Optimize your GPU. Find the value the easy way.
Optimize your GPU. Find the value the easy way. |
|
| log in |
Message boards : Number crunching : Optimize your GPU. Find the value the easy way.
Previous · 1 . . . 8 · 9 · 10 · 11 · 12 · 13 · Next
| Author | Message |
|---|---|
|
I`ve got 700 already. | |
ID: 1287930 ·  | |
I`ve got 700 already. You've crunched a WU that took 7 times longer than normal? | |
| ID: 1287934 · | |
|
Josef W. Segur  Volunteer developer Volunteer tester Send message Joined: 30 Oct 99 Posts: 3930 Credit: 934,263 RAC: 14  | |
I've been running V1.4 for 24 hours now!:D Are these "real" WUs? Or am I doing something wrong? I chose: The PG set are shortened WUs, the FG set are full length. The names indicate the angle range with an implied decimal point after the first digit. So FG00091 and FG00134 are VLAR tasks which would not normally be sent to CUDA, the others so far are rarities; the set was chosen to provide samples across the range, not to represent typical distribution. The _v7 in the filenames indicates that the autocorrelation search supported by SETI@home v7 will be done in addition to the other four search types. That will mean that even the WUs with more typical AR will take longer to process than SETI@home Enhanced tasks of the same AR. Joe | |
| ID: 1288101 · | |
|
This has me confused.. Just for fun I tried this on two computers both with GTX 260's and both comp's came back at .5 as the optimal setting.. So I decided to actually try it on one comp and so far it is a tiny bit faster running at .5.. I'm lucky in that I have a group of work units that are all the same day and angle.. It crunched 8, two at a time, at 6:35 each.. And so far it's crunched 4 or 5 more, 1 at a time, at 3:44 each.. | |
| ID: 1288268 · | |
 S@NL - eFMer - www.efmer.eu/boincVolunteer tester  Send message Joined: 7 Jun 99 Posts: 497 Credit: 110,883,844 RAC: 170,102  | |
This has me confused.. Just for fun I tried this on two computers both with GTX 260's and both comp's came back at .5 as the optimal setting.. So I decided to actually try it on one comp and so far it is a tiny bit faster running at .5.. I'm lucky in that I have a group of work units that are all the same day and angle.. It crunched 8, two at a time, at 6:35 each.. And so far it's crunched 4 or 5 more, 1 at a time, at 3:44 each.. Normally it's even slower running more than 1. My guess it got something to do with feeding the GPU. Maybe the CPU doesn't got the time to feed the GPU properly. And with 2 CPU tasks it may get more time allocated. This means leaving a CPU core dedicated to the GPU feeding may help. Or set a higher priority on the feeder task. You can try this program: http://www.efmer.eu/forum_tt/index.php?topic=198.0 and set the priority higher. ____________ TThrottle Control your temperatures. BoincTasks The best way to view BOINC. Anza Borrego Desert hiking. | |
| ID: 1288334 · | |
This has me confused.. Just for fun I tried this on two computers both with GTX 260's and both comp's came back at .5 as the optimal setting.. So I decided to actually try it on one comp and so far it is a tiny bit faster running at .5.. I'm lucky in that I have a group of work units that are all the same day and angle.. It crunched 8, two at a time, at 6:35 each.. And so far it's crunched 4 or 5 more, 1 at a time, at 3:44 each.. Yeah me too. I've got the same dumb look on my face you do. I tried to kickstart this conversation quite a few posts up when Fred tested his laptop and all I got was... tumbleweeds. So I too decided to to try 2 at a time (like you said "just for fun") on the tiniest of mobile GPUs from the 200 series and whadayaknow... 2 was a bit faster than one for me too. I thought it was big news, but no-one seems to care. The good news is that thanx to Fred we finally have a much needed Seti benchmarking tool for idiots. Fred, I know you have been "advertising" this tool as a one-off way to check "count" but I think you have created a monster of a benchmarking tool for us Seti noobs without even realizing it!:D From a post I made over 180 days ago in the Lunatics Windows Installer v0.40 Release Notes thread: ... If I had a "pretty" benchmark app I would go OCD on the thing. I'd check every WHQL nVidia driver from late 250's to current 290's. I'd turn Windows eye-candy, services and processes on and off. I'd play with nVidia settings. And whenever Jason-G would come out of left-field with, "Oh, you know, it could be your Wi-Fi that's interfering with crunching" I'd check that too:) I say "pretty" because I don't know how to work with black & white windows. I need installers and progress bars and buttons I can click on and things to hold my hand:) Of course I'm sure you guys have next-to NO time for such a thing...And thanx to a thread Jim_S created a couple of days ago, I got the idea that you can even use this tool to check how many CPU cores are optimal for GPU feeding. So Fred, you just crossed out the biggest item on my Seti wishlist! Thank you!:D PS @Zule I have NO idea how this got by our gurus. My theory ATM is that no-one thought to check/benchmark "count" when the 270 drivers came out. Those drivers came out almost 3 years after your 260 did... | |
| ID: 1288363 · | |
|
Richard Haselgrove Volunteer tester Send message Joined: 4 Jul 99 Posts: 7452 Credit: 39,523,206 RAC: 32,194  | |
|
The answer has been posted on the boards before, but it's worth saying again. Here's one relatively recent iteration, from message 1259625: it seems that multiple simultaneous WUs don't scale well unless on Fermi or later cards. So, the bullet points are: * 200 series can run two at once, and have always been able to - provided there's enough memory * The benefits are marginal at best, and much lower than the speed-up you'd get on a Fermi | |
| ID: 1288379 · | |
I think Fredo may have come up with the answer when he said it might help if you dedicated a CPU to it.. I'm currently only crunching GPU on this comp so I have a Phenom II 965 feeding the 260... As for the drivers, I've done some benchmarking and 275.33 are the fastest for me. Seen a few other people say the same thing, yet everyone wants those latest drivers;) | |
| ID: 1288423 · | |
So we accidentally found a flaw in the testing process since when the tool runs the CPU is dedicated to the GPU. Uninformed users may take the results from 200 series cards and below and use them in normal operation. Without a dedicated core/cpu like me it may have negative results.. Btw.. Love the tool and love Boinc Tasks:) | |
| ID: 1288431 · | |
|
@Zule I think Fred misunderstood what you were saying. It looks like he answered to "why is my GPU slower at doing 2 WUs at a time" which isn't what you asked:) I could be wrong though... Fred? | |
| ID: 1288444 · | |
|
GTX660 (Palit factory OC'ed, Base clock 1006Mhz, Boost clock 1072Mhz) | |
| ID: 1289086 · | |
GTX660 (Palit factory OC'ed, Base clock 1006Mhz, Boost clock 1072Mhz) Looks like a toss up between 2 & 3. Only a few seconds gained for the extra WU for 3 at a time, but overtime i guess it would all add up. ____________ Grant Darwin NT. | |
| ID: 1289099 · | |
|
There is nothing automatic about this test. Notice how nearly 100% of this discussion involves Volunteer Testers. Perhaps version 1.5 or 1.6 could actually be automatic. There are millions of programs out there that all one need do is download and run and they work, perhaps one day this will become one. | |
| ID: 1289280 · | |
|
Hey Snowmain! | |
| ID: 1289299 · | |
GTX660 (Palit factory OC'ed, Base clock 1006Mhz, Boost clock 1072Mhz) Times from my MSI GTX560 factory OC. I thought GTX660 was faster than a GTX560, wonder if we tested using the same WU? 15 September 2012 - 14:59:07 Start, devices: 1, device count: 2 (0.50) --------------------------------------------------------------------------- Results: Device: 0, device count: 2, average time / count: 305, average time on device: 152 Seconds (2 Minutes, 32 Seconds) 15 September 2012 - 15:05:41 Start, devices: 1, device count: 3 (0.33) --------------------------------------------------------------------------- Results: Device: 0, device count: 3, average time / count: 448, average time on device: 149 Seconds (2 Minutes, 29 Seconds) | |
| ID: 1289308 · | |
|
S@NL - eFMer - www.efmer.eu/boinc Volunteer tester Send message Joined: 7 Jun 99 Posts: 497 Credit: 110,883,844 RAC: 170,102 | |
Are you using 1.4 ? Simply unpack the zip in a folder and run it. Sometimes something unexpected happens. Use explorer and try to open this folder: C:\videocardtest\SetiPerformance_32_64_1_4\zip\x41g\ Try opening C:\Users\Jim\AppData\Roaming\eFMer\SetiPerformance and check what is in slot0. Helping me solve this will help others and will make a better V 1.5 ____________ TThrottle Control your temperatures. BoincTasks The best way to view BOINC. Anza Borrego Desert hiking. | |
| ID: 1289309 · | |
|
S@NL - eFMer - www.efmer.eu/boinc Volunteer tester Send message Joined: 7 Jun 99 Posts: 497 Credit: 110,883,844 RAC: 170,102 | |
What I see so far is that the 6xx series is actually slower on the g. It has more cuda cores but they are less capable. The Beta Z series suggest that the 6xx series will be a bit faster with that program version. ____________ TThrottle Control your temperatures. BoincTasks The best way to view BOINC. Anza Borrego Desert hiking. | |
| ID: 1289310 · | |
|
S@NL - eFMer - www.efmer.eu/boinc Volunteer tester Send message Joined: 7 Jun 99 Posts: 497 Credit: 110,883,844 RAC: 170,102 | |
I can't find them. \r\n \r\n\r\n (D A) only. Give me a specific location. ____________ TThrottle Control your temperatures. BoincTasks The best way to view BOINC. Anza Borrego Desert hiking. | |
| ID: 1289316 · | |
|
MY advise..... | |
| ID: 1289317 · | |
It is not 'specific', it is on every line break:
Offset(h) 00 01 02 03 04 05 06 07 08 09 0A 0B 0C 0D 0E 0F
00000000 30 39 2F 32 35 2F 32 30 31 32 2C 20 32 30 3A 30 09/25/2012, 20:0
00000010 36 3A 32 33 20 2D 2D 20 53 74 61 72 74 75 70 0D 6:23 -- Startup.
00000020 0D 0A 30 39 2F 32 35 2F 32 30 31 32 2C 20 32 30 ..09/25/2012, 20
00000030 3A 30 36 3A 32 34 20 2D 2D 20 53 74 61 72 74 20 :06:24 -- Start
00000040 72 65 61 64 69 6E 67 20 63 6F 6E 66 69 67 2E 78 reading config.x
00000050 6D 6C 0D 0D 0A 30 39 2F 32 35 2F 32 30 31 32 2C ml...09/25/2012,
00000060 20 32 30 3A 30 36 3A 32 34 20 2D 2D 20 3C 64 61 20:06:24 -- <da
00000070 74 61 3E 20 3C 6E 61 6D 65 3E 4D 42 36 5F 77 69 ta> <name>MB6_wi
00000080 6E 5F 78 38 36 5F 53 53 45 33 5F 4F 70 65 6E 43 n_x86_SSE3_OpenC
00000090 4C 5F 41 54 69 5F 48 44 35 5F 72 33 39 30 3C 66 L_ATi_HD5_r390<f
000000A0 6F 6C 64 65 72 3E 4D 42 36 5F 77 69 6E 5F 78 38 older>MB6_win_x8
000000B0 36 5F 53 53 45 33 5F 4F 70 65 6E 43 4C 5F 41 54 6_SSE3_OpenCL_AT
000000C0 69 5F 48 44 35 5F 72 33 39 30 3C 65 78 65 3E 4D i_HD5_r390<exe>M
000000D0 42 36 5F 77 69 6E 5F 78 38 36 5F 53 53 45 33 5F B6_win_x86_SSE3_
000000E0 4F 70 65 6E 43 4C 5F 41 54 69 5F 48 44 35 5F 72 OpenCL_ATi_HD5_r
000000F0 33 39 30 2E 65 78 65 3C 70 61 72 61 6D 65 74 65 390.exe<paramete
00000100 72 3E 20 2D 2D 64 65 76 69 63 65 20 25 64 65 76 r> --device %dev
00000110 69 63 65 20 2D 69 6E 73 74 61 6E 63 65 73 5F 70 ice -instances_p
00000120 65 72 5F 64 65 76 69 63 65 20 34 20 2D 70 65 72 er_device 4 -per
00000130 69 6F 64 5F 69 74 65 72 61 74 69 6F 6E 73 5F 6E iod_iterations_n
00000140 75 6D 20 32 30 20 0D 0D 0A 30 39 2F 32 35 2F 32 um 20 ...09/25/2
Offset(h) 00 01 02 03 04 05 06 07 08 09 0A 0B 0C 0D 0E 0F
00000000 53 74 61 72 74 69 6E 67 20 61 75 74 6F 6D 61 74 Starting automat
00000010 69 63 20 74 65 73 74 3A 20 28 41 50 36 5F 77 69 ic test: (AP6_wi
00000020 6E 5F 78 38 36 5F 53 53 45 32 5F 4F 70 65 6E 43 n_x86_SSE2_OpenC
00000030 4C 5F 41 54 49 5F 72 31 33 31 36 29 0D 0D 0A 32 L_ATI_r1316)...2
00000040 35 20 53 65 70 74 65 6D 62 65 72 20 32 30 31 32 5 September 2012
00000050 20 2D 20 32 30 3A 31 38 3A 33 37 20 53 74 61 72 - 20:18:37 Star
00000060 74 2C 20 64 65 76 69 63 65 73 3A 20 31 2C 20 64 t, devices: 1, d
00000070 65 76 69 63 65 20 63 6F 75 6E 74 3A 20 31 20 28 evice count: 1 (
00000080 31 2E 30 30 29 20 0D 0D 0A 32 35 20 53 65 70 74 1.00) ...25 Sept
00000090 65 6D 62 65 72 20 32 30 31 32 20 2D 20 32 30 3A ember 2012 - 20:
000000A0 31 38 3A 33 37 20 53 65 74 3A 20 41 50 36 5F 77 18:37 Set: AP6_w
000000B0 69 6E 5F 78 38 36 5F 53 53 45 32 5F 4F 70 65 6E in_x86_SSE2_Open
000000C0 43 4C 5F 41 54 49 5F 72 31 33 31 36 20 2C 20 45 CL_ATI_r1316 , E
000000D0 78 65 3A 20 41 50 36 5F 77 69 6E 5F 78 38 36 5F xe: AP6_win_x86_
000000E0 53 53 45 32 5F 4F 70 65 6E 43 4C 5F 41 54 49 5F SSE2_OpenCL_ATI_
000000F0 72 31 33 31 36 2E 65 78 65 20 2C 20 50 61 72 61 r1316.exe , Para
00000100 6D 65 74 65 72 73 3A 20 20 2D 2D 64 65 76 69 63 meters: --devic
00000110 65 20 25 64 65 76 69 63 65 20 2D 75 6E 72 6F 6C e %device -unrol
00000120 6C 20 33 20 0D 0D 0A 32 35 20 53 65 70 74 65 6D l 3 ...25 Septem
00000130 62 65 72 20 32 30 31 32 20 2D 20 32 30 3A 33 31 ber 2012 - 20:31
00000140 3A 31 32 20 52 75 6E 74 69 6D 65 3A 20 44 65 76 :12 Runtime: Dev
00000150 69 63 65 3A 20 30 2C 20 63 6F 75 6E 74 3A 20 30 ice: 0, count: 0
00000160 2C 20 37 35 34 20 73 65 63 6F 6E 64 73 0D 0D 0A , 754 seconds...
00000170 32 35 20 53 65 70 74 65 6D 62 65 72 20 32 30 31 25 September 201
00000180 32 20 2D 20 32 30 3A 33 31 3A 31 32 20 44 65 76 2 - 20:31:12 Dev
00000190 69 63 65 3A 20 30 2C 20 43 6F 75 6E 74 3A 20 30 ice: 0, Count: 0
000001A0 2C 20 66 69 6E 69 73 68 65 64 2E 0D 0D 0A 2D 2D , finished....--
____________   - ALF - "Find out what you don't do well ..... then don't do it!" :) | |
| ID: 1289374 · | |
Message boards : Number crunching : Optimize your GPU. Find the value the easy way.
| Copyright © 2013 University of California |
