Message boards :
Number crunching :
Tesla Personal Supercomputing Gpu's
Message board moderation
Previous · 1 · 2
Author | Message |
---|---|
NX Send message Joined: 15 Apr 09 Posts: 4 Credit: 119,902 RAC: 0 |
Tim's "Asus P6T7 WS SuperComputer." is a more advanced version of my MOBO. he can run quad GPU's 1 more card than mine @ x8 http://www.asus.com/Motherboards/Intel_Socket_1156/P7P55_WS_Supercomputer/ from what I can tell I can run 3 way SLI @ 8x or 2 SLI @ 16x ( full speed ) 2 x PCIe 2.0 x16 (@ x16 or x8) 2 x PCIe 2.0 x16 (@ x8) 1 x PCIe x16 (@ x4) 1 x PCIe x1 (@ x1) Well i see what some are saying. I guess just wishful thinking of my part wanting to use a Tesla card. It would greatly speed up the crunching Maybe if the Bionic software gave full support the Tesla line of GPU's I know each mobo out there has some kind of "bottleneck" when come to GPU, CPU and other stuff. I probably could be better off with two 690's whenever the price comes down or set aside at least some money for one to get started.[/url] |
juan BFP Send message Joined: 16 Mar 07 Posts: 9786 Credit: 572,710,851 RAC: 3,799 |
Some more tips (valid only for SETI crunch, not gaming, etc.) Of course PCIe 2.0 x 16 is faster than PCIe 2.0 x 8 but for crunching SETI thats make no "big real" difference because most of the job is done by the GPU internaly, the bottleneck in this case is the ability of the CPU feeds the GPU´s when you have a lot of WU running in parallel at the same time and the memory to keep all that tasks in the RAM all the time (about 100 MB per task, switching to swap file even on a SDD is almost unconcebible). Something easy to do with the new generation of I5/I7 CPU´s and big 8Gb RAM sticks. I have few hosts with a low ending I5 (2310) feeding 2x690 on a cheap MB (ASUS P8B75-M LE with 2 PCIe 3.0 slots) and works perfect. However, the 690 runs on PCIe 3.0 it´s works even faster but as spected there are no big gain again, a simple usage of an optimized app, more WU per GPU, OC on GPU/CPU/Memory etc. will produce a lot more gain. But be clear, that´s for SETI crunching only, diferent projects have diferent CPU/GPU/IO rate of usage, so in some cases the CPU/IO speed makes a lot of diference. And optimized cruncher running only SETI and by haighly optimizing the PCIe usage could obtain a 2% of increase on its RAC, but you could do nothing else on that host, the video respounse times make impossible to work in it even with simple aplications. And remember, at least for some time, until the low speed DL´s and the 100WU limit where rised, the real bottleneck of a heavy cruncher is that limit. Actualy (at least at my side of the world due the constant conection problems with the SETI lab - nothing related to my ISP) an I5 with 2x690 running just 2WU (could do 3) at a time (a total of 8 GPU WU/time) could crunch more WU than it receives (imagine on a 3x690, or in a shorties storm) and constanly switches back to Eistein (the backup project). (edit) About Tesla, the gain could be obtained if SETI move double precision, but who needs double precision if you could do the math with single precision? Don´t expect any news on that area for a while. They have so few resources to even try to spend them on that task. Go for Tesla only if you use/need a high end specific software that needs this impressive but highly expensive card, no the SETI case. |
Tim Send message Joined: 19 May 99 Posts: 211 Credit: 278,575,259 RAC: 0 |
The issue is not limited to windows : It is a CPU /MB capacity issue. No I am running 4x590 as you can see at my valid wu's. I didn't have the problem from 275.33 and earlier versions, but I don't complain. It is working perfectly with that version, so .... if it is working... don't change it. About the CPU, I don't think it is CPU issue even with 4 CPU's. The rig is working 24/7 with the 980x at 20% max. (i do only GPU wu's , not Cpu). And the 980x is more than enough. The only thing I deactivate from the BIOS is the sound chip. I do not need sound on that pc. |
juan BFP Send message Joined: 16 Mar 07 Posts: 9786 Credit: 572,710,851 RAC: 3,799 |
Thanks for your answer and help. Sorry was my mistake, we are talking about 4x690 on one host and somebody refear you use 4GPUs on your MB, and i don´t check that info. Now i see you use 4x590 that´s exacty what i say, 4x590 works (on MB who have the PCIe slots avaiable of course), 4x690 still NOT know one who put to work in the SETI world at least (plain normal GPUs without any BIOS enhancements), any other combination works. (edit) Yes Tesla GPU´s works normaly for SETI crunch, they are just to expensive compared to common comercial GPUs (GForces or ATIs) to do the job |
shizaru Send message Joined: 14 Jun 04 Posts: 1130 Credit: 1,967,904 RAC: 0 |
...I guess just wishful thinking of my part wanting to use a Tesla card. It would greatly speed up the crunching Maybe if the Bionic software gave full support the Tesla line of GPU's If this is in reference to what I said: a) They are far better at Double Precision (which however SETI does not use) maybe it would have been better if I said: "...which however SETI does not need" AFAIK Seti works with Tesla cards. |
zoom3+1=4 Send message Joined: 30 Nov 03 Posts: 65738 Credit: 55,293,173 RAC: 49 |
The issue is not limited to windows : It is a CPU /MB capacity issue. Juan, you can call Me Victor, most everyone else does. Then I'll change that main psu to an EVGA SuperNOVA NEX1500 Classified 1500W Modular Gold rated power supply(1500w @ 125A w/115vac or 1650w @ 133A w/230vac), the 450w psu can power the 4th card still, the EVGA is a single rail psu. 70% of 125A is 87.5A @ 1500w, 70% of 133A is 93.1A @ 1650w. I'd love to try the x41zb build, but that isn't available to Me, I have x41g and x41u, I'm currently using x41g. My current Silverstone 1500w Silver rated Power Supply will have to be supplemented by a 650w Tt video card psu that I have for the 3rd GTX590 card, the 650w psu is a 2 bay psu, unlike the 450w which uses just 1 bay. The T1 Trust, PRR T1 Class 4-4-4-4 #5550, 1 of America's First HST's |
Team kizb Send message Joined: 8 Mar 01 Posts: 219 Credit: 3,709,162 RAC: 0 |
|
zoom3+1=4 Send message Joined: 30 Nov 03 Posts: 65738 Credit: 55,293,173 RAC: 49 |
@ Tim It's that or He'd be having a thermonuclear meltdown... Oh and glad see everybody... The T1 Trust, PRR T1 Class 4-4-4-4 #5550, 1 of America's First HST's |
Lee Gresham Send message Joined: 12 Aug 03 Posts: 159 Credit: 130,116,228 RAC: 0 |
I have been pondering since the Supercomputing Gpu's came out few years ago if is really i mean really worth it forking out $4000 for one of those cards. With all the video cards out there like the GTX 500's and GTX 600's look like cheap alternative to Nvidia's Supercomputing Gpu's. ever improving cuda technology I want to make a super cruncher. But which one will it give me some good bang for my buck? Back to the question, here is some info to put the Tesla in perspective I built an i7 960 machine in the summer of 2010 with lots of ram a 1200 watt ps and a Tesla C1060. The C1060 has 4GB ram and about 244 shaders. The PC uses a GT 220 for video that does not run any projects. Prior to Seti's latest hickups the i7 pc's RAC were usually between 21K and 24K with the Tesla cranking out a work unit every 10 to 12 Minutes. Another pc I'm running is a core 2 quad 2.66GHZ with 6GB ram and an EVGA GTX 470 with 1.2GB ram and 448 shaders . It's RAC runs neck and neck with the i7 and it cranks out work units at about 7 to 9 minutes . 2 other pc's I run use LGA 775 P4s @ 3.8 GHZ. One is a GTX 560Ti and the other a GTX 560. The Ti was at 17.5 RAC and the plain 560 was in the 15K range. All my pcs run Seti 100% cpu and gpu time 7/24. The Tesla has been a work horse but I think any fermi from the 560s up would out run it. I'd like to retire the Tesla with another GTX 470 and see what happens. Delta-V |
Tim Send message Joined: 19 May 99 Posts: 211 Credit: 278,575,259 RAC: 0 |
@ Tim No, it is on air. I will try to post a picture. Tim |
Lee Gresham Send message Joined: 12 Aug 03 Posts: 159 Credit: 130,116,228 RAC: 0 |
Alas, services will be Friday for the GTX 560 that died without work during the outage. Delta-V |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
On the other side of town, AMD says they just doubled the speed of their Parallel Processing APP. They have a reference behind the claim, I'm still looking for the meaning of that(2). SUNNYVALE, Calif. 12/4/2012 "The APP SDK 2.8 includes dozens of new and improved samples for OpenCL™, Aparapi and C++ AMP that deliver significantly faster performance than APP SDK 2.7 – up to 2.3x faster(2) on average in nine key benchmarks..." The rest of the story is here, AMD Paves Ease-of-Programming Path to Heterogeneous System Architecture with New APP SDK 2.8 and Unified Developer Tool Suite In case you haven't seen them, there are some nice videos here Radeon HD 7990 And GeForce GTX 690: Bring Out The Big Guns |
Tron Send message Joined: 16 Aug 09 Posts: 180 Credit: 2,250,468 RAC: 0 |
Juan wrote: Sorry was my mistake, we are talking about 4x690 on one host and somebody refear you use 4GPUs on your MB, and i don´t check that info. The confusing post was mine ,when I refered to a different mother board than Tim is using. The motherboard that should support 4X 690 gpu's is an Asus Z9PE-D8 WS dual cpu lga 2011,.. based on it's specifications not actual testing. |
juan BFP Send message Joined: 16 Mar 07 Posts: 9786 Credit: 572,710,851 RAC: 3,799 |
Juan wrote:Sorry was my mistake, we are talking about 4x690 on one host and somebody refear you use 4GPUs on your MB, and i don´t check that info. By specs yes i agree with you, in real world i still want to see one working with 4x690, if anyone do that please share with us how to do, that will be very interesting. All the info i have is not a MB problem is a GPU dessing problem and only happens on the 690, but you can´t said is a limitation, very few people have a real use for 4x690 on a single host. I try few diferent MB but not the model you show, they are not avaiable here, at least in the oficial market. |
©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.