Posts by jason_gee


log in
1) Message boards : Number crunching : Cuda50 Task Invalid Against Two Other Cuda50s (Message 1764577)
Posted 1 day ago by Profile jason_gee
I've got nothign specifically against best signal printing, though remain to be convinced adding more random numbers would be helpful.

I'm in favour of keeping things simple, but for things like this one (where everything looks the same, but isn't) those details would be useful.


Indeed a tough call. On one hand I would say no reportabla signals means "whocares?". On the Other hand I would say precision matters, and sadly you lost to two bad apple applications this time around.
2) Message boards : Number crunching : Cuda50 Task Invalid Against Two Other Cuda50s (Message 1764575)
Posted 1 day ago by Profile jason_gee
How is it so?


Even with no 'reportable' signals, there is still a best spike, autocorrelation, triplet, pulse, and gaussian in the result file (total 5). The current implementations across the board lose precision the further below reportable threshold you get, and it seems the compiler options used to make the apple stock builds may not have yet factored in the precision enhancements Eric made in v8.


Ah, OK.
I see those in the apple-darwin stderr outputs, but not in the Lunatics stderr output.


If you're referring to the Cuda builds missing these, I'm sad to report our missing brother in arms Joe Segur was largeley responsible for those. I had indicated before his disappearance that I wanted to reduce stderr pollution (by removing stylised ascii art lunatics logo for starters), and he may have taken that to heart. I've got nothign specifically against best signal printing, though remain to be convinced adding more random numbers would be helpful.
3) Message boards : Number crunching : Cuda50 Task Invalid Against Two Other Cuda50s (Message 1764571)
Posted 1 day ago by Profile jason_gee
How is it so?


Even with no 'reportable' signals, there is still a best spike, autocorrelation, triplet, pulse, and gaussian in the result file (total 5). The current implementations across the board lose precision the further below reportable threshold you get, and it seems the compiler options used to make the apple stock builds may not have yet factored in the precision enhancements Eric made in v8.
4) Message boards : Number crunching : Panic Mode On (102) Server Problems? (Message 1764336)
Posted 1 day ago by Profile jason_gee
+1, I find random things sometimes helpful in technical matters.
5) Message boards : Number crunching : Panic Mode On (102) Server Problems? (Message 1764030)
Posted 2 days ago by Profile jason_gee
Sometimes it's easily forgotten that consumer Geforce lines card's don't Have ECC memory. If you can check back on the original host_GPU, chances are it'd come up clear, or otherwise give some clues for further diagnoses, if a repeatable event is found on the same unit. If not repeatable, then could easily be dealing with soft-error (i.e. more to do with ambient, cosmic and semiconductor packaging radiation, While hardware, driver or firmware issues should tend to be more repeatable).
6) Message boards : Number crunching : Update on Linux 64 -Nividia-V8-MB ????? (Message 1763891)
Posted 3 days ago by Profile jason_gee
Yeah, watching the test machine (my Linux Host) for a bit before forwarding at least one Linux x64 build to Eric. Seems OK, with the majority of inconclusives seeming explainable, leaving inconclusive to pending ratio better than the 5% target.

Had been further tied up unexpectedly, which delayed this and the Mac builds being packaged up for Eric with suitable documentation. Since there are still some unknowns for the readmes for both platforms, I'll probably field what I can in the morning via my website downloads. That way I can cheat by letting the community tell me what needs to go in all the question marks in the readmes, lol.
7) Message boards : Number crunching : How Does This Make Sense? (CUDA42 vs CUDA50 on Similar Machines) (Message 1763714)
Posted 4 days ago by Profile jason_gee
If really 'stuck' Workarounds are to either force the issue on the one stuck on the 'wrong' app (I think a project reset *might* reset those numbers, though haven't checked), or by running anonymous platform.


Verrryyyyy Interesting!

OK, I give up, I will go with Lunatics in a week or two; again, can I force the issue there or not? Given the APR numbers quoted above, plus the fact that I have done a few thousand v8 on each machine already, would it be wise to force i7-4820K to CUDA42 (if doable)?


I would try both with using Lunatics, selecting Cuda50, then See what the APRs 'stabilise' at. Then for the sakes of comparison Cuda4.2. If you were able to compare actual runtime medians and variance, you'd probably see some overlap in real performance depending on work mix, making the server's confusion partly understandable. Though as humans we can look at the runtimes and say one or another is clearly better, the server's view has some fairly bad cataracts.
8) Message boards : Number crunching : How Does This Make Sense? (CUDA42 vs CUDA50 on Similar Machines) (Message 1763712)
Posted 4 days ago by Profile jason_gee
Application details for i7-4790K
cuda42: 100.99 GFLOPS
cuda50: 127.37 GFLOPS

Application details for i7-4820K
cuda42: 113.13 GFLOPS
cuda50: 100.49 GFLOPS


The numbers make it simpler to picture. 'real' vales are probably around 106 GFlops for the Cuda42 app, and 115 GFlops for the Cuda50 one, So way closer to one another than the server's foggy glasses, so the apps look the same.
9) Message boards : Number crunching : How Does This Make Sense? (CUDA42 vs CUDA50 on Similar Machines) (Message 1763701)
Posted 4 days ago by Profile jason_gee
Distilling down things as best I can,
- what app is 'chosen' is based on what the server estimates your tasks will take.
- those estimates are connected to CreditNew, which we know is unstable
- there isn't a 'huge' performance difference between Cuda4.2 and Cuda5, at least compared to how much those estimates are unstable

Let's say there were 20%, +/- 10%, difference between the applications on the same host (which there wouldn't be that much, but it'll illustrate. Next, let's say the machines are truly identical in usage/loading, temps, clocks, and the Angle ranges of work they receive.

Even under that impossible scenario, for one app server estimates will have ~+/- 37% variation, and logic says 50 % will be on the high side (estimate 37% too long), and 50% on the low side (estimating 37% short). So the choice is swamped by noise.

Performing some statistical voodoo, then under these ideal circumstances you get some probabilities.
Chance of given host receiving the 'correct' application (A), given B is the wrong application:
P(A|Bwrong)= ( PAestimateLow x PBestimateHigh) / PBestimateHigh
--> ( 50% x 50% ) / 50% = 50%

so 50-50 chance for each of your machines to get the right app or the wrong app, and you had two coin tosses.

For the formula to change to be less of a useless coin toss, the estimates would need to be closer than the 'actual' difference on the same host. So the +/37% server estimate slop would need to be improved to better than the +/- 10% difference between apps, i.e. by a factor of four or more times closer to actual. Quite doable in engineering terms, though I don't think the Will is there at the moment.

If really 'stuck' Workarounds are to either force the issue on the one stuck on the 'wrong' app (I think a project reset *might* reset those numbers, though haven't checked), or by running anonymous platform.
10) Message boards : Number crunching : OpenCL NV MultiBeam v8 SoG edition for Windows (Message 1763301)
Posted 6 days ago by Profile jason_gee
Assuming these are offline runs, can you run with the -poll command line option on the Cuda50 application ? effect should be full CPU core use (similar to the OpenCL mid-Low ARs) and faster elapsed. if noticably faster, and if CPU core Use is acceptable for some people/situations (though probably case specific), I would consider enabling the feature as an option through mbcuda.cfg

Another thing to consider with the Cuda50 build on Later GPUss, specifically with the VLAR timing, is default pulsefind settings are set fairly gentle. Could be worth upping pfblockspersm to 15, and pfperiodsperlaunch to 200, which would be more aggressive settings than suitable as defaults. [ process priority may need similar attention]
11) Message boards : Number crunching : OpenCL NV MultiBeam v8 SoG edition for Windows (Message 1763127)
Posted 6 days ago by Profile jason_gee
Cuda app certainly won't be 'fighting' for anything unless you tell it to :)
12) Message boards : Number crunching : Cuda50 Task Invalid Against Two Other Cuda50s (Message 1763073)
Posted 7 days ago by Profile jason_gee
Always hard to gauge on few results and little time, but basically good results can 'conspire', and your app could be good or bad. So many new apps, too early to point bones.
13) Message boards : Number crunching : Panic Mode On (102) Server Problems? (Message 1763041)
Posted 7 days ago by Profile jason_gee
EDIT- just noticed some longer running than estimated WUs on my i7 as well.


Yeah, In the scheme of things, that's the estimation component in the scheduler of CreditNew, so expect it to be no more stable/accurate than credits. (moral being never send scientists to do an Engineer's Job)

For the settings, yeah process priority impact *might* be significantly swamping other settings. It's all very system dependant though, so if your Core2Duo happens to have as much trouble feeding 2 750's as mine does 1 980, then I wouldn't expect 'max performance'. The ability to respond to software interrupts as fast isn't there, motherboard chipset plays a role there, irrespective of actual CPU utilisation.
14) Message boards : Number crunching : I'm Trying to Build an OSX CUDA App... (Message 1763038)
Posted 7 days ago by Profile jason_gee
*probably* will eventually, though in the world of science funding making presumptions can be dangerous, unless you live in Germany where the Chancellor is a Physics doctorate so knows the deal. At the very least I'd be throwing similar precision refinements as I make Cuda support for AP, which can trickle back to other applications. I suppose a lot will depend on the nature of the se other telescope searches though, of which I have no knowledge other than 'bigger data'
15) Message boards : Number crunching : I'm Trying to Build an OSX CUDA App... (Message 1763029)
Posted 7 days ago by Profile jason_gee
Yeah, I get as excited as the next person to see the 980 tear it up, lol, and there's a lot more to come that's been tried, and some not even tried yet. I know Petri knows what's going on and continues working on things too :D.

I think the next big cheek clench will be as GBT/Breakthrough data starts to flow, then we get to find out if the v8 apps even hold up (let alone the servers)
16) Message boards : Number crunching : Panic Mode On (102) Server Problems? (Message 1763026)
Posted 7 days ago by Profile jason_gee
On those 750's with the zi app, I notice that you're running at default (generic) settings:
pulsefind: blocks per SM 4 (Fermi or newer default)
pulsefind: periods per launch 100 (default)
Priority of process set to BELOW_NORMAL (default) successfully


The Lower the angle range, the more impact those settings have. I would suggest for display devices and a device like that, pfblockspersm of 15, and pfperiodsperlaunch of 200, might inmprove things. On non-display GPUs (or even the display one if you don;t experience notable slowdown) you could up the process priority. All those settings are in the xxx_mbcuda.cfg file, using the sample provided as a guide. If display lag doesn't get too bad, then such settings should reduce CPU feeding requirement somewhat.

For Core2Duo, yes MBv8's increased precision makes feeding harder (I use Core2Duo to feed a 980 on the main development rig, so know the pain). Not something that can be necessarily optimised out (at least short term), because the [CPU-Side] precision increase is there for reasons.
17) Message boards : Number crunching : I'm Trying to Build an OSX CUDA App... (Message 1763017)
Posted 7 days ago by Profile jason_gee
Patience :) Xbranch worked out by playing the 'long game' (spacesuited tortoise on my website graphic isn't there by accident, lol).

[Straight Build has a lot of Caveats/issues to iron out for a widescale release]

Integration of that, and unspecified other stuff, is there as a proving ground for some new technologies & techniques. v8 transition dust settles (without servers blowing up every week), then you'll get a roadmap :)
18) Message boards : Number crunching : Windows 10 - Yea or Nay? (Message 1762384)
Posted 9 days ago by Profile jason_gee
...but please, please, please keep this thread alive and up to date, cause it is momentary really the best resource for keeping your (my) flawless running 7s and 8.1s systems relatively clean - very much appreciated - thanks a lot! :)

It's just a shame that it now has no relation to what the thread was originally started for.


Pretty similar sentiments I've heard from Ex M$ employees over the last year.
19) Message boards : Number crunching : Why won't V44 Recognize my Video card (Message 1762382)
Posted 9 days ago by Profile jason_gee
I *think* Jim might be expecting the installer to detect the kinds of GPU installed. Please correct me if I am wrong Jim. No you have to select the applications to install with the radio buttons. It is of course entirely technically possible to detect such hardware and automate selections, though it seems shortcut malware/antivirus heuristics don't like such approaches, so please know what hardware you have (and select accordingly) :) Most likely future detection automation will rely on post installation features.
20) Message boards : Number crunching : Notes on Nvidia driver bloat [361.60 Hotfix] (Message 1762167)
Posted 9 days ago by Profile jason_gee
OK, Thanks. Will add the missing ones later, and see if I have any of them. Worth me crossreferencing once I can spend more time at the machine. Some I had came from a subsequent updated post, and indeed are expected on 8/8.1 or 10 itself only. Probably will attempt to 'unbreak' Win update for the exercise at some point, though for now happy with 361.75 clean install, broken Windows update, and the handy combined lists of updates to watch/flag. Some part, or the combined total seems to have made a big difference on what is a quite dated core2Duo.


Next 20

Copyright © 2016 University of California