Posts by jason_gee


log in
1) Message boards : Number crunching : Q: How much time for a 0.42 MB task? (Message 1714412)
Posted 13 days ago by Profile jason_gee
on ye old stock x41zc, Running 2 up, looks like about 10 minutes elapsed (so 5 minutes effective) on reference cooling 980SC with no particular tweaks. newer code by yourself and yours truly looks like achieving maybe 30% improvement (give or take) sp fingers crossed should end up a fraction better than the 4m11s by the time we're both done, but definitely in the ballpark ( either 1 up, or 2 up average if we still need a bit of latency hiding).

And the release date approx,to us who are not coders, will be when?


We have rough plans to have closed alpha testing underway in about 3 weeks, and a probably 4 week sprint for release polishing x41 finalisation following that.. so roughly 7 weeks all up.

i.e. we're about 3 weeks into a 6 week alpha sprint.

Importantly the team is new to Agile development processes, so we are taking very long sprints ( normally 2 weeks, 6 and 4 for us) so we learn about the process. We're also keeping things pretty laid back, with our conflicting schedules, but initial indications seem to be pointing to that this Agile stuff works and is no joke (even if you do it very slowly ;) )

What we hope to acheive long term, is less dependance on me personally, more automation, and higher quality. Petri's performance work (and a few bits of mine) is an extra bonus, before we get to strip it all back to the bones and reengineer it better for x42.
2) Message boards : Number crunching : Q: How much time for a 0.42 MB task? (Message 1714404)
Posted 13 days ago by Profile jason_gee
on ye old stock x41zc, Running 2 up, looks like about 10 minutes elapsed (so 5 minutes effective) on reference cooling 980SC with no particular tweaks. newer code by yourself and yours truly looks like achieving maybe 30% improvement (give or take) sp fingers crossed should end up a fraction better than the 4m11s by the time we're both done, but definitely in the ballpark ( either 1 up, or 2 up average if we still need a bit of latency hiding).
3) Message boards : News : Dan Werthimer radio interview (Message 1713419)
Posted 15 days ago by Profile jason_gee
Dan has a radio voice :-D
4) Message boards : Number crunching : Why this CPU task invalid so soon? (Message 1709939)
Posted 22 days ago by Profile jason_gee
Thanks! Some (evolving) back story is that while we largely expect code that's been working for a long time on a lot of systems is 'correct', that it doesn't necessarily cope with all possible inputs gracefully opens up some possibilities to make it better. No-one really expects us to do that for cases where there is some obvious instability, but we do have some luxuries in our seti@home cases of a lot of people looking at numbers for a long time.

In this particular case, it probably shows handling of denormal numbers and other ( I usually use flush to zero modes, not sure what floating point modes this particular codebase has enabled) could be used to at least shield postprocessing code from rubbish, and perhaps flag some warning to the user, reprocessing/retries, or something of that sort.

We reasonably rely a fair bit on Boinc's redundancy mechanism to weed out a lot of result data corruption of this sort, but it doesn't really handle some cases, like where the source of difficulty might be in the source data distributed to both hosts, or completely different hosts generate a NaN in the same number (e.g. a sum used for normalisation) via completely different avenues/circuits.

It'll be something to think about, given that we have a reasonable idea of what numbers look sane at different points, and expecting some number of machines fall off the rails over time is pretty reasonable.
5) Message boards : Number crunching : Why this CPU task invalid so soon? (Message 1709801)
Posted 23 days ago by Profile jason_gee
@Keith: Can you possibly put in some tests of the CPU with prime95 ? (before and after the reseating/resinking/TIM-replacement etc).

The NaNs, once in there become problematic for the comparisons to select best.

[Edit:] from scattered sources, I'm getting the general impression the FX-8350 can sometimes need a small voltage bump for Prime95 stability, over motherboard 'auto' settings. Some mentions of Bios updates. If something like that is going on here, then it would probably be something that would slowly worsen with time. My Intel Core2Duos have drooped a little oversome years use, so wouldn't be a strange or unusual thing necessarily.
6) Message boards : Number crunching : Windows 10 - Yea or Nay? (Message 1709244)
Posted 24 days ago by Profile jason_gee
Would be good to devise some table of unwanted services


http://www.blackviper.com/service-configurations/black-vipers-windows-10-service-configurations/
7) Message boards : Number crunching : Why this CPU task invalid so soon? (Message 1709111)
Posted 24 days ago by Profile jason_gee
It looks almost as though that best autocorrelation wasn't saved across that restart. That host could be a suitable victim for multiple restart experiments, since all the weaknesses found in the stderr truncation debate, may well apply [but to state or result files]. That'd be important to know about for application developers IMO, even if Boinc developers prefer to keep the onus for intact files on the client. [afaik the client has no interaction with the state/result files until completion, so any damage there lies solely within the domain of the app and boincapi linked into it]

[Edit:] It ^could^ well mean (hypothesising), anything less than full commit AND a patient client, is just a workaround (which we knew), but problems manifesting in a repeatable way would be surprising, interesting and a smoking gun at the same time.
8) Message boards : Number crunching : Windows 10 - Yea or Nay? (Message 1708813)
Posted 25 days ago by Profile jason_gee
http://www.blackviper.com , and a good NAT router with no NSA backdoors :D. Can be trickier if you have a static IP

[For the ads, adblockplus on Google Chrome, with the youtube ads disabled via chrome debug prompt]
9) Message boards : Number crunching : 560Ti offer (Message 1708261)
Posted 26 days ago by Profile jason_gee
...Don't know much about the 5850, but for MB Cuda the 5870 in my Mac seems to be roughly half the performance of the 780...

I think you're being a generous from what I've seen. The question is what would work best on a dual core P4 running XP. A 5850 running the old Catalyst 12.1 or a 560Ti running the latest NV driver. Yes, you can run the newer NV drivers in XP, you can't run anything newer than 12.1 with ATI.
Compare the times from here, http://setiathome.berkeley.edu/results.php?hostid=7423385&offset=60


Mmmm, yeah well I try to keep an open mind, given I pushed neither GPU (yet).
You're saying 5850 XP drivers is problematic ? I can see that, given Ati.amd would have faced the same OS tech changes that nv did in a different way.

Dual core p4, I think then maybe the 560ti Cuda only would completely own just about anything else the machine would take ( except maybe a 680). Could be tricky to feed the 560ti enough, very tricky.
10) Message boards : Number crunching : 560Ti offer (Message 1708251)
Posted 26 days ago by Profile jason_gee
ooh, a good ol' fashioned GPU battle, lol :D

For my part, I can say the 560ti can either be pretty beasty for what it is, and punch way above its weight (power cost and performance are relatively good), but can also be challenging to drive, because some manufacturers set the core voltage too low for glitch free operation at clocks greater than reference.

Don't know much about the 5850, but for MB Cuda the 5870 in my Mac seems to be roughly half the performance of the 780. That should place the 560ti closer to a 5870 than a 5850 I would have thought.... at least for MB

A wildcard here is the 560ti was the first nv GPU to get the compute/VRAM balance right, extending on the successful bits of the 460. Those refinements should make it best with more than one task at a time (with all the voltage, clocks, and power caveats implied above)
11) Message boards : Number crunching : Windows 10 - Yea or Nay? (Message 1707992)
Posted 27 days ago by Profile jason_gee
good thing is, looks like blackviper already updated his site with win10 info, which would make all the MS data mining and other OS pretend niceness pretty much obsolete, if you just disable the unwanted stuff.
12) Message boards : Number crunching : SETI/BOINC Milestones [ v2.0 ] - XXIX (Message 1707255)
Posted 29 days ago by Profile jason_gee
well that explains why one WU does not use up the whole card
thnx for the explanation


FWIW there are other more sophisticated methods that can automate this process, but by nature come in future applications and supporting tools. The general gist is if you can balance these things by hand, using some sort of logic and tools, then there is a way to replicate that. How far the Boinc infrastructure can be stretched is another matter.
13) Message boards : Number crunching : SETI/BOINC Milestones [ v2.0 ] - XXIX (Message 1707233)
Posted 29 days ago by Profile jason_gee
@jason_gee

how many wu's can the 900 series cards do at the same time without overloading the card and slowing everything down ?



Depends on the system feeding them, and the app+settings. Best efficiency would be in the region of 2 to 4 MBs, which will change. At least two are needed to hide serial CPU portions of processing, and driver latencies (waits)

The trick is to add enough tasks at once to just peak power demand, with the tradeoff of extra tasks creating system contention. Compare to the approach that doesn't work well, of attempting to use up VRAM.

As Apps further mature, the optimal # of tasks will likely reduce toward a probably ideal of 2, and optionally use more resources for those 2 instances.

More than 2 instances at the moment, is solely due to the app needing to catch up with the hardware.
14) Message boards : Number crunching : SETI/BOINC Milestones [ v2.0 ] - XXIX (Message 1707199)
Posted 29 days ago by Profile jason_gee

Jason would be more of one to ask about why nvidia does so well here, but these days he sometimes gets to speaking in such technaleze that even I cannot understand it.

No offense, dear Jason.


LoL, none taken :D Yeah I have become a bit lazier with decoding the technobabble over the years, just because there're plenty of people to chip in, in some cases, and sometimes I found no simple way to present the information ( or in some cases organise it into proper words, lol)

Could be part ageing, part mad scientist kicking in I suppose.

In the case of nVidia GPUs doing better here relative to on other projects, for multi beam part is the optimisation of the Cuda libraries, another that we don't really use much double precision at all (which is knobbled except on the highest end dedicated compute models), and of course the odd couple of my own techniques, grown from research, thrown in (some of which raistmer adapted for OpenCL use too).

I've looked at some limited portions of other project Cuda sources, and they tend to be mostly devoid of some key optimisations, and/or depend on double precision a lot more. Recent gradual moves, having had some help from Petri33 with some streaming, are pointing towards ways to capture and automate these techniques more, so that they'll scale to next generation hardware better. That would be probably too costly or disruptive for some projects, since often it means throwing out large portions of code that works and starting again.
15) Message boards : Number crunching : Panic Mode On (99) Server Problems? (Message 1706917)
Posted 31 Jul 2015 by Profile jason_gee
Win7 dev machine cleaned up about 63 ghosts while I was sleeping... so resend_trigger still works. Can turn that back off now. Not sure when the ghosts accumulated, as been a pretty busy time for me, but all tidy again.
16) Message boards : Number crunching : Panic Mode On (99) Server Problems? (Message 1706826)
Posted 30 Jul 2015 by Profile jason_gee
will be good to check out. Just feeling my way for now. No huge surprises.

Now I would call the temps off the back a little warmer than lukewarm. Yeah the CPU app definitely could do with an update. Still deciding what Cuda GPU i'd like to whack in there. Probably will multithread all 16 threads and combine the radeon and the 680 results just to confuse credit new. Could be a lark
17) Message boards : Number crunching : Panic Mode On (99) Server Problems? (Message 1706817)
Posted 30 Jul 2015 by Profile jason_gee
This case seems to have convenient places to warm hamburgers, though not warm enough yet.
18) Message boards : Number crunching : Panic Mode On (99) Server Problems? (Message 1706814)
Posted 30 Jul 2015 by Profile jason_gee
lol, one step at a time. The magic mouse is growing on me, but this keyboard must be made for midgets or something.

So far my win7 dev machine with a 980 is singing it's soothing song. No such comforting sound out of the mac yet, but am sure if I push it it'll have limits. Not all that worried about where they are, just that I can spread Xbranch into more areas, since I believe I have a better grasp of the synthetic-aesthetic, enough to push things a little further anyway.
19) Message boards : Number crunching : Panic Mode On (99) Server Problems? (Message 1706808)
Posted 30 Jul 2015 by Profile jason_gee
interestingly I have a smattering of VLARs now on the Mac Pro {allocated to CPUs}. I'm babying it, but Kudos to Urs/Raistmer for the radeon munching away, and the stock CPUs doing their thing. Have enabled 4 of the sixteen threads for now, just to see how hot it gets in here. So Far looks like very ugly computers with a half sucked lozenge look, do pretty good :D
20) Message boards : Number crunching : CPU warnings (Message 1706799)
Posted 30 Jul 2015 by Profile jason_gee
The last time I saw crazy errors like that, was when a motherboard was failing, and coincidentally the (i5) CPU fan had seized. I would eliminate the easy stuff, like CPU fan is working (check) thermal paste isn;t dry, CPU reseated carefully.... Then it could still be the CPU itself, but seems on the low probability side compared to possible motherboard /power issues.


Next 20

Copyright © 2015 University of California