Posts by Hydropower

1) Message boards : Number crunching : v0.18/v0.19 issues here (Message 31915) Posted 3 Oct 2009 by Hydropower Post: I'm up and running.
2) Message boards : Number crunching : v0.18/v0.19 issues here (Message 31897) Posted 3 Oct 2009 by Hydropower Post: I am happy to hear it is now working. I just bought an ATI card today (incredible I know) and have the drivers working for Folding@Home as a test. However, I seem not to get any work from Milkyway. I suspect it has to do with the fact that I run an AGP HD3850 on an old machine with Duron CPU. It does not support SSE2, only SSE. My questions: 1. do I need to manually download the GPU app ? 2. If so, where is it ?! 3. Is there a GPU app for ATI which does NOT require SSE2 ? thanks ! Hydro.
3) Message boards : Number crunching : CUDA Application for 32 bit Windows (Message 31176) Posted 19 Sep 2009 by Hydropower Post: @scibuff: The device Device Emulation (CPU) from the command line cannot be used Your CPU is using the emulation library for CUDA, hence boinc thinks it is a GPU, but the library has 'older' compute capability, so it is rejected. I am not familiar with the command line, but some searching for 'emulation' might help. I recommend switching that option off, wherever it is set.
4) Message boards : Number crunching : Problems running more than one GTX 295 (Message 31172) Posted 19 Sep 2009 by Hydropower Post: Hi. I'd take out the third card first to see if you now have enough power to run the remaining cards on the PCIe bus. Green lights on the card mean little, merely that some form of power is enough to drive some electronics, not that the card has enough power to actually operate. been there done that. Also, check the bios version of your motherboard with Asus and verify that there are no issues solved regarding 4 GTXes in newer bios revisions. Which power supplies do you have (make and model) and how are they (exact wiring, what goes where) connected to your graphics cards ? This is an essential part of the success. Please do read the pointer to gpugrid. (and thanks for that max :) ). regards, Hydro. We do have 4 monitors running with a desktop all over them, so we don't think it's a windows problem {on the other hand,we do have a huge problem getting the ASUS M4A79T Deluxe amd3 790 chipset mobo to recognize the third GTX 295, but will leave that for later}
5) Message boards : Number crunching : Probability of burning a 4870x2 ? (Message 30683) Posted 13 Sep 2009 by Hydropower Post: GTX 280 was severely burned, I could hear little metallic clicks coming out of the chip! The card can run in 2d mode, but once it hits 3d or cuda, I'd get a screen full of rainbows! Have you taken the card out and cleaned it with compressed air ? Check for bugs falling out of the card that may have got caught in it and stopped airflow, just to see if that caused it. See this interesting video: http://www.youtube.com/watch?v=s727hZ_VU9Q
6) Message boards : Number crunching : 4x HD4870X2, has anyone tried it yet ? (Message 30677) Posted 13 Sep 2009 by Hydropower Post: I hadn't seen the thread (thanks for that), but know of him and his website. I read it every now and then. Back when I designed my system I found it quite inspiring. The main issue could be driver and application related, that is why I asked the question. I know the hardware tricks, and for NVidia I also know the software side (been there done that), but I have no experience with ATI cards.
7) Message boards : Number crunching : 4x HD4870X2, has anyone tried it yet ? (Message 30394) Posted 8 Sep 2009 by Hydropower Post: Thanks for that info, I have heard that with Vista it is hard to get cards working right (also NVidia) so great to hear you actually have two double cards running. I am amazed you can get by with the 700 W.
8) Message boards : Number crunching : 4x HD4870X2, has anyone tried it yet ? (Message 30300) Posted 6 Sep 2009 by Hydropower Post: Just found the sapphire link 5 minutes ago :) (thanks) http://www.tomshardware.com/reviews/radeon-hd-4870-x2,2073-2.html I initially found a link to the HD4850x2, which had 4 connectors, but that seemed to blow the hot air into the case, not something I'd want. A pity for the multi-wu capability. I could imagine that the two or more GPU threads would run alternating to the main CPU performing tasks on the other thread. Like pipelining in a CPU, this wouuld be pipelining tasks to, in and back from the GPU. fill the bucket, give the bucket to the firefighter, empty the bucket by the firefighter, hand back the empty bucket. 4 stage pipeline.
9) Message boards : Number crunching : 4x HD4870X2, has anyone tried it yet ? (Message 30297) Posted 6 Sep 2009 by Hydropower Post: @laviathan: yes, that 64 cpu thing had caught my eye too. I can only imagine they rigged the response, use a virtual system and clustered it or they use real systems running in a cluster configuration. @Cluster Physik: from the NVidia adventure I can tell that over there the max number of GPUs supported is 8, eventhough SLI supports only 4. I would expect ATI to have done the same if the demo systems have 8 GPUs. I'd be running it under XP64. Which brands would have those 4 outputs ? Yes, the powersupply jungle is treacherous :) Finding the right one is an important key to success. Hydropower works very well generally. Thanks for the pointer to collatz. I read more on multiple workunits per GPU for ATI, Does the current CUDA MW app support multiple workunits as well ? If so, how do I activate that feature ?
10) Message boards : Number crunching : 4x HD4870X2, has anyone tried it yet ? (Message 30273) Posted 6 Sep 2009 by Hydropower Post: Hello Peter, Verstapp sounds like Verstappen, known from the area around St. Odilienberg. 'Ich hub och ne tied in roermung en hearle gewoend'. :) Love Australia. I visited last year for work, had a good view of Sydney and surrounding areas, Adelaide, a quick stay in Darwin and Alice. The Hydropower machine was supposed to run on watercooling, hence the name, until I found out that the air did just as well and saved over 1K Euro on the build.
11) Message boards : Number crunching : 4x HD4870X2, has anyone tried it yet ? (Message 30268) Posted 6 Sep 2009 by Hydropower Post: Laviathan, Thank you for your detailed post, much appreciated. I would also find it funny if you built it and then MW project shut down. :) Yes that would be hilarious :-) I had a good laugh on that one :) For the hydropower machine I use 4 700 W powersupplies, with a sync box for the on-off. Power consumption fluctuates with all cards running per project, for GPUGrid it was around 1300 Watts, for MW it is stable at 737 Watts for the entire machine with a peak to 1000, I have a power meter attached. For the new setup I would use two 1000W Silverstones, I have elaborately researched the various models and types. Money to burn, well, not really but a careful consideration. Some people go on vacations, I save the money and build rigs like these. I actually do run it as a heater, starting september temps go down here in Holland, until around april with freezing winters so the heating cost saved will partially compensate for the energy bill. Green crunching so to speak. I will not be running games on it (loud 'ahhhh' and head-shaking from the audience). Interesting point with the workunit turnaround times. I do think that Boinc lets you keep a 'supply' of work measured in time rather than workunits, so it should not be too much an issue. GPUGrid used not to supply spare workunits until it became apparent that the project would benefit from it, now there is an optimal supply of work units. 6 TFlops per card, now, that IS fast. From the list of top computers in MW, I see that UL1 has quite a few 16 CPU machines, I wonder what GPUs he is using. I'll have to ask... Ah, AL1 says in a post I'm sorry, but I have to correct you on this: I'm running 'only' three single 4870s, no 4870X2s. I once had an 4870X2, but I wasn't able to get the GPUs to run at more than 50% load each and so it turned out to be slower than a single 4870. I also tried running four single 4870s on one board, but this only looked fine on paper...in reality something slowed this configuration down and therefore it was less productive than using three cards... there you go. Regards, Hydro
12) Message boards : Number crunching : 4x HD4870X2, has anyone tried it yet ? (Message 30248) Posted 5 Sep 2009 by Hydropower Post: Has anyone built a 4-card HD4870X2 machine ? If so, what are the pitfalls to watch ? Are there other projects besides MilkyWay and FAH that specifically target ATI cards ? I am weighing the pros and cons of building a system like that. (alternate considerations are quad HD5000, to be released coming friday or the upcoming G300) I just need the info on the ATI cards, I already have all the other components.
13) Message boards : Number crunching : CUDA Application for 32 bit Windows (Message 30245) Posted 5 Sep 2009 by Hydropower Post: Things to test with error 0x01: - Overclocked ? use default clock settings (Q6700 o/c'd to 2.8GHz well...) - Overheated ? increase airflow to the GPU when temps are over 80 degrees (NVIDIA) - Broken ? Run the OCCT program (try google) with the error checking option. Most likely caues in my experience.
14) Message boards : Number crunching : v0.18/v0.19 issues here (Message 30233) Posted 5 Sep 2009 by Hydropower Post: It was built with Microsoft Visual Studio 2003. This version was done for testing in the first instance. SSE is only useful for single precision, so it doesn't provide any benefit for MW (using double precision calculations) and one sees the differences between the different compilers/switches used for that and the stock app. You should try the x87 version (built with an Intel Compiler), maybe that one is a bit faster than the stock app In fact SSE2 and up do provide packed double precision operation. (SIMD) so that might provide a good speedup. See: table B-25. formats and encodings of SSE2 floating point instructions. Intel document 253667.pdf Regards, Hydro.
15) Message boards : Number crunching : CUDA Application for 32 bit Windows (Message 30231) Posted 5 Sep 2009 by Hydropower Post: Well, the new button will not do any good, it doesn't even get to showing buttons most of the time. And this version of boinc worked just fine with GPUGrid, so it's the app's way of minimizing communication. 'Maximizing runtime on the GPU' is the positive approach to describing the same problem. FTP access works just fine, so it is the video updating that is the bottleneck. Thanks for you input though.
16) Message boards : Number crunching : CUDA Application for 32 bit Windows (Message 30215) Posted 4 Sep 2009 by Hydropower Post: (i7 with 4x GTX295, 8 GPUs) Thank you for developing this CUDA version ! I notice that the CUDA app almost completely freezes the system though. It would really help if there was more screen updating inbetween, at least for the main video adapter. In the current setting it took me an hour to regain control over the machine, after yanking the network cable to make it run out of workunits. I also notice that Boinc manager has a hard time communicating with the apps, it almost constantly shows a popup "Communicating with BOINC client, please wait". this prohibits control over Boinc. in case of a problem one cannot stop the application gracefully.