Posts by Stephen Uitti
log in
1) Message boards : Number crunching : Errors on CUDA workunit (Message 20821)
Posted 920 days ago by Profile Stephen Uitti
My 64 bit AMD Linux machine running a GTX 650 ti turned in a unit that took 50 hours instead of the usual 36 hours on July 9th, with a validation error. During the run, BOINC was reporting huge times (many hundreds of hours) to completion. Task 19120250.

The current running task, 19405307 has BOINC claiming 800 hours to go after 15 hours running. As it's at 30%, that works out to about 51 hours total, maybe 36 hours remaining. I'm guessing it will also be invalid.

The units claim to be Large Collatz Conjecture v6.06 (cuda40). The GTX 650 ti is, as near as i can tell, a CUDA 3.0 board. It has always worked before.

My 64 bit AMD Linux machine running a GTS 250 is now sporting 162 tasks in status Error while computing. All since July 9th. Not good. I've "Remove"'d the project, and added it back with "Add Project"... It says it won't get another task until tomorrow. Typical task: 19421338 which is a Large Collatz Conjecture v6.07 (cuda55). This card is so old (released in 2009), it's hard to even figure out what CUDA level it is. Maybe 1. Certainly not 5. The machine is configured only for "UDA for nVidia GPUs (cuda40/cuda55)" plan tasks.

My backup GPU project for this card is Moo!, which is currently out of units. But this card crunched through a whole bunch of those for a couple days. I don't think it's the card...
2) Message boards : Number crunching : Mixed nvidia and AMD drivers (Message 20336)
Posted 1043 days ago by Profile Stephen Uitti
I had to reload the OS (Mint 13 - an Ubuntu derivative) on the AMD A8-6600K. I disabled the Nvidia card, and did not install the Nvidia driver. I installed AMD's APU (GPU on the chip) driver. It does great video. On BOINC startup, i get:

28-Feb-2015 10:30:49 [---] cc_config.xml not found - using defaults
28-Feb-2015 10:30:49 [---] Starting BOINC client version 7.2.42 for x86_64-pc-linux-gnu
28-Feb-2015 10:30:49 [---] log flags: file_xfer, sched_ops, task
28-Feb-2015 10:30:49 [---] Libraries: libcurl/7.22.0 OpenSSL/1.0.1 zlib/1.2.3.4 libidn/1.23 librtmp/2.3
28-Feb-2015 10:30:49 [---] Data directory: /u/suitti/BOINC
28-Feb-2015 10:30:49 [---] CAL: ATI GPU 0: AMD Radeon HD 7500/7600/8500/8600 series (Devastator) (CAL version 1.4.1848, 768MB, 692MB available, 1080 GFLOPS peak)
28-Feb-2015 10:30:49 [---] Creating new client state file
28-Feb-2015 10:30:49 [---] Host name: pensar
28-Feb-2015 10:30:49 [---] Processor: 4 AuthenticAMD AMD A8-6600K APU with Radeon(tm) HD Graphics [Family 21 Model 19 Stepping 1]
28-Feb-2015 10:30:49 [---] Processor features: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm constant_tsc rep_good nopl nonstop_tsc extd_apicid aperfmperf pni pclmulqdq monitor ssse3 fma cx16 sse4_1 sse4_2 popcnt aes xsave avx f16c lahf_lm cmp_legacy svm extapic cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw ibs xop skinit wdt lwp fma4 tce tbm topoext perfctr_core arat cpb npt lbrv svm_lock nrip_save tsc_scale vmcb_clean flushbyasid decodeassists pausefilter pfthreshold bmi1
28-Feb-2015 10:30:49 [---] OS: Linux: 3.2.0-23-generic
28-Feb-2015 10:30:49 [---] Memory: 7.00 GB physical, 3.91 GB virtual
28-Feb-2015 10:30:49 [---] Disk: 435.38 GB total, 233.71 GB free

...which shows that it sees the APU. But it doesn't seem to want to crunch units on it (Collatz or otherwise). Has anyone gotten the AMD APU running at all under Linux? If so, is there some sort of hint?

My son has an AMD A10 6800K under Windows 10 (his choice), and it gets 40k to 60k units a day with Collatz. I'm expecting less from my A8, should i get it working. If i have to pick just one, i'll go with the 80k/day Nvidia. Even if i can't get the APU to work with the Nvidia GPU under Linux, i could possibly run the Nvidia card on another box, if i could get the APU crunching under Linux.
3) Message boards : Number crunching : nvidia gts 250 (Message 20133)
Posted 1131 days ago by Profile Stephen Uitti
Host 142392 really has an NVIDIA GTS 250. It finished a large this afternoon (EST, i think).

15-Dec-2014 13:25:47 [Collatz Conjecture] Computation for task large_collatz_2391964306953104794739_26388279066624_0 finished

Then, the logs show that it can't seem to find any work it can do. That's because it no longer thinks it has a GTS 250. I restarted BOINC, and it said "No usable GPUS found". The machine had been running for 12 days. I restarted Linux, and BOINC, and it's still on "No usable GPUS found".

And yet, the screens work. The NVIDIA X-Server tool seems happy with it. Sigh. If i recall, this is what i had to see if BOINC was going to see the GTS.

If i recall, it was a PITA to get the driver installed. Actually, since that was a year ago, i don't recall it very clearly. And, it's been a very, very stable year for the card and the driver. There isn't any chance i have to upgrade the driver, is there?

Mint 13 xfce.

Stephen.
4) Message boards : Number crunching : Mixed nvidia and AMD drivers (Message 19754)
Posted 1266 days ago by Profile Stephen Uitti
I've just discovered that my AMD A8-6600K APU does OpenCL. That is, the video chip on the 4 core CPU should be able to crunch Collatz. This system has a 2011 vintage Nvidia GTS 250 crunching, and doing a good job. But the video on the CPU chip could be similar performance, from what little i was able to find. The AMD driver suggests that i should uninstall my Nvidia driver first. Of course, i don't want to do that. What i want to do is run both. For various reasons, i don't want to goof with it right away, maybe in a week or two. I'm curious if anyone has any experience with this mix of drivers. Oh, it's an Ubuntu Linux system (Mint, really, but Mint is derived from Ubuntu). One crazy idea is to run one of the cards in VirtualBox. So one copy of Linux would have the Nvidia driver, and the other would have the AMD driver. There's plenty of RAM and cores. But i think i'd rather have it all in the host system.

At the moment, my guess is 80k credit a day. Though anyone who is crunching with an AMD A8-6600K APU might be able to tell me better.
5) Message boards : News : Server Back Online (Message 19557)
Posted 1334 days ago by Profile Stephen Uitti
Opps. My bad. My SETI@Home computing preferences for disk were very strange, and BOINC thought i didn't have a disk space for BOINC, despite 11 GB free. Fixed.

And why only 11 GB free? I've no idea what i had in mind for that unused 438 GB partition that is formatted, empty, but unmounted... sigh.

I've been totally loving the 4-5 day large collatz units.
6) Message boards : News : Server Back Online (Message 19556)
Posted 1334 days ago by Profile Stephen Uitti
I've got a machine, ID 142392, with a GTS 250 that has just turned in a large and several solo units. The machine was set to not accept CPU units, just GPU units. It can't seem to get new units assigned to it. I even opened it up to CPU units, but nothing. There's nothing running on the GPU. I even restarted the machine. The logs say:

26-May-2014 20:11:42 [Collatz Conjecture] Sending scheduler request: Requested by user.
26-May-2014 20:11:42 [Collatz Conjecture] Requesting new tasks for NVIDIA
26-May-2014 20:11:43 [Collatz Conjecture] Scheduler request completed: got 0 new tasks

These would be the first units after a flood of units now completed after the big change. It is running an outdated version of BOINC, but i couldn't get the current version to run, as i didn't have the shared libraries installed. I also have another machine with Linux Mint 13, which is running the latest boinc, so the plan is to grab the shared libraries from there... And though i'd really like to upgrade (there are several reasons), my guess is that this isn't the issue. It gets seti units, for example.
7) Message boards : News : Missing Apps (Message 19263)
Posted 1362 days ago by Profile Stephen Uitti
The better configuration should be:

Micro project: for ARMs only.
Mini Project: for CPU only
Solo & Large Project: for GPU only.


My slowest GPU, an ATI on an older laptop, takes about 12 hours on a Mini 6.04 ATI4. My faster GTS 250 should take about 14 minutes for a Mini, as it's 50x faster. And, i have a GTX 650 ti, which should be 5x faster than that, or 750x faster than the ATI. The 650 ti isn't the fastest GPU out there. So, i don't think you can simply say Minis are for CPUs only. My laptop ATI GPU is similar in speed to the laptop's CPUs - perhaps 1x to 2x faster. I'd really like control for each host. An automatic assignment should work, based on a benchmark. But the benchmark would have to be the time taken for previous Collatz work units, not the BOINC benchmarks. For example, my GTS 250 got a 78 hour large work unit, but the estimate was easily ten times that.

The way i'd do it is, for a new machine, issue a micro work unit. If it takes less than ten minutes, issue a mini work unit. If that takes less than an hour then issue a solo unit, and if that takes less than an hour, go for large units - unless the user has said they don't want them. I could see someone having a fast laptop that will be on the net, or even powered on, only sometimes, and long units aren't a good match, or some other reason. I generally like long units on my desktops.

Let's say i have a unit in progress. It's showing 3:04:00 elapsed at 45%. Converting the time to decimal time, 4 (minutes) / 60 = .066. Add the 3 hours, and you get 3.066 hours. Divide this by the percent done: 3.066 / 0.45 = 6.81 hours. That's the total time for the unit. Subtract the 3.066 it has already done, 6.81 - 3.066 = 3.75 hours remaining. These calculations use the job at hand with the CPU/GPU at hand. They've been very accurate for me. But when i compare my result with the Remaining (estimated), it might say anything from 2:30 to 7:20 instead of 3:45. I've gotten reasonable accuracy even when less than 1% of a work unit has been completed, though the more, the better.
8) Message boards : News : Missing Apps (Message 19259)
Posted 1363 days ago by Profile Stephen Uitti
Machine ID 141349 is a laptop with an ATI video card. A large solo task has started, and should finish in 147 days. Machine ID 142392 has an NVIDIA GTS 250, and a large solo task takes about 3 days, and should finish tomorrow night. I see in Collatz Conjecture Preferences that i can turn off large solo tasks, but 3 days seems OK. Can this be done on a machine by machine basis?




Main page · Your account · Message boards


Copyright © 2018 Jon Sonntag; All rights reserved.