Message boards :
Number crunching :
Random Musings About the Value of CPUs vs CUDA
Message board moderation
Previous · 1 . . . 7 · 8 · 9 · 10 · 11 · Next
Author | Message |
---|---|
ML1 Send message Joined: 25 Nov 01 Posts: 21241 Credit: 7,508,002 RAC: 20 |
... Vanitas vanitatum is a (bad) Latin translation of the Hebrew words Hebel habalim. whose translation might be Vaporware. There is also: Hevel Havalim All very subtle, and very apt in the various flavours of meaning of transience and meaninglessness... Good interesting quote Tulio. Thanks also to Who? ;-P Also a question of context? This is a good one to add into wikipedia. Meanwhile, hopefully the debugging and optimisations race on apace! Happy fast crunchin', Martin See new freedom: Mageia Linux Take a look for yourself: Linux Format The Future is what We all make IT (GPLv3) |
Blurf Send message Joined: 2 Sep 06 Posts: 8964 Credit: 12,678,685 RAC: 0 |
Just a friendly request to stop the sniping back and forth. If you don't agree with someone's post, don't respond. If you take personal issue with a post, either deal with the person via Private PM or Red-X it and the moderators will look at it. If the sniping continues, this thread will be locked. Any concerns over this post should go to the mod list directly via red-x or an email to seti_moderators@ssl.berkeley.edu Thanks. |
eaglescouter Send message Joined: 28 Dec 02 Posts: 162 Credit: 42,012,553 RAC: 0 |
All pointless if the system cannot work properly: http://setiweb.ssl.berkeley.edu/forum_thread.php?id=51240 It's not too many computers, it's a lack of circuit breakers for this room. But we can fix it :) |
kelhorn Send message Joined: 6 Jan 09 Posts: 1 Credit: 1,799,952 RAC: 0 |
Correct me if I'm wrong, but I think some of you just lost the meaning of s@h. It isn't about getting a massive amount of credits, being 1st in tables or any such thing. IMO it's just a pure joy of taking part in worldwide project, that might lead to a biggest discovery ever made. I don't care, if the CUDA works perfectly or not, or how many points it'll make per day, I guess I'll stick to my 1,6 Celeron M. If some of you think, that the whole BOINC is just about getting more&more points, then why don't you upgrade your gear constantly and focus on 3dMark, seeing all the digits getting there higher every time should give just as much satifaction as s@h. It doesn't matter what gear do you have, does it take you 3k sec for one chunk or 300k sec, what matters now is the contribution and faith in receiving phonecall from c3p0;] |
OzzFan Send message Joined: 9 Apr 02 Posts: 15691 Credit: 84,761,841 RAC: 28 |
Correct me if I'm wrong, but I think some of you just lost the meaning of s@h. It isn't about getting a massive amount of credits, being 1st in tables or any such thing. IMO it's just a pure joy of taking part in worldwide project, that might lead to a biggest discovery ever made. I don't care, if the CUDA works perfectly or not, or how many points it'll make per day, I guess I'll stick to my 1,6 Celeron M. If some of you think, that the whole BOINC is just about getting more&more points, then why don't you upgrade your gear constantly and focus on 3dMark, seeing all the digits getting there higher every time should give just as much satifaction as s@h. It doesn't matter what gear do you have, does it take you 3k sec for one chunk or 300k sec, what matters now is the contribution and faith in receiving phonecall from c3p0;] Welcome to the boards! |
-= Vyper =- Send message Joined: 5 Sep 99 Posts: 1652 Credit: 1,065,191,981 RAC: 2,537 |
Correct me if I'm wrong, but I think some of you just lost the meaning of s@h. It isn't about getting a massive amount of credits, being 1st in tables or any such thing. IMO it's just a pure joy of taking part in worldwide project, that might lead to a biggest discovery ever made. I don't care, if the CUDA works perfectly or not, or how many points it'll make per day, I guess I'll stick to my 1,6 Celeron M. If some of you think, that the whole BOINC is just about getting more&more points, then why don't you upgrade your gear constantly and focus on 3dMark, seeing all the digits getting there higher every time should give just as much satifaction as s@h. It doesn't matter what gear do you have, does it take you 3k sec for one chunk or 300k sec, what matters now is the contribution and faith in receiving phonecall from c3p0;] Well said and welcome.. When i first began with seti that was my main goal also but after the years it has become merly an obsession instead of getting the highest rank possible but maintaining 100% accurate results.. I for my self has been overclocking my computers since i got Commodore Amigas and that is now 18 years ago i started doing this and has been doing it since then.. So all in all the goal is now for me to have a decent powerusage with my main machine home and to get it done with aircooling alone maintaining its highest peak and then i back it up a bit just to be safe and sound and my electricity bill wont skyrocket too much.. It's pointing upwards more and more anyhow :) Well ... Many of us nowadays use s@h as both an extreme hobby to get the most out of it (RAC) and maintaining validness of the work beeing done. Cuda at present time is in my oppinion considered "dodgy" but that will also be corrected later on. The good thing about it that the code if it calculates through an entire WU produces accurate results and if not the code fails in the beginning and eventually knocks out the gfx driver so all in all that is good but the validators at berkeley need to know somehow becuase if two graphiccards calculate the WU and fails and then a Cpu person comes along and chews the wu in 10000 seconds that person won't be getting the credits what he deserves for doing that and that is truly an annoying fact.. But in the end when the staff which by all means are doing a hell of a job maintaining this huuuge project corrects this they tend to even this out with scripts which corrects things that have occured and so on so i have total faith and respect for all crew members at s@h berkeley.. Hope you get an insight of why some of us tend to bark and getting annoyed on this forums.. The solution in nearly all cases in the years has been around the corner from my point of view.. Kind regards Vyper _________________________________________________________________________ Addicted to SETI crunching! Founder of GPU Users Group |
Iona Send message Joined: 12 Jul 07 Posts: 790 Credit: 22,438,118 RAC: 0 |
Correct me if I'm wrong, but I think some of you just lost the meaning of s@h. It isn't about getting a massive amount of credits, being 1st in tables or any such thing. IMO it's just a pure joy of taking part in worldwide project, that might lead to a biggest discovery ever made. I don't care, if the CUDA works perfectly or not, or how many points it'll make per day, I guess I'll stick to my 1,6 Celeron M. If some of you think, that the whole BOINC is just about getting more&more points, then why don't you upgrade your gear constantly and focus on 3dMark, seeing all the digits getting there higher every time should give just as much satifaction as s@h. It doesn't matter what gear do you have, does it take you 3k sec for one chunk or 300k sec, what matters now is the contribution and faith in receiving phonecall from c3p0;] I'll add to that and say that I don't run my PCs 24/7 on S@H....thats not what they were thrown together for. Potentially, in GPUs, there is a huge amount of computing power that has been untapped - until now. Yes, I've also seen some pretty annoying instances, of 'CUDA wingmen' being credited for work that they should have got nothing for - I've had that happen to me, due to a bad OCZ RAM module on a P4 Prescott. The term, level playing fields, comes to mind. It is obviously going to take time for the reason for the 'failures' to be deduced and the fix to be implemented. Having said that, that could mean that the latest driver for a GPU, may not work properly with the app and have to 'roll-back' - we've probably all seen instances of this in the past. I have, hence my switch to ATI cards over 5 years ago. Personal 'glory', is not why I run this project, but through it I've found some truly amazing people - for that alone, I'm thankful. For myself, when the CUDA app is 100% reliable and not driver specific, if it takes a 'CUDA PC' 3 mins to do the same job, as it takes my PC 40 mins, well, thats the way of things. I will not, however, be rushing out to buy a CUDA card, or perform a huge m/b, CPU and RAM upgrade - because I can't justify that cost to myself, for my primary purposes for using the PCs. Nor will I try to justify it. Both PCs here, are more than adequate for their purposes with some life left in them. All I'm doing, is putting that spare time to a constructive purpose and, S@H is that constructive purpose. Don't take life too seriously, as you'll never come out of it alive! |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14679 Credit: 200,643,578 RAC: 874 |
Several days ago, and a long way down this thread, we had some disagreements about whether CUDA would add a worthwhile contribution to RAC. Nothing much came of it. Regular readers will be able to guess what happened next. Yep, I went out on Friday and bought myself a 9800GT. It was the cheapest no-brand-name value offering at the local trade counter, about 90 quid GB - say 100 Euros, 140 dollars US if the current exchange rates are to be believed. I put it in my Q9300, which has a Foxconn motherboard and some decent dual-channel DDR2 800 RAM running XP at stock (2.5GHz). So an unremarkable, middling-to-high and reasonably well-balanced cruncher. I burnt the card in for 24 hours with some stock work, then settled down for a serious 24-hour timing run (taking notice of all the various hints y'all have been providing for the last three weeks). For the timing run, I chose: LHC (host 9681553) - 2 cores. This uses a stock app, and the old-fashioned benchmark*time credit scoring, so shows a BOINC baseline for the system. Astropulse (host 4292666) - 2 cores. I ran the Lunatics r84 optimisation, a stable build one step beyond the current release but already outrun by the latest test builds. SETI Beta (host 28361) - the only safe place to run CUDA, in my opinion. LHC had lots of work, by a lucky conincidence. I was able to keep two cores fully occupied for all but the last six minutes of my 24 hours. The BOINC credit claim was a consistent 405 per core per day, for an idealised RAC of 1620 for the whole machine. But we all know that the CUDA app needs a bit of CPU time to keep it supplied. So I also measured the "Wall time" for the tasks (from starting and finishing entries in the BOINC logs), and derived a "Wall RAC" (credit per calendar day) for LHC of exactly 401 per core, 1604 overall. Evidently LHC lost about 1% to CUDA and system overhead. I did the same calculations for Astropulse, for the two complete and uninterrupted runs during the test period. With this app, I got a "CPU RAC" (taking CPU timings only) of 1382 per core, or an idealised RAC of 5530 for the whole machine. "Wall time" measurements, as before, brought this down to a "Wall RAC" of 1322 per core, 5290 overall - a loss of 4.35% to CUDA and system overhead. What about CUDA? I ran Raistmer's r396mod with the priority enhancement, to get proper 4+1 running under BOINC v6.4.5, and I aborted all VLAR tasks (known to crash) before I started - well, I meant to abort them all, but I missed a couple, so there was a slightly shaky start and a couple of reboots at the beginning of the run. All adds to the realism! Once I got going, I completed 66 MB tasks in 24 hours - just over 20 minutes each. 59 of them validated immediately against stock CPU apps (thus getting the CPU credit 'rate for the job'): 5 validated against other CUDA apps (getting inflated credit), and just 2 were still pending at the end of the run. My actual Beta credit increased by 3226 during the 24 hours: adjusting the CUDA overclaims back to CPU levels, and assuming the two pendings both validate eventually, I'm claiming a RAC of 3319 for the 9800GT. So, adding a CUDA card to an efficient cruncher, it appears that I might suffer a loss of 240 RAC from the CPUs (under 5%), and a gain of 3319 from the GPU. That's a pretty impressive payback from a close-to-stock CUDA application, and the validation rate (after some heavy VLAR aborting) isn't bad for a Beta app. Can't wait for it to be ready for release on Main! ;-) |
popandbob Send message Joined: 19 Mar 05 Posts: 551 Credit: 4,673,015 RAC: 0 |
Thanks Richard for the update. As a few of you may remember I was going to do a run of my own. Well that didn't start off so swell.. A few BSOD's a few VLAR's and I detached and ran GPUgrid for a while. Now I am back at it currently running CUDA + WCG... Aborting all VLAR and running the script to end any hung starts.. Now watch the RAC FLY! (Lets see If I can hit top 10...) Do you Good Search for Seti@Home? http://www.goodsearch.com/?charityid=888957 Or Good Shop? http://www.goodshop.com/?charityid=888957 |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14679 Credit: 200,643,578 RAC: 874 |
Addendum to the RAC analysis: During the second 24 hours (without any restarts, which probably helped), I got a CUDA credit award of 3347, plus 354 extra pendings. It's getting complicated to go back through them all and do the CUDA/CPU credit adjustments, so let's just leave the raw score as a nice round 3,700 credit per day. Now that LHC has finally finished its mammoth work issue, I switched the two spare CPU cores to Einstein, to check out their new v6.10 application - I think it's the first time an SSE2 optimisation has been released as the stock app for Windows in the current S5R4 run. (But don't all rush at once - they're talking of starting S5R5 in the next day or two). Using the same calculations as before, I got an idealised "CPU RAC" for the Q9300 of 2664, and a "Wall RAC" of 2591 - a loss of 2.7% to CUDA and system overhead. |
Francois Piednoel Send message Joined: 14 Jun 00 Posts: 898 Credit: 5,969,361 RAC: 0 |
Addendum to the RAC analysis: Let's be clear, the title of the thread is "Random Musings About the Value of CPUs vs CUDA", I have the right to say that the CPU is faster, because it is. |
W-K 666 Send message Joined: 18 May 99 Posts: 19406 Credit: 40,757,560 RAC: 67 |
Addendum to the RAC analysis: But if you were to equip a EVGA X58 3X SLI Core i7 Motherboard with three gtx 280 graphics cards the graphics cards would out perform the cpu cores. |
Francois Piednoel Send message Joined: 14 Jun 00 Posts: 898 Credit: 5,969,361 RAC: 0 |
Addendum to the RAC analysis: may be, we will see if it can make it to the Top 20. |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14679 Credit: 200,643,578 RAC: 874 |
I really like your name change to what it is now :D I am disappointed that the first appearance of "Bogus CUDA claims" in this thread appears alongside a quotation of my attempt to put real-life CUDA productivity into a mathematically-valid context. If Francois wants to question my methods, he is welcome to a copy of my BOINC daemon log and recording spreadsheet for audit purposes. However, I am glad that he has chosen the appropriate word "censured" (definitions), and not any similar word he might have been thinking of. |
Dirk Sadowski Send message Joined: 6 Apr 07 Posts: 7105 Credit: 147,663,825 RAC: 5 |
Top 20 hosts Current host #14: Intel(R) Core(TM)2 Quad CPU Q9450 @ 2.66GHz + 1 x GTX 280: Avg. credit 8,594.86 Congrats to mr.kjellen! :-) |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
Ah, my mod installed on one of top 20 hosts *DANCE* Leaving with this thought :D :D :D :D |
Dirk Sadowski Send message Joined: 6 Apr 07 Posts: 7105 Credit: 147,663,825 RAC: 5 |
Congrats to you also! ;-D Happy holidays for you! :-) EDIT: BTW. [200. post in this thread! ;-D] |
Alinator Send message Joined: 19 Apr 05 Posts: 4178 Credit: 4,647,982 RAC: 0 |
Yep... In addition, since Mr. Kjellen has been around and posting recently, it would be interesting to find out if he recalls what his 9450 was pulling before he went to CUDA Boost. ;-) Alinator |
Byron S Goodgame Send message Joined: 16 Jan 06 Posts: 1145 Credit: 3,936,993 RAC: 0 |
There's actaully several CUDA capable pc's on that list...Sure didn't take long. Wonder how long now before they're all CUDA. |
ML1 Send message Joined: 25 Nov 01 Posts: 21241 Credit: 7,508,002 RAC: 20 |
... There's actaully several CUDA capable pc's on that list...Sure didn't take long. Wonder how long now before they're all CUDA. Don't tell you-know-Who... ;-P I wonder if he's holding back his 26-core wonder to wipe us all off the face of the leader boards at the last moment for a F.... media frenzy...? :-) (Or not, as is to be expected in reality.) Happy fast GPU crunchin', Martin See new freedom: Mageia Linux Take a look for yourself: Linux Format The Future is what We all make IT (GPLv3) |
©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.