined: 17 Feb 01 Posts: 34258 Credit: 79,922,639 RAC: 80 |
BTW, anyone know the specifics on how many machines M$ allows you to install Window 8? You can run windows only on one machine unless you bought a multi license. It generates a code on startup and each hardware change changes this code also. It affects Mobo, CPU, GPU and HD. With each crime and every kindness we birth our future. |
rob smith Send message Joined: 7 Mar 03 Posts: 22202 Credit: 416,307,556 RAC: 380 |
As Mike says - unless you bought a "multi computer" license you are only allowed to install it on ONE computer. Bob Smith Member of Seti PIPPS (Pluto is a Planet Protest Society) Somewhere in the (un)known Universe? |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
I tracked down the license and it says you have to deactivate it on the first computer before installing it on a second. No great loss I suppose since I don't intend on using it on the first computer anyway. I might keep it around long enough to test the next nVidia SoG App, then install it on the other machine. It's nice you don't have that problem with Linux. Having to buy a copy of Windows for a few machines just to run a Science Program could get expensive. Why on Earth should I have to spend money on an OS just to spend even more money running just a Science Program on it. I dunno, just doesn't make sense to me, I suppose it makes cents to someone. |
rob smith Send message Joined: 7 Mar 03 Posts: 22202 Credit: 416,307,556 RAC: 380 |
Because M$ are greedy wotzitz and want their $$$ for every computer regardless of its use. Bob Smith Member of Seti PIPPS (Pluto is a Planet Protest Society) Somewhere in the (un)known Universe? |
Jeff Buck Send message Joined: 11 Feb 00 Posts: 1441 Credit: 148,764,870 RAC: 0 |
Okay, just in case the waters surrounding the Reported Gaussian vs. Best Gaussian were starting to become clearer, let's see if I can muddy them back up a bit. Out of curiosity, I decided to try to see just how high a percentage of tasks that I've processed recently have mis-matched Gaussians and Best Gaussians. I ran an analysis on all of the tasks in my archives for the month of June for all my hosts and got the following results: Total number of tasks processed with some version of the Petri/TBar Special App: 42,438 Number of those tasks that had Reported Gaussians: 4,935 (11.6%) Number with Reported Gaussians where the Best Gaussian did not match one of the reported ones: 135 (2.7%, or 0.3% of total tasks) Of remaining tasks w/o Reported Gaussians (37,503), number that had Peak=0: 23,442 (62.5%, or 55.2% of total) Total number of tasks processed with some version of the OpenCL SoG App (mostly, or perhaps all, r3584): 11,831 Number of those tasks that had Reported Gaussians: 1,413 (11.9%) Number with Reported Gaussians where the Best Gaussian did not match one of the reported ones: 22 (1.6%, or 0.2% of total tasks) Of remaining tasks w/o Reported Gaussians (10,418), number that had Peak=0: 5,503 (52.8%, or 46.5% of total) Now, to my mind, the numbers for the Special App are encouraging in that the overall percentage of tasks where the Best Gaussian doesn't match any of the reported Gaussians seems extremely low. What struck me as odd, though, following our discussions about which apps follow which code paths when it comes to determining Best Gaussian, is that the SoG app also shows the discrepancy, though to a lesser degree. Assuming that there might be a flaw in my analysis, I went through and located some of those 22 that didn't match. Here's a sample of the relevant portions of the task detail for a couple. I should be able to pull more if anybody wants to see them. Task 5835902425 - just reported yesterday, so still on the server for a bit longer. SSE3xj Win32 Build 3584 ... Pulse: peak=2.760845, time=55.88, period=1.069, d_freq=1418896636.33, score=1.024, chirp=13.642, fft_len=64 D: threshold 0.01462163; unscaled peak power: 0.01487383 exceeds threshold for 1.725% Pulse: peak=2.740437, time=55.88, period=1.069, d_freq=1418896636.21, score=1.016, chirp=16.371, fft_len=64 D: threshold 0.014575; unscaled peak power: 0.01474594 exceeds threshold for 1.173% Spike: peak=24.41263, time=88.92, d_freq=1418897559.1, chirp=42.693, fft_len=32k Spike: peak=24.32981, time=88.92, d_freq=1418897559.11, chirp=42.767, fft_len=32k Gaussian: peak=2.973351, mean=0.5321659, ChiSq=1.414494, time=94.79, d_freq=1418891937.93, score=0.3563112, null_hyp=2.292758, chirp=52.299, fft_len=16k Triplet: peak=9.929982, time=66.85, period=2.333, d_freq=1418895023.54, chirp=57.469, fft_len=512 Best spike: peak=24.41263, time=88.92, d_freq=1418897559.1, chirp=42.693, fft_len=32k Best autocorr: peak=17.53636, time=33.55, delay=3.3631, d_freq=1418897204.28, chirp=21.455, fft_len=128k Best gaussian: peak=2.516772, mean=0.4707989, ChiSq=1.323055, time=89.76, d_freq=1418898860.6, score=0.8768578, null_hyp=2.265836, chirp=-34.812, fft_len=16k Best pulse: peak=2.760845, time=55.88, period=1.069, d_freq=1418896636.33, score=1.024, chirp=13.642, fft_len=64 Best triplet: peak=9.929982, time=66.85, period=2.333, d_freq=1418895023.54, chirp=57.469, fft_len=512 Spike count: 2 Autocorr count: 0 Pulse count: 2 Triplet count: 1 Gaussian count: 1 Task 5792349744 - reported June 7th, so long gone from the server, but I can supply the full Stderr if desired SSE3xj Win32 Build 3584 ... Spike: peak=25.5063, time=20.13, d_freq=1419891401.01, chirp=10.051, fft_len=128k Spike: peak=26.94442, time=20.13, d_freq=1419891401.01, chirp=10.055, fft_len=128k Spike: peak=26.81781, time=20.13, d_freq=1419891401.01, chirp=10.059, fft_len=128k Spike: peak=25.12685, time=20.13, d_freq=1419891401.01, chirp=10.062, fft_len=128k Autocorr: peak=18.17017, time=46.98, delay=0.97044, d_freq=1419891672.72, chirp=-19.274, fft_len=128k Autocorr: peak=17.98606, time=46.98, delay=0.97044, d_freq=1419891671.68, chirp=-19.296, fft_len=128k Autocorr: peak=17.92366, time=87.24, delay=6.1074, d_freq=1419890490.43, chirp=-23.93, fft_len=128k Autocorr: peak=18.01424, time=87.24, delay=6.1074, d_freq=1419890490.27, chirp=-23.932, fft_len=128k Spike: peak=25.76125, time=60.4, d_freq=1419896740.37, chirp=22.839, fft_len=128k Spike: peak=26.84739, time=60.4, d_freq=1419896740.37, chirp=22.843, fft_len=128k Spike: peak=26.66767, time=60.4, d_freq=1419896740.37, chirp=22.847, fft_len=128k Spike: peak=25.17525, time=60.4, d_freq=1419896740.37, chirp=22.85, fft_len=128k Gaussian: peak=3.449904, mean=0.5257975, ChiSq=1.390573, time=88.08, d_freq=1419890600.86, score=1.034913, null_hyp=2.292815, chirp=74.889, fft_len=16k Best spike: peak=26.94442, time=20.13, d_freq=1419891401.01, chirp=10.055, fft_len=128k Best autocorr: peak=18.17017, time=46.98, delay=0.97044, d_freq=1419891672.72, chirp=-19.274, fft_len=128k Best gaussian: peak=4.452988, mean=0.5278533, ChiSq=1.157371, time=81.37, d_freq=1419890059.41, score=1.830914, null_hyp=2.212165, chirp=18.308, fft_len=16k Best pulse: peak=9.583758, time=49.6, period=3.12, d_freq=1419893416.48, score=0.9969, chirp=36.896, fft_len=128 Best triplet: peak=0, time=-2.121e+011, period=0, d_freq=0, chirp=0, fft_len=0 Spike count: 8 Autocorr count: 4 Pulse count: 0 Triplet count: 0 Gaussian count: 1 So, any thoughts.......anyone? |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
Thanks. Still need to read it... Repo version is in sync enough?
Yes, it's possible. On early stages of AstroPulse GPU development I always got invalids with unroll more than 10 for NV GPU of that time (9600GSO perhaps or 8600). No driver restarts just invalid signals. That's how overload showed itself there. Maybe smth similar in your case. SETI apps news We're not gonna fight them. We're gonna transcend them. |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
Because M$ are greedy wotzitz and want their $$$ for every computer regardless of its use.I think I'm going to give WinToUSB a try. According to the M$ license agreement it's perfectly fine to move the OS to a different Computer, it simply says 'Every time you transfer the software to a new computer, you must remove the software from the prior computer.' By golly, that's exactly what you do when you unplug the cable from one machine and plug it into another. The WinToUSB just allows you to use the feature from the Enterprise edition that allows 'Windows To Go' on a USB drive. I just happen to have two unused Portable USB2 drives laying around, and the Win8.1 Img I used to create a USB Installer. Shouldn't be a problem. I'll bet it even works on my Mac, if for some reason I wanted to run Windows on my Mac... |
Brent Norman Send message Joined: 1 Dec 99 Posts: 2786 Credit: 685,657,289 RAC: 835 |
A bit on an interesting one, same CUDA80 app different result for pulses, Tbar found one more than me. Everything is identical other than that one pulse. 9831637 Pulse: peak=10.53939, time=66.18, period=3.487, d_freq=1418784915.72, score=1.066, chirp=29.566, fft_len=64 Pulse: peak=10.61994, time=66.18, period=3.487, d_freq=1418784902.84, score=1.074, chirp=31.677, fft_len=64 Pulse: peak=4.742024, time=66.18, period=1.395, d_freq=1418784890.02, score=1.07, chirp=33.789, fft_len=64 Pulse: peak=1.68661, time=78.21, period=0.3278, d_freq=1418788250.91, score=1.006, chirp=-33.789, fft_len=16 Pulse: peak=1.712173, time=78.21, period=0.3278, d_freq=1418788200.57, score=1.021, chirp=-42.237, fft_len=16 (extra one at end) Pulse: peak=1.68855, time=78.21, period=0.3278, d_freq=1418788150.22, score=1.007, chirp=-50.685, fft_len=16 My 1070: SETI@home using CUDA accelerated device GeForce GTX 1070 Using pfb = 64 from command line args Unroll autotune 15. Overriding Pulse find periods per launch. Parameter -pfp set to 15 Pulse count: 5 Tbarr's 750Ti: SETI@home using CUDA accelerated device GeForce GTX 750 Ti Unroll autotune 5. Overriding Pulse find periods per launch. Parameter -pfp set to 5 Pulse count: 6 |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
That's the one I mentioned back here, https://setiathome.berkeley.edu/forum_thread.php?id=80636&postid=1874698#1874698 I checked a couple of those tasks and the CPUs said it was your machine. This is the results I got from My CPU on that task; Autocorr: peak=17.92262, time=100.7, delay=0.5504, d_freq=1418791064.98, chirp=19.893, fft_len=128k Autocorr: peak=18.57429, time=100.7, delay=0.5504, d_freq=1418791066.84, chirp=19.911, fft_len=128k Autocorr: peak=18.70623, time=100.7, delay=0.5504, d_freq=1418791068.7, chirp=19.93, fft_len=128k Autocorr: peak=18.3377, time=46.98, delay=5.9057, d_freq=1418787840.88, chirp=-26.005, fft_len=128k Pulse: peak=10.53938, time=66.18, period=3.487, d_freq=1418784915.72, score=1.066, chirp=29.566, fft_len=64 Pulse: peak=10.61993, time=66.18, period=3.487, d_freq=1418784902.84, score=1.074, chirp=31.677, fft_len=64 Pulse: peak=4.742021, time=66.18, period=1.395, d_freq=1418784890.02, score=1.07, chirp=33.789, fft_len=64 Pulse: peak=1.68661, time=78.21, period=0.3278, d_freq=1418788250.91, score=1.006, chirp=-33.789, fft_len=16 Triplet: peak=9.775889, time=72.35, period=1.16, d_freq=1418788341.39, chirp=37.485, fft_len=128 Triplet: peak=9.783492, time=72.35, period=1.16, d_freq=1418788341.47, chirp=38.541, fft_len=128 Triplet: peak=9.772717, time=72.35, period=1.16, d_freq=1418788341.61, chirp=39.597, fft_len=128 Triplet: peak=9.789353, time=72.35, period=1.16, d_freq=1418788341.68, chirp=40.653, fft_len=128 Triplet: peak=9.951018, time=72.35, period=1.16, d_freq=1418788341.82, chirp=41.709, fft_len=128 Pulse: peak=1.712173, time=78.21, period=0.3278, d_freq=1418788200.57, score=1.021, chirp=-42.237, fft_len=16 Triplet: peak=10.09993, time=72.35, period=1.16, d_freq=1418788341.9, chirp=42.765, fft_len=128 Triplet: peak=10.21768, time=72.35, period=1.16, d_freq=1418788342.04, chirp=43.821, fft_len=128 Triplet: peak=10.31642, time=72.35, period=1.16, d_freq=1418788342.11, chirp=44.877, fft_len=128 Triplet: peak=10.38861, time=72.35, period=1.16, d_freq=1418788342.19, chirp=45.932, fft_len=128 Triplet: peak=10.43168, time=72.35, period=1.16, d_freq=1418788342.33, chirp=46.988, fft_len=128 Pulse: peak=1.68855, time=78.21, period=0.3278, d_freq=1418788150.22, score=1.007, chirp=-50.685, fft_len=16 Best spike: peak=23.16136, time=60.4, d_freq=1418788202.54, chirp=2.7358, fft_len=128k Best autocorr: peak=18.70623, time=100.7, delay=0.5504, d_freq=1418791068.7, chirp=19.93, fft_len=128k Best gaussian: peak=3.512909, mean=0.5294697, ChiSq=1.151221, time=84.72, d_freq=1418785084.33, score=-1.700733, null_hyp=2.010488, chirp=-93.857, fft_len=16k Best pulse: peak=4.858808, time=66.18, period=1.395, d_freq=1418784902.84, score=1.096, chirp=31.677, fft_len=64 Best triplet: peak=10.43168, time=72.35, period=1.16, d_freq=1418788342.33, chirp=46.988, fft_len=128 Spike count: 0 Autocorr count: 4 Pulse count: 6 Triplet count: 10 Gaussian count: 0 Time cpu in use since last restart: 8847.0 seconds The only way to tell which machine is correct it to run the task on a CPU, the CPU is usually right, unless it's one of the Usual Suspects. Have you seen any when not using -pfb 64? |
©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.