[TBR] Viva La Roast! (Perpetual Thread V)

Jump to latest Follow Reply
H

Hobbes

Ars Praefectus
3,274
  • #1,166
    APSBiker":wmuqrrfy said:
    Surprising results from the 5770, I thought it would be a little faster.

    Even better off spending the same amount ($100) on a 4770, overclocking the snot out of it, and getting 500 Mkeys/sec. 50W or so ain't bad either!
    Click to expand...
    I stumbled across this post of GPU benchmarks. There's no 5770 data, but Mustard's benchmark fills that gap. The data seems consistent with what we've talked about. What I find astounding is the performance of the 5870 and 5970! :eek: I haven't found any posts related to the 5970, but I have found discussion around the 5870. I guess 58XX and above are the way to go.
     
    TheJet

    TheJet

    Ars Praefectus
    4,597
    Subscriptor++
  • #1,168
    Village":2f6oy227 said:
    I haven't been paying attention. Does this mean I need to move to that candidate for retroactive abortion of a program that is BOINC?
    Click to expand...

    NO, you most certainly should not, neither 'BOINC-enabled' project is sanctioned by the distributed.net folks, and there's no guarantee that credit granted through these programs will be retained. If someone is interested in getting BOINC capabilities added to the client [and this extends to existing application writers], they MUST contact distributed.net and work with them. This notion of running separate 'projects' that integrate BOINC with d.net is getting ridiculous, there is NO reason for going this route, especially when the d.net folks have been ASKING FOR MONTHS/YEARS for help performing just such an integration that both retains the integrity of our distribution mechanisms and our statistics.

    The latest BOINC enabled GPU client is crazy, they definitely give the notion of actual affiliation with distributed.net [even in the domain name], and they most certainly are not. Run the software at your own risk, there are NO guarantees that what is being run will not seriously harm your computer.
     
    M

    MrJackson2000

    Ars Praefectus
    3,635
  • #1,170
    Got a Gigabyte 5870 yesterday;
    Code: 
    [Apr 01 22:38:09 UTC] RC5-72: using core #0 (IL 4-pipe c).
[Apr 01 22:38:14 UTC] RC5-72: Benchmark for core #0 (IL 4-pipe c)
                      0.00:00:02.35 [1,883,096,940 keys/sec]
[Apr 01 22:38:14 UTC] RC5-72: using core #1 (IL 4-pipe c alt).
[Apr 01 22:38:21 UTC] RC5-72: Benchmark for core #1 (IL 4-pipe c alt)
                      0.00:00:03.83 [1,131,424,436 keys/sec]
[Apr 01 22:38:21 UTC] RC5-72: using core #2 (IL 4-pipe 2 threads).
[Apr 01 22:38:27 UTC] RC5-72: Benchmark for core #2 (IL 4-pipe 2 threads)
                      0.00:00:02.97 [1,486,292,704 keys/sec]
    Win7-64 w/ a bunch of stuff running.
     
    H

    Hobbes

    Ars Praefectus
    3,274
  • #1,175
    You really are starting to zoom! :D <- Happy, but green with envy.

    Your real world throughput still shows at less than half of your benchmark, which is amazing! Is that because of the difficulties you just mentioned? Even without full utilization, you've increased the team's output by about 25%. And I think you'll be able to hold your #2 spot for a while - and maybe take 1st!
     
    M

    MrJackson2000

    Ars Praefectus
    3,635
  • #1,180
    Hobbes":onyoecgt said:
    Over 27,000 blocks, #1 on the team yesterday, and you're still about 600 Million Keys/Second shy of your latest benchmark. :scared:
    Click to expand...
    Amazing isn't it? I'm still in disbelief...
    Running at priority 2 I don't notice any lag on day to day stuff so I'll be leaving it at that for now.

    Gonna be low tomorrow, got new mobo/cpu (gotta keep OGR going too :devious: ) in today so I'll be rebuilding that machine tonight.
     
    S

    SkyCaptain

    Wise, Aged Ars Veteran
    120
  • #1,183
    dnetc@dnetc.net = 7,220,720 yesterday on rc5 on the daily participant listing. So since I didn't look at the listing the day before, I'm wondering if that was for their 13 days worth (their listed time in existence) or if that is just really one day. So then the next thing is that after reading TheJet's comment regarding the dnetc@dnetc.net wrapper project, if that is daily output, is that going to modify distributed.net's thoughts on that wrapper project???

    Then the next thing is that I shot a note to the rc5 desk asking when they would have the personal stats stuff back to being able to be edited as as it sits now there is no team assignments allowed. Thus I can't get my mustard account under the ars team as it currently is not assigned to a team. The answer I got back was "I don't know, but probably a long time." which I don't consider a very informative answer really. So is anyone able to find out what the real story is with the "server maintenance" issue that has been going on for several weeks now??????

    And lastly, a benchmark for a 5770 on XP-64 using the ATI 10-3 driver/catalyst and 517 client which is greatly improved over the previous 10-2 cat and 516 client, and the -bench is with no overclocking, and the card is an XFX manufactured card:

    dnetc v2.9108-517-CTR-10021520 for ATI Stream on Win32 (WindowsNT 5.2).

    Apr 10 17:06:29 UTC] RC5-72: using core #0 (IL 4-pipe c).
    [Apr 10 17:06:37 UTC] RC5-72: Benchmark for core #0 (IL 4-pipe c)
    0.00:00:04.89 [888,052,312 keys/sec]
    [Apr 10 17:06:37 UTC] RC5-72: using core #1 (IL 4-pipe c alt).
    [Apr 10 17:06:47 UTC] RC5-72: Benchmark for core #1 (IL 4-pipe c alt)
    0.00:00:07.71 [559,879,137 keys/sec]
    [Apr 10 17:06:47 UTC] RC5-72: using core #2 (IL 4-pipe 2 threads).
    [Apr 10 17:06:55 UTC] RC5-72: Benchmark for core #2 (IL 4-pipe 2 threads)
    0.00:00:05.89 [783,343,317 keys/sec]
    [Apr 10 17:06:55 UTC] RC5-72 benchmark summary :
    Default core : #0 (IL 4-pipe c)
    Fastest core : #0 (IL 4-pipe c)

    So while the double precision issue is a show stopper for some folks in buying them, when you take into account the comparatively low power utilization and heat generation of the 5770, it is a decent crunching card now for rc5.
     
    C

    CanSpice

    Ars Tribunus Angusticlavius
    7,147
  • #1,192
    CanSpice":td3mc95j said:
    So we just happened to get a 44-node cluster here. Each node has two E5520 Xeons (2.26GHz, quad-core) in them. I'm thinking of doing a burn-in... :devious:
    Click to expand...

    Hrm. If I'm reading this correctly, if I do a 'dnetc -bench', that only tests one core, right? So I should take that number and multiply it by 4 to get the number of nodes/sec/CPU, then multiply that by 2 to get nodes/sec/server, then multiply that by 44 to get my total nodes/sec, right?

    'dnetc -bench' says I get 44,640,642 nodes/sec, so let's just round that to 44.5 Mnodes/sec. Thus I should get 178 Mnodes/sec/CPU, 356 Mnodes/sec/server, and a total of 15.7 Gnodes/sec. Does that sound right?

    If I look at last night's stats that's about double what the top person produced (and about four times what all of TBR produced). :devious:
     
    A

    Accs

    Ars Legatus Legionis
    12,557
    Subscriptor
  • #1,194
    CanSpice":1sxb5x6s said:
    If I'm reading this correctly, if I do a 'dnetc -bench', that only tests one core, right? So I should take that number and multiply it by 4 to get the number of nodes/sec/CPU, then multiply that by 2 to get nodes/sec/server, then multiply that by 44 to get my total nodes/sec, right?
    Click to expand...
    If Hyper-Threading is disabled, this is correct. If Hyper-Threading is enabled, the actual results will be lower.
     
    C

    CanSpice

    Ars Tribunus Angusticlavius
    7,147
  • #1,196
    Hyperthreading is disabled on these servers. I've watched them reboot enough times that that line is imprinted on my brain right now. :)

    Unfortunately the server room seems to have a little issue with cooling at the moment, so I won't be able to stress-test with the whole cluster. Plus one node is down (the 'health LED' on the front is blinky blinky red, which is a Bad Thing).
     
    You must log in or register to reply here.
    Jump to latest Follow Reply
    Jump to latest Follow Reply