21)
Message boards :
Number crunching :
MilkyWay_GPU
(Message 20459)
Posted 25 Apr 2009 by jedirock Post: I have my HD4870 + Q6600 (95W) OC'd to 3.26GHz + 4x DDR2-1066 + 2 SATA + 2 IDE on a 750W PC Power and Cooling Silencer unit. It's plenty overkill, as this machine draws about 420W on full load, but that's fine with me. Yes, it has 2x 6-pin and 2x 6/8-pin PCIe connectors on it. |
22)
Message boards :
Number crunching :
MilkyWay_GPU
(Message 20454)
Posted 25 Apr 2009 by jedirock Post:
Sounds about right. I have my HD4870 + Q6600 (95W) OC'd to 3.26GHz + 4x DDR2-1066 + 2 SATA + 2 IDE on a 750W PC Power and Cooling Silencer unit. It's plenty overkill, as this machine draws about 420W on full load, but that's fine with me. |
23)
Message boards :
Number crunching :
MilkyWay_GPU
(Message 19965)
Posted 22 Apr 2009 by jedirock Post: I went with the 8800GT because my mobo looks like it only supports a single pci-e card. You only need a single PCIe slot for the dual slot cards. The second slot isn't taken up by another connector, but by a bigger cooler. |
24)
Message boards :
Number crunching :
MilkyWay_GPU
(Message 19294)
Posted 18 Apr 2009 by jedirock Post: ID #30. :-D And team founder to boot. And a friendship request from Misfit already too... No. |
25)
Message boards :
Number crunching :
Posting restrictions
(Message 19048)
Posted 16 Apr 2009 by jedirock Post: Hey folks, It's probably because he's the admin. Thanks Dave, should be appreciated. I haven't seen any spam, but presumably that's because you've been moderating it faster than I've been reading it. :) |
26)
Message boards :
Number crunching :
3rd.in - optimized apps
(Message 18601)
Posted 13 Apr 2009 by jedirock Post: Are there really any performance difference between jedirock's release and the stock client...? None taken. I wasn't expecting that much difference anyway, just tweaking of the GCC flags, which were fairly good already. |
27)
Message boards :
Application Code Discussion :
GPU/CUDA Milkyway
(Message 18390)
Posted 12 Apr 2009 by jedirock Post: from what i hear the nvidia gpu card will have to be a gtx 260 or better,,, Yep. Only the GT200 based cards can do double precision calculations. So that means the GTX 260 or higher (275, 280, 285, 295, etc). The GTS 250 and other cards similar to that are just rebranded 9xxx series cards, so they don't have the necessary hardware. |
28)
Message boards :
Number crunching :
Interview with Gipsel
(Message 18280)
Posted 11 Apr 2009 by jedirock Post: Isn't this the same thing that was posted here: http://milkyway.cs.rpi.edu/milkyway/forum_thread.php?id=750 some three weeks ago? D'oh, I missed that. Only reading two forums comes and bites me in the rear... Oh well, he can always use more fame, as you said. :) |
29)
Message boards :
Number crunching :
Interview with Gipsel
(Message 18213)
Posted 10 Apr 2009 by jedirock Post: An interview with Gipsel from Planet3DNow on the MW clients. http://www.brightsideofnews.com/news/2009/3/22/interview-milkywayhome-meets-the-power-of-graphics.aspx |
30)
Message boards :
Number crunching :
GPU units hanging
(Message 18131)
Posted 10 Apr 2009 by jedirock Post: Remember you ether have to restart BOINC or in BOINC manager - I've always had to do: Advanced->Read config file to get changes to the cc_config to apply, but it does work. |
31)
Message boards :
Number crunching :
ATI 4750/4770/4890 support
(Message 17960)
Posted 8 Apr 2009 by jedirock Post: nVidia, with CUDA, is well supported and publicised, and are in a better position software wise to be utilised. Hence why more projects support it. No, I think what happened was that Nvidia gave the BOINC devs some money and a couple cards to implement CUDA first. Nvidia also wanted it to coincide with a press release from them, which is why BOINC 6.4.x was released in such a sorry state, but with CUDA crunching. |
32)
Message boards :
Number crunching :
4870 and 3870
(Message 17958)
Posted 8 Apr 2009 by jedirock Post: It should Crossfire-X, though both would run at 3870 quality (at least for gaming). For crunching, it's an unknown factor. Give it a go? =P No, you cannot Crossfire a 38xx and a 48xx card together. |
33)
Message boards :
Number crunching :
3rd.in - optimized apps
(Message 17853)
Posted 7 Apr 2009 by jedirock Post: The 661mb File I Downloaded was a zip file too, but when opened it had an .iso file inside and that's how I burned it as an .iso file to a CD. Did you check for the renamed files in C:\Windows\System32? They're described in the app readme. |
34)
Message boards :
Number crunching :
3rd.in - optimized apps
(Message 17852)
Posted 7 Apr 2009 by jedirock Post: Again, the only Driver that will install is the v8.60 Driver that Diamond put on the CD that came with the Video Card. This Driver isn't even listed @ the ATI Site so I don't know where it came from So Catalyst 9.3 uses the 8.591 driver, while the 4890 needs the 8.6 driver, and Catalyst 8.6 came with a different driver... Who the heck came up with that numbering system? |
35)
Message boards :
Number crunching :
3rd.in - optimized apps
(Message 17834)
Posted 7 Apr 2009 by jedirock Post: So far this is all I can get from the 4890 using the 8.60 Drivers which are the only ones I can get to install so far You'll need 8.12 or later to run the GPU app. I ran 8.12 for a while on my machine, and 9.3 seems to be working just fine with the fix specified in the readme file included with the app. |
36)
Message boards :
Number crunching :
3rd.in - optimized apps
(Message 17564)
Posted 4 Apr 2009 by jedirock Post: what version would be right for a asus a8n sli-prem(amd cpu) and a 9800gtx with latest patch The GT200 based Nvidia cards are capable of double-precision, but they have less shader units dedicated to the task than comparable ATI cards. |
37)
Message boards :
Number crunching :
ATI 4750/4770/4890 support
(Message 17319)
Posted 2 Apr 2009 by jedirock Post: Only x8xx cards are being supported, as they support double precision. This project relies on double precision accuracy, that x7xx cards do not provide. I made a thread about the issues of single vs. double precision here : http://milkyway.cs.rpi.edu/milkyway/forum_thread.php?id=764 I don't know how that stands against the HD4770 though. Reputed performance puts it above the HD4830 and almost to the HD4850. It comes with GDDR5 memory as well. If those stats are indicative of anything, it may well be capable of double-precision. |
38)
Message boards :
Number crunching :
All jobs fail on one computer
(Message 17306)
Posted 2 Apr 2009 by jedirock Post: One of the stacktraces I looked at from a workunit said it was using the SSE4.1 application, which, as you're probably aware, the E4500 doesn't support. Make sure you're using the SSSE3 or lower application. |
39)
Message boards :
Number crunching :
Optimized OS X Applications
(Message 17278)
Posted 1 Apr 2009 by jedirock Post: I will post over on his forums and see if he can look at your code. It's the stock source with a tweaked Makefile, which isn't going to help him if he's using icc. |
40)
Message boards :
Number crunching :
Optimized OS X Applications
(Message 17248)
Posted 1 Apr 2009 by jedirock Post: The question that comes to mind is how much difference is there between the Linux app and the OS X app. Crunch3r seems to have a good fix for releasing the optimized AP apps for both Linux and OS X. I believe Crunch3r also has the Intel compiler. If he does have MW compiles for OS X, it'd be interesting to compare their speed and what ICC can actually do compared to GCC. |
©2024 Astroinformatics Group