Message boards :
Number crunching :
AMD 11.6 seems to be Ok on 5XXX
Message board moderation
Author | Message |
---|---|
Send message Joined: 5 May 11 Posts: 233 Credit: 351,414,150 RAC: 0 |
Just posting an observation which may (or may not) help anyone thinking of switching to the 11.6 driver set. Seems fine, no particular speed up noted, not that there should be in the BOINC world according to the release notes. It went in first time to my 5850, but I had a hassle loading it up on the 5970s. With the latter it would not go in after a simple Control Panel delete of the old set, and ended up having to do a complete clean out with Driver Sweeper ( I did have the 11.5a & 11.5b hotfixes loaded, may have had something to do with that - dont know, sheer guess). After that, it went in first time. Been running overnight, no dramas, and crunch times finally settled back to my norm earlier this morning. Life goes on :) Unremarkable change with minor hassle. EDIT: Could be a different outcome with 6XXX cards, according to the release notes looks like lots of activity concerning 6XXX, so loading and outcome may well be different, worth reading release notes if you have a 6XXX. Regards Zy |
Send message Joined: 20 Apr 11 Posts: 388 Credit: 822,356,221 RAC: 0 |
Hi, Awesome, thanks for this info! I actually need to upgrade my own host with latest ATI driver to see if it helps with some crashes I've seen. So this makes me a bit less nervous on doing that. :) -w |
Send message Joined: 5 May 11 Posts: 233 Credit: 351,414,150 RAC: 0 |
The Devil may be in the detail - after the initial non-load on the twin 5970s, I got real picky, and let loose a registry cleaner at each significant point in the process after the initial attempt. That may have had an effect in reaching the successful outcome after the initial failures, or may have been a total waste of time, I have no idea - however, all went in at the end of the day .... The Registry Sweeper did pick up lots of so called "errors" at various points in the reloads, albeit lots were the usual "empty file" so called errors. It was zero "errors" at the start of it all prior to install attempts, but the resultant "errors" probably to be expected, as there were some failed Driver loads by that stage. In the great scheme of things, it was a non-event really, a minor hassle. Regards Zy |
Send message Joined: 2 May 11 Posts: 21 Credit: 173,527,396 RAC: 0 |
Hi, after the 11.4 fiasco i went back to 11.3 which works great here on my twin 5970's box so if theres no actual benifit speed wise and i'm already getting great reliability i think i'll stick with 11.3 lol best regards Ian ----> Please Join team Scotland HERE |
Send message Joined: 5 May 11 Posts: 233 Credit: 351,414,150 RAC: 0 |
Yup - if it aint broke, dont fix it :) Regards Zy |
Send message Joined: 6 May 11 Posts: 15 Credit: 692,725,672 RAC: 2 |
Hi, I could't help myself; had to go with this new 11.6 set. It had no gain for my HD6970 though. On different projects I think the 11.6 build is less stabile. I'll think it's time to roll back to the 11.4 preview I used before! |
Send message Joined: 18 May 11 Posts: 46 Credit: 1,254,302,893 RAC: 0 |
v10.12 works great on all my boxes (on Moo!, MilkyWay, Collatz, PrimeGrid). Any particular reason to upgrade? |
Send message Joined: 5 May 11 Posts: 233 Credit: 351,414,150 RAC: 0 |
No real reason in terms of the 11.X driver sets, cant compare the 10.X ones. There has been a lot of activity for 6XXX in the change from 11.5 to 11.6, no idea if its significant in BOINC terms for 6XXX. Otherwise an unremarkable upgrade from a BOINC perspective, its well into the "if it aint broke dont fix it" category :) EDIT: One caviat ..... there is a question mark for 4XXX users. Its looking like the OpenCL driver set (2.4) downloaded with 11.6 will prevent 4XXX users running any AMD OpenCL 2.4 apps. Still trying to get 100% fact, but previous rumblings from AMD indicated this would happen, and already some PG 4XXX users switching to 11.6 fell over on the PPS Sieve app, so its looking likely. Still has "rumour" status until confirmed, but if anyone is a 4XXX user, be wary until that aspect is verified or not. Regards Zy |
Send message Joined: 18 May 11 Posts: 46 Credit: 1,254,302,893 RAC: 0 |
Zydor thanks for the info. All my ATI cards are 4770, 5850 & 5870 so it looks like 10.12 is a good place to stay for now. It works well when running with mixed ATI / NVidia too. |
Send message Joined: 24 May 11 Posts: 7 Credit: 5,249,967 RAC: 0 |
Hello i've problem with my System. I've 2 5850 Cards all tasks run about 1h20min. I test different cataclyst and the new boinc 6.12.33 at the moment. Also with or without crossfire-nothing change? Can anybody help me? My other projects MW or Collatz run fine. Thanks a lot. |
Send message Joined: 5 May 11 Posts: 233 Credit: 351,414,150 RAC: 0 |
Hello i've problem with my System. That is on the slow side - you should be getting around 1200-1400 seconds with two 5850s. You were getting around 2600 seconds in May, which whilst still slow for the installation, is still way faster than current speed. What changed in your PC since May? Do you run any CPU Wus at the same time, if so, what are they and how many? Have you stopped the BOINC Client, and applications, then restarted it in the last few days? (that stop start will bring into effect a change to the Moo application making life a little easier) Regards Zy |
Send message Joined: 24 May 11 Posts: 7 Credit: 5,249,967 RAC: 0 |
Hello Zy On my harware i change nothing.I only update any staff on the softwareside like cataclyst and so. My main project is DNETC but that's offline so i change between MW or Collatz. Now i want growin up my creds on Moo but there is that prob. I always restart my comp. after each channge. Greetings Christian (Krypto2) |
Send message Joined: 5 May 11 Posts: 233 Credit: 351,414,150 RAC: 0 |
Do you run any CPU Wus at the same time, if so, what are they and how many? EDIT: (Look at my second machine to get an idea of 5850 times you can expect - bare in mind it only has one 5850, yours has two, so yours will be faster times - when we get it sorted ) Regards Zy |
Send message Joined: 24 May 11 Posts: 7 Credit: 5,249,967 RAC: 0 |
Yes i run 6 cores with Einstein@home at the mom. |
Send message Joined: 5 May 11 Posts: 233 Credit: 351,414,150 RAC: 0 |
Ok - drastic change, only so we easily see any effect, suspend all CPU WU applications, then come back here with some times when Moo apps have completed. When you have suspended the CPU Project(s), reboot the PC, and when its settled do an "Update" to Moo Project from the Machine BOINC Client. All that is to make real sure we only have Moo running, and Moo is synched and uptodate. GPUs need CPU support, the level of support varies from GPU Project to GPU Project, but the CPU capacity must be there, if its not, the GPU WU gets way slowed down. How much of a slow down will vary from Project to Project as it depends how efficient the CPU application is, and how much support the GPU application needs. What we are doing in the test above, is running Moo without any CPU Wus running, that will eliminate any CPU WU slowdown, and also give Moo a clear run so we can see better whats going on. Regards Zy |
Send message Joined: 24 May 11 Posts: 7 Credit: 5,249,967 RAC: 0 |
O.K. well we will see. Thanks a lot Zy. |
Send message Joined: 5 May 11 Posts: 233 Credit: 351,414,150 RAC: 0 |
One thing to note - I will be away from the PC from around 1600 UK time to around 1930 UK time - dont want you to think I've done a runner when you come back with the test times. I anticipate significantly improved times for WUs when run on their own, if there is not .... its thinking caps on as something real strange is happening, and we would have to dig deeper. If times drastically improve close to what are expected, start running CPU WUs one by one (ie run one first - see waht GPU time is, then run two at a time - see what GPU time is etc etc, until you find the "sweet spot"). I would anticipate with two 5850s needing one CPU free ..... maybe two, so end result would be running ok with 4 or 5 CPU WUs running. If the latter is the case, you will find similar performance increases at other Projects without 6 CPU WUs running (ie only 3,4,5 or 6 whatever it ends up at that Project(s) - not same speed increase, but some - certainly enough to take the hit of less CPU WUs running, the speed increase of the GPU app would more than make up for a minor loss in CPU WU credits. If I am not back, one of the other guys no doubt can pick this up to save you hanging on for a reply, whether or not there is a speed improvement in the test runs. I Should be back at around 1930 UK Time (GMT + 1) Regards Zy |
Send message Joined: 24 May 11 Posts: 7 Credit: 5,249,967 RAC: 0 |
O.K. Let us see tomorrow cause i also away from my Comp. from 1900CEST. I don't know when i'm back. Thank you very much again for the moment. |
Send message Joined: 2 May 11 Posts: 21 Credit: 173,527,396 RAC: 0 |
hi, when running twin cards setting the core value to 3 in your moo prefs helps heaps if you havent done it best regards Ian ----> Please Join team Scotland HERE |
Send message Joined: 25 Jun 11 Posts: 4 Credit: 100,365,728 RAC: 0 |
No real reason in terms of the 11.X driver sets, cant compare the 10.X ones. There has been a lot of activity for 6XXX in the change from 11.5 to 11.6, no idea if its significant in BOINC terms for 6XXX. 11.6 is running moo! fine on my hd 4850. amazon has a daily deal for me, a xfx hd6950 for under $200 I may just break down and order it. |