Message boards :
Number crunching :
Cost effectiveness?
Message board moderation
Previous · 1 · 2
Author | Message |
---|---|
Grant (SSSF) Send message Joined: 19 Aug 99 Posts: 13755 Credit: 208,696,464 RAC: 304 |
Its more like You can also get a Sandy Bridge capable motherboard for much, much less, but the example quoted was high end AMD v high end Intel, so those were the examples i used. Grant Darwin NT |
HAL9000 Send message Joined: 11 Sep 99 Posts: 6534 Credit: 196,805,888 RAC: 57 |
I assumed cost effective meant cheaply built. Slapping several GPU's on your computer will cost you on the back end. Your electric bill will go up. Depending on your time frame new parts will come out and change what part is the best performance per watt. The "best bang for the buck" and "best PPW" questions come up every few months. Tom's Hardware did an efficiency comparison back in Jan which might be of interest to you. Also a run with medium to high end graphics cards. Based on the power usage, performance, & cost of the 965, 1100T, and 2500K. I'd pick the 2500K for a CPU. The prices of the CPUs I just looked on Newegg quickly and found 965=~$160, 1100T and 2500K=~#230. I'm sure with more searching better prices could be found, but they probably scale about the same. If you add a GPU for processing as well I think your 100w system power might want to be more like 150-250. Depending on which card you choose to go with. SETI@home classic workunits: 93,865 CPU time: 863,447 hours Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[ |
Grant (SSSF) Send message Joined: 19 Aug 99 Posts: 13755 Credit: 208,696,464 RAC: 304 |
If you add a GPU for processing as well I think your 100w system power might want to be more like 150-250. Depending on which card you choose to go with. To give people a rough idea. My 1.5kVA UPS claims it's load is around 34% (with the monitors off, network switch & modem still on). That's an i7 2600 at default settings (4 cores, 8 threads all doing Seti) with a GTX460 (doing 2 Work Units at a time). 34% of 1.5kVA = 510VA. Assuming a Power Factor of around .8 gives us 408W. Grant Darwin NT |
spitfire_mk_2 Send message Joined: 14 Apr 00 Posts: 563 Credit: 27,306,885 RAC: 0 |
The GT 240 is equivalent to old nVidia 9600 cards, you do not want to waste your money on GT 240. I am shopping for GTS 250, 70-80 dollars for used one. One time saw a refurbished GTX 260 or 270 for 100 dollas, I should have bought it. |
hbomber Send message Joined: 2 May 01 Posts: 437 Credit: 50,852,854 RAC: 0 |
I was never interested in exactly measuring how one AMD CPU would cripple my RAC, but I made my conclusions based on some other numbers I've seen. With MAR WU, CPU time used with my 2500K(OCed, let's not forget it) is 35-40 seconds. On clocked 1090Т @3.6, I've seen CPU time of 120-140 seconds, or rougly three times longer. For VHARs on my CPU it takes 17-20 seconds(where there is fixed part, unit preparation phase which is fixed - 11-12 seconds for any unit). If we aply same ratio, this will give us 60 seconds for above 1090T processor. Lets assume and say CPU time for 1090T is 30 seconds longer. Such VHAR units, my 460 finished is 2:30 mins. Those thirty seconds are 1/5, or 20% of total time, used for crunching. On my 470s, VHARs were bellow, or 2 minutes long, shorter than on 460s. So the ratio increases from 1/5 to 1/4. And so on, in this manner, for faster cards. I know those numbers are crude, but they can give a basic idea. My choice is to run two cores only, leaving one free core per each GPU. If I run more cores, causing GPUs to wait time to time, bcs CPU being busy, it increases CPU time for a unit, and increases its total time also. But not much. It was proven again while measuring average GPU usage for a certain amount of time. Anyway, running 4 cores would bring me higher RAC, but not as high as adding two more CPU cores to crunch to my current crunching "setup" - two cores only and one core for each GPU left free. I wrote a program to increase GPU processes priority to High from Bellow Normal, when running 4 CPU cores, but results were inconclusive. And, in context of 1090T, I suspect this impacts on GPU performance, depending on free CPU cores u leave and its overall speed, are even greater. One more thing to know - leaving CPU with no work, just using GPUs for crunching, and in same time C-states for CPU are on, allowing it to lower its frequency and voltage + working Speedstep, will actually increase nearly twice CPU time per unit. I suspect Speedstep is most guilty of it. Bcs need of CPU cycles in SETI is sporadic, it makes CPU to increase and decrease its speed many times, very often. And bcs those transitions need time, the often they are, the worse it gets - lost time in transitions matters. When I left my 920 with constant multiplier of 21, there was no time lost. With C1E enabled it doubles CPU time per unit - from 30 seconds to 60 seconds. Btw, this may explain so long CPU times for those units I saw on above mentioned 1090T, but since I'm not aware how things work with AMD, I cannot say anything more precisely. And also, on my 2500K system, turning on-of power saving functions, or using fixed vCore coltage leads to different results, compared to my 920 systems. I havent found any system in it yet, but I only know how it doesnt increase CPU time per unit - not not turn on Speedstep and don't use offset voltage. |
Mike Send message Joined: 17 Feb 01 Posts: 34264 Credit: 79,922,639 RAC: 80 |
Have tried that out already. Didn´t change much on GPU times. But RAC dropped over 1000 in case running 5 cores instead of 6. Run times on GPU didn´t improve by 5 minutes per task. So not worth on my 1090T. May be different from host to host. With each crime and every kindness we birth our future. |
©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.