Message boards :
Number crunching :
Lunatics Windows Installer v0.44 - new release for v8 (required upgrade)
Message board moderation
Previous · 1 · 2 · 3 · 4 · 5 · 6 . . . 7 · Next
Author | Message |
---|---|
Ulrich Metzner Send message Joined: 3 Jul 02 Posts: 1256 Credit: 13,565,513 RAC: 13 |
MutliBeam ;) Aloha, Uli |
jedimstr Send message Joined: 23 Oct 00 Posts: 33 Credit: 16,828,887 RAC: 0 |
If I understand correctly, these apps are the same as the stock apps, and you expect to be releasing optimized apps later. My shift on the desk. Yes, that's right - but the transition back from Anonymous Platform to stock isn't as easy as in the opposite direction. Just to clarify this point further, does this mean if we are using the stock apps currently with modern GPUs/CPUs, there is no performance advantage going with the Lunatics install at this point in time? |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14679 Credit: 200,643,578 RAC: 874 |
Just to clarify this point further, does this mean if we are using the stock apps currently with modern GPUs/CPUs, there is no performance advantage going with the Lunatics install at this point in time? Yes for GPUs No for CPUs. The installer does contain optimised CPU applications which have not been deployed as stock. The newer your CPU, the more likely it is to benefit: if your CPU/OS can run AVX - as any Intel Core i3/i5/i7 of third generation or later, running Windows 7 or later - should be able be able to do, you'll get the most improvement. |
BilBg Send message Joined: 27 May 07 Posts: 3720 Credit: 9,385,827 RAC: 0 |
Just to clarify this point further, does this mean if we are using the stock apps currently with modern GPUs/CPUs, there is no performance advantage going with the Lunatics install at this point in time? Maybe there is "advantage going with the Lunatics" even for GPUs - Apps may be the same but the server may not choose the best variant (or take too much time to settle on "the best") - There are no Settings for stock to select exactly what you want to run on every device, e.g. only SETI@home v8 on CPU + only AstroPulse on ATI AMD + only SETI@home v8 CUDA 5 on NVIDIA Â - ALF - "Find out what you don't do well ..... then don't do it!" :) Â |
TimeLord04 Send message Joined: 9 Mar 06 Posts: 21140 Credit: 33,933,039 RAC: 23 |
Just reporting in... Got two AP v7 WUs now running under Lunatics 0.44. Will monitor, and report Success/Failure... I anticipate that they'll run fine, though. :-) (They're new, NOT re-sends. v7 7.10) TL TimeLord04 Have TARDIS, will travel... Come along K-9! Join Calm Chaos |
Jord Send message Joined: 9 Jun 99 Posts: 15184 Credit: 4,362,181 RAC: 3 |
I have to say, keeping an eye on the times in BOINC Manager, I'd run over 100 tasks already with Mike's r3299 for my ATI HD7870 GPU. Those ran just the 'generic' ati_opencl_sah, whereas r3300 from the Lunatics installer runs opencl_ati5_sah. The latter runs them slower. Per task by about 30 seconds to over 2 minutes. e.g. some *11aa* tasks ran for 17m 54s on r3299, but run in 18m 22s on r3300. Some *11ac* ran in 6m 50s on r3299, but run in 9m 09s on r3300. The load and temperatures are about the same, I haven't seen them go over 60C yet, be it r3299 or r330. Better than Fallout 4 with its 67C and high-fan-racket. :) |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14679 Credit: 200,643,578 RAC: 874 |
I have to say, keeping an eye on the times in BOINC Manager, I'd run over 100 tasks already with Mike's r3299 for my ATI HD7870 GPU. Those ran just the 'generic' ati_opencl_sah, whereas r3300 from the Lunatics installer runs opencl_ati5_sah. No, not quite right. 'ati_opencl_sah' and 'opencl_ati5_sah' don't run anything at all - they are simply plan_class labels. Eric hasn't deployed any ati_ plan classes this time - they were mainly to fool v6 BOINC clients (which didn't have native OpenCL scheduling) into running OpenCL as if they were CAL applications. It looks like we've finally thrown away that ancient kludge. The slowdown between r3299 and r3330 (those numbers are significant) is probably because of the increased arithmetic precision of the final released applications - see if your rate of inconclusive validations goes down. |
Mike Send message Joined: 17 Feb 01 Posts: 34376 Credit: 79,922,639 RAC: 80 |
This depends on the GPU in use Jord. On my 380 3330 is faster. Not to forget some important bugs were fixed. With each crime and every kindness we birth our future. |
Jord Send message Joined: 9 Jun 99 Posts: 15184 Credit: 4,362,181 RAC: 3 |
Richard wrote: No, not quite right. 'ati_opencl_sah' and 'opencl_ati5_sah' don't run anything at all - they are simply plan_class labels. And because I put that into my app_info.xml file, the old one had <plan_class>ati_opencl_sah</plan_class> in it, the new one has all possible ones, even one my GPU cannot run: <plan_class>opencl_ati_nocal</plan_class> (Those are, if all is well, only for the AMD GPUs that have no hardware support for, and therefore no detection of CAL. My HD7870 still does have CAL support and does the detection, so in essence it's useless that it's included in my app_info.xml file, only there to confuse. I know my GPU can run these tasks with that label, as they're just generic tasks with no extra bits and bobs, just saying that it shouldn't be included in an app_info.xml file of a GPU that's perfectly capable of doing the detection the plan_class says it cannot do. If we're to confuse anyway... But perhaps that a separate application is needed here, the _nocal_ one, that people with newer R5s, R7s, R9s can choose out of (not the rebranded HD5xxx/69xx/78xx/79xx). Richard wrote: The slowdown between r3299 and r3330 (those numbers are significant) is probably because of the increased arithmetic precision of the final released applications - see if your rate of inconclusive validations goes down. Sorry I missed out on the extra 3 in the name. But then I didn't misspell Multibeam and v7/v8... :<P>> My rate of inconclusives though? The one I have is due to this guy who has 119 of them and seems to make a sport of having them. I don't think my system has ever thrown out any inconclusive results, unless I was tinkering with it at a time I shouldn't have and something went wrong with a key file. And that must've been in 1985. So. Mike wrote: On my 380 3330 is faster. Edit: err oops... there was something I wanted to say but completely forgot. Why did I quote you, Mike? Tell me? :) |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14679 Credit: 200,643,578 RAC: 874 |
All those plan_class names are included in the installer's AI stubs because at some time in the next six months: Some guy might join the SETI@Home project using an old (or new, or middling) ATi (or AMD) GPU, and get the scheduler's usual mish-mash of task types. And find the unexpected work/thermal load burns up that GPU. And go out to buy a replacement. And then find these message boards. And immediately get told that they "should" be running Lunatics (it happens, but not from me). And run the installer. And wondrously, all that mish-mash of strange stock tasks becomes runnable again. But I'm not including the full gamut of Beta and archival plan_classes and version numbers as well. |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
There were NO such thing in OpenCL app. Precision exactly the same from initial v8 introduction. All precision issues came from stock Linux x64 build and were corrected. OpenCL was with "gold standard" ;) So any such performance differencies quite subjective and could fall only into instruction layout differencies between binaries. |
Jimbocous Send message Joined: 1 Apr 13 Posts: 1856 Credit: 268,616,081 RAC: 1,349 |
Very small visual 7ypo (which don't really need correction) on ATI and Intel pages: Mutlibeam:) edit: late to the show again ...sigh |
Jim_S Send message Joined: 23 Feb 00 Posts: 4705 Credit: 64,560,357 RAC: 31 |
Very small visual 7ypo (which don't really need correction) on ATI and Intel pages: Yup. ;o) I Desire Peace and Justice, Jim Scott (Mod-Ret.) |
Jimbocous Send message Joined: 1 Apr 13 Posts: 1856 Credit: 268,616,081 RAC: 1,349 |
Mutlibeam:) Better than Motleybeam, for sure ... |
Jeff Buck Send message Joined: 11 Feb 00 Posts: 1441 Credit: 148,764,870 RAC: 0 |
Actually.....we're all late to the party. Here's the same window from v0.43b. I guess none of us really pay attention to details when we're in a hurry to install the latest and greatest! ;^) |
HAL9000 Send message Joined: 11 Sep 99 Posts: 6534 Credit: 196,805,888 RAC: 57 |
Actually.....we're all late to the party. Here's the same window from v0.43b. Well you are not only late to the party, but you also missed the main error in the 0.44 installer. You guys might want to give Richard's reply to BilBg's post a read. SETI@home classic workunits: 93,865 CPU time: 863,447 hours Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[ |
Jim_S Send message Joined: 23 Feb 00 Posts: 4705 Credit: 64,560,357 RAC: 31 |
Ummmmm...plan_class labels. I Desire Peace and Justice, Jim Scott (Mod-Ret.) |
janneseti Send message Joined: 14 Oct 09 Posts: 14106 Credit: 655,366 RAC: 0 |
I have installed this new release and everything is working fine. But now I cant see any of the v8 workunits in the SETI webpage despite I crunched and reported several. Only v7 workunits are showing. Any ideas? |
Cruncher-American Send message Joined: 25 Mar 02 Posts: 1513 Credit: 370,893,186 RAC: 340 |
I have installed this new release and everything is working fine. Problem is that the replica dB is currently offline. That is where the WU info comes from. |
janneseti Send message Joined: 14 Oct 09 Posts: 14106 Credit: 655,366 RAC: 0 |
I see. Thanks. |
©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.