Message boards :
Number crunching :
AMD Opteron based PC
Message board moderation
Author | Message |
---|---|
Vipin Palazhi Send message Joined: 29 Feb 08 Posts: 286 Credit: 167,386,578 RAC: 0 |
One of my colleagues has recently been bitten by the seti bug and is now planning to put together a dedicated cruncher. To give him a head start, I offered my old GTX 260 to him. Being an AMD aficionado, he is planning to assemble an opteron based desktop. Although I too like AMD, I have no experience with server based PC. I have suggested AMD FX 8150 and any suitable AM3+ motherboard and to get a few GPUs. This is what he has come up with AMD Opteron 6284 SE: ~ $1480 Supermicro H8SGL or H8SGL-F: ~ $1040 The prices are what I saw online, although shipping to the middle east where we temporarily reside is not usually offered. That, apart from the high cost of the components, would make this a difficult system to assemble. I also read that this processor doesnt come with a cooling fan making it necessary to purchase an aftermarket cooler. Any advise on which one to go for, which at the same time will not block the PCI-E slot to be used for the GPU? Would any of you please give your valuable suggestions in this regard? ______________ |
Mike Send message Joined: 17 Feb 01 Posts: 34258 Credit: 79,922,639 RAC: 80 |
I´d suggest this one. http://www.noctua.at/main.php?show=productview&setlng=en&products_id=38 With each crime and every kindness we birth our future. |
tullio Send message Joined: 9 Apr 04 Posts: 8797 Credit: 2,930,782 RAC: 1 |
My SUN WS has an Opteron 1210 running 24/7 at 1.8 GHz since January 2008. It has its own cooler. But I bought an Opteron 1220 at 2.8 GHz and am planning to install it in fall, when room temperature goes down. I have bought an ASUS Silent Square EVD cooler, which looks quite impressive. Tullio |
fataldog187 Send message Joined: 4 Nov 02 Posts: 42 Credit: 1,271,261 RAC: 0 |
I like AMD too but wouldn't it be a lot more cost effective to get a much cheaper CPU and some NVIDIA GPU's instead? |
tbret Send message Joined: 28 May 99 Posts: 3380 Credit: 296,162,071 RAC: 40 |
One of my colleagues has recently been bitten by the seti bug and is now planning to put together a dedicated cruncher. To give him a head start, I offered my old GTX 260 to him. Being an AMD aficionado, he is planning to assemble an opteron based desktop. Although I too like AMD, I have no experience with server based PC. I have suggested AMD FX 8150 and any suitable AM3+ motherboard and to get a few GPUs. If this is truly about being bitten by a crunching bug, then you owe it to your friend to point-out that a $1,000 in video cards is worth at least $10,000 worth of CPUs. You would be doing your friend a great service by telling him to look at the top, oh, 500 crunchers before spending that kind of money on an Opteron and the requisite motherboard. |
Cosmic_Ocean Send message Joined: 23 Dec 00 Posts: 3027 Credit: 13,516,867 RAC: 13 |
I was using a 2p Opteron 2222SE configuration for three years (was 2p 2210 for the first two years). They were good at what they did, but clock-for-clock, Intel is the way to go for crunching. Especially when compared to Bulldozer. Having two cores share the FPU cripples floating-point performance unless you only run on half the cores, then you're just hoping that the OS will spread the tasks out evenly so that only one task is running on one FPU. If you want crunching performance and throughput and are willing to spend more money, go with Intel and a couple GPUs. If you're trying to save some money but still want some good crunching performance, go AMD and a couple GPUs. AMD CPUs and boards tend to be a bit cheaper than Intel stuff, but to each his own. Linux laptop: record uptime: 1511d 20h 19m (ended due to the power brick giving-up) |
Vipin Palazhi Send message Joined: 29 Feb 08 Posts: 286 Credit: 167,386,578 RAC: 0 |
If you want crunching performance and throughput and are willing to spend more money, go with Intel and a couple GPUs. Thank you for all the responses. As I mentioned, this guy is a hardcore AMD fanatic, he has vowed never to touch an Intel CPU. If this is truly about being bitten by a crunching bug, then you owe it to your friend to point-out that a $1,000 in video cards is worth at least $10,000 worth of CPUs. I have finally managed to convince him to settle for an AMD FX processor and an nVidia GTX 680. I also had to tell him the 'NEZ' story so that he doesnt start installing BOINC on all the office servers and PCs. Now I need to go with him to make sure he does not overspend on opterons - it is in his nature to overspend and then later crib about it. ______________ |
Link Send message Joined: 18 Sep 03 Posts: 834 Credit: 1,807,369 RAC: 0 |
I have finally managed to convince him to settle for an AMD FX processor (...) Buldozer? I don't know, if others (with experience with those CPUs) can confim that, but I don't *think* that shared FPUs are a good thing for SETI... AMD has also nice 6-core CPUs (with 6 FPUs instead of 4), if it needs to be high end AMD CPU. Check if you can find anything in the top computers list... |
skildude Send message Joined: 4 Oct 00 Posts: 9541 Credit: 50,759,529 RAC: 60 |
Having two cores share the FPU cripples floating-point performance unless you only run on half the cores, then you're just hoping that the OS will spread the tasks out evenly so that only one task is running on one FPU. EHHHHHHH Wrong. I've not seen any decrease is production speed on my FX 8150. shared FPU or not these things are very fast. I have to assume that the FPU has enough bandwidth to handle minmize any slowdown that could have occurred. $ for $ and part for part AMD is the value choice especially if you intend to just have a GPU farm. In a rich man's house there is no place to spit but his face. Diogenes Of Sinope |
Mike Send message Joined: 17 Feb 01 Posts: 34258 Credit: 79,922,639 RAC: 80 |
I had a 1090T and now a FX 8150. Overall performance of the FX is better no matter what some test may suggest. With each crime and every kindness we birth our future. |
Cosmic_Ocean Send message Joined: 23 Dec 00 Posts: 3027 Credit: 13,516,867 RAC: 13 |
Okay, so I guess I should clarify my statement about the shared FPU thing. If you compare an FX-6100 to an X6 1100T, the FX will churn out more RAC, but not by a whole lot. There are only a few various "real world tests" where 1100T actually does better, but in general, Bulldozer does win..with a small (<20% margin). I ran a battery of tests on my FX-6100 when I got it, and found some interesting results. Rather than typing it all over again.. I dug up my old post about it. However, it must be said that when I run only three cores, I end up averaging about 8%/hour per AP (comes out to 6 APs/24 hours usually). If I run all six cores, it does slow down to about 5.75%/hour. It was noted by somebody when I pointed that out a while back that 8*3 = 24, where 5.75*6 = 34.5, which is actually about 43% more productive, albeit longer crunching durations. Either way, I still feel that shared-FPU cripples the true potential, because it could be 8*6 = 48, which is 39% more than what is actually observed right now, and follows pretty closely in that 33% range that I noted with various floating-point tasks. Linux laptop: record uptime: 1511d 20h 19m (ended due to the power brick giving-up) |
tbret Send message Joined: 28 May 99 Posts: 3380 Credit: 296,162,071 RAC: 40 |
Having two cores share the FPU cripples floating-point performance unless you only run on half the cores, then you're just hoping that the OS will spread the tasks out evenly so that only one task is running on one FPU. If I were going to let my CPU do much crunching, there is no way that I'd buy a current-offering AMD CPU. My machine with the FX-8120 in it has been in a computer with a combination of things that don't seem to like each-other, or I've got some sort of driver problem I don't see going-on. I'm working on it slowly, so... you know, you have to take my observations with maybe a half-grain of salt. That being said: I'm satisfied with what the FX-8120 does, but if (for crunching) someone were to just give me either an 1100T Phenom II, or an FX-8120, my choice, I'd get the 1100T. Six FPUs is still better than four. BUT... I'm really impressed with the FX-8120s multi-tasking ability more than my 1100T or my 1090T. THAT being said: I bought a CPU for a single-GPU machine recently and went with an Athlon II x 4 because, really, for number-crunching, I cannot justify spending money on CPUs at all. As long as it's enough to feed the GPU and maybe do the random AstroPulse that comes its way (since there is no optimized NVIDIA AstroPulse app, yet), that's all I really want. And I'm just letting it do AstroPulse to be contributing a little something to that project, not because I'm trying to increase overall RAC. Honestly, I'm not seeing much advantage to any of these CPUs over a late-model P4 for feeding a single GPU. But, if I were the friend in this case, and I were contemplating an expensive Opteron and that motherboard, I'd definitely go with a high-powered i7 and its motherboard. The cost should be a fraction - albeit a significant fraction - of the server-parts cost and absolutely smoke the number-crunching ability of anything we can do with our AMD CPUs. (I am running almost exclusively AMD CPUs, and as I said, I just bought another because that's the economics of the thing. It's not like I'm an Intel cheerleader of some sort. Remove the price considerations and I'd be buying the clearly superior -- in every way but price -- Intel CPUs.) |
Mike Send message Joined: 17 Feb 01 Posts: 34258 Credit: 79,922,639 RAC: 80 |
If you overclock your FX it should easily outperform the 1100T. I'm running my FX 8150 @4.4 GHZ and 1100T has no chance reaching my processing times. With each crime and every kindness we birth our future. |
Fred J. Verster Send message Joined: 21 Apr 04 Posts: 3252 Credit: 31,903,643 RAC: 0 |
If you overclock your FX it should easily outperform the 1100T. FX 8150 : 3225.33 miljoen ops/sec 'integer' snelheid 9785.89 miljoen ops/sec. Whetstone and Drystone, respectively. A little, 3323 ops/sec and 10235 'integer' lower as an i7-2600, but still fast. But out off subject, since they're INTEL :-) |
mmstick Send message Joined: 14 Jan 12 Posts: 1 Credit: 23,316 RAC: 0 |
Having two cores share the FPU cripples floating-point performance unless you only run on half the cores, then you're just hoping that the OS will spread the tasks out evenly so that only one task is running on one FPU. Oh please, you don't know much about processor architecture do you? I own 3 Phenom II X6s and one FX-8120. 4 FX FPUs != 6 Phenom II FPUs (You are trying to compare six 128-bit FPUs to four 256-bit FPUs that have higher IPC). Secondly, my FX-8120 running 4GHz at a simply voltage of 1.27V consumes about the same as my Phenom II X6s running at 3.5GHz, but processes a good 30% faster in a majority of tasks. If the project is compiled using FMA4/AVX/XOP the gap increases. Finally, the only problem with FX is the cache misses, I suspect that if AMD had used better cache it would be demonstrating much higher IPC than Phenom II in single threaded benchmarks, but due to the amount of misses the processor must be overclocked a good amount to make up for the fault. |
Vipin Palazhi Send message Joined: 29 Feb 08 Posts: 286 Credit: 167,386,578 RAC: 0 |
Thanks for all the posts. The guy has indeed settled down for an AMD FX-8150 after seeing that I too have one of those, although I use it as my main PC for other uses as well. I would like to ask a slightly different question now. We managed to pick up a used GTX 460 at a good bargain through the classifieds. It has been running ever since. His motherboard has one more usable PCI-E slot and we have been trying to get hold of another similar card to go in there. However, it seems only the older GTX 250 cards are available. Since I only use single card on my rigs, I have no clue if two such cards can be used on a single motherboard to crunch seti? Can they use common drivers? And if so, what will be shown under 'You Computers > GPU'? ______________ |
Fred J. Verster Send message Joined: 21 Apr 04 Posts: 3252 Credit: 31,903,643 RAC: 0 |
Thanks for all the posts. The guy has indeed settled down for an AMD FX-8150 after seeing that I too have one of those, although I use it as my main PC for other uses as well. If the GTX460 is the 1st GPU, BOINC will 'see' 2 460 GPUs. But I wouldn't mix these two completely different architectures in one system. Even you can use the same driver and with only SETI MB WUs, there wouldn't be much of a difference compaired to 2 460s, if running only 1 task per GPU. On other projects, Einstein/GPUgrid etc. it's a whole different story. You'll degrade your 460 (FERMI) to a 200 series GPU! |
Vipin Palazhi Send message Joined: 29 Feb 08 Posts: 286 Credit: 167,386,578 RAC: 0 |
Thanks Fred. I will keep it in mind not to club the 400 and 200 series. A GTX480 has just now become available on the classifieds. I might as well persuade my colleague to get that and pair with his existing 460. From what I understand, two differently numbered cards from the same group should not have any issues in working together. ______________ |
Fred J. Verster Send message Joined: 21 Apr 04 Posts: 3252 Credit: 31,903,643 RAC: 0 |
Always glad to help ;-) And you're right about mixing any 400 series since they've the same architecture. (FERMI). Compute Capabillity 2.0. Although the 460 (7 CUDA-cores) and higher gives a better/higher throughput compaired to 450 and lower. A 480 is still a good choise (15 CUDA-cores). |
Fred E. Send message Joined: 22 Jul 99 Posts: 768 Credit: 24,140,697 RAC: 0 |
BOINC will only use the "most capable" gpu until you put an option in The user -provided file cc_config.xml. That option is: <use_all_gpus>0|1</use_all_gpus> If 1, use all GPUs (otherwise only the most capable ones are used). New in 6.6.25 Another Fred Support SETI@home when you search the Web with GoodSearch or shop online with GoodShop. |
©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.