Announcing projects 5905 and 5906 (Core 14)

[ihaque]

Projects 5905 and 5906 just went up on advmethods. 5905 is an extension of project 5904 and 5906 is an extension of 5900. Since 5900/4's already been through the whole testing cycle and these projects are extensions of those, we'll be moving through the usual testing regime more quickly than normal.

Please note, the behavior of this core is different from what you may expect. The following behaviors are NORMAL, and the points credit takes it into consideration:

* Varying time per frame
* Low GPU temperatures
* Fluctuating CPU usage (somewhat higher than GPU core 11).


5905 WU's are worth 1888 points, with a preferred deadline of 3 days and final deadline of 6 days.
5906 WU's are worth 472 points, with a preferred deadline of 2 days and final deadline of 3 days.

Thanks for folding!

[dempaSD]

Thanks for these!! Can't wait to get a slew of them on my GPU's real soon... It is summer and I really really really don't want any more of those overheating 511 point WU's. Any chance project will run out of those soon? Please? I'll prefer 5900-5906 before any other proj thanks to preserving GPU hardware.

Thanks!

[extrasalty]

I hope that reduces the number of 511 pointers out there. I'm getting nothing but those lately. And not only their heat output sucks, they are THE slowest on my card.

[Kibbo]

I really like the heat characteristics when working on these WUs. My 285s stay around 70 degrees C with fans at 40%. When I first sat down at my main computer today I was convinced that the system had crashed from overheating or something because it was so quiet.

[ihaque]

Since no one's complained, I'm releasing these to the public.

[extrasalty]

I love these units (59xx), but they all seem to terminate prematurely with a driver crash and recovery. I know I'm on beta OS, with beta drivers, but other units are way more stable. For example this morning:

Code: Select all

[15:01:29] Completed 77%
[15:07:18] Completed 78%
[15:13:09] Completed 79%
[15:19:08] Completed 80%
[15:25:01] Completed 81%
[15:30:47] Completed 82%
[15:36:33] Completed 83%
[15:42:21] Completed 84%
[15:48:19] Completed 85%
[15:54:08] Completed 86%
[15:59:20] SEH code: 3221225477
[15:59:20] Run: exception thrown during GuardedRun
[15:59:20] Run: exception thrown in GuardedRun -- Gromacs cannot continue further.
[15:59:20] Going to send back what have done -- stepsTotalG=8000000
[15:59:20] Work fraction=0.8686 steps=8000000.
[15:59:24] logfile size=0 infoLength=0 edr=0 trr=23
[15:59:24] - Writing 641 bytes of core data to disk...
[15:59:24] Done: 129 -> 126 (compressed to 97.6 percent)
[15:59:24]   ... Done.
[15:59:24]
[15:59:24] Folding@home Core Shutdown: UNSTABLE_MACHINE
[15:59:32] CoreStatus = 7A (122)
[15:59:32] Sending work to server
[15:59:32] Project: 5905 (Run 8, Clone 41, Gen 0)

...All I did was open remote desktop connection, and when it started redrawing slowly and hard drive started working, I knew what has happened.
Just few minutes ago same story- it was working fine until I started an update in Linux VMware window:

Code: Select all

[02:23:41] Completed 21%
[02:29:39] Completed 22%
[02:33:13] SEH code: 3221225477
[02:33:13] Run: exception thrown during GuardedRun
[02:33:13] Run: exception thrown in GuardedRun -- Gromacs cannot continue further.
[02:33:13] Going to send back what have done -- stepsTotalG=8000000
[02:33:15] Work fraction=0.2259 steps=8000000.
[02:33:17] logfile size=0 infoLength=0 edr=0 trr=23
[02:33:17] - Writing 641 bytes of core data to disk...
[02:33:17] Done: 129 -> 126 (compressed to 97.6 percent)
[02:33:17]   ... Done.
[02:33:18]
[02:33:18] Folding@home Core Shutdown: UNSTABLE_MACHINE
[02:33:23] CoreStatus = 7A (122)
[02:33:23] Sending work to server
[02:33:23] Project: 5905 (Run 0, Clone 177, Gen 0)

A few other times it crashed during DXVA playback.
Beyond that, another peculiarity I noticed about the early 5900s is that realtime priority for the GPU client leads to a quick BSOD- this is something that I have tried when they came out and is not what I have currently.

[AZBrandon]

I love these units (59xx), but they all seem to terminate prematurely with a driver crash and recovery. I know I'm on beta OS, with beta drivers, but other units are way more stable.

Are you overclocking your card, or have a factory overclocked card? I made the same assumption as you - that the problem was the WU's. I backed off my card from an overclock to 1890mhz to the reference stock 1836mhz and haven't lost a WU since. If I recall the thread correctly from last month, there is additional error checking in the fah_core14 WU's which makes them better at identifying even very tiny GPU instability, so while you may not fail the fah_core11 based units, you might still be getting errors in them, but they're going undetected. The new units simply detect the errors better and thus will bail out more frequently if you're on the edge of stability or have a marginal video card that passed QA at the factory when it maybe shouldn't have.

[bruce]

I love these units (59xx), but they all seem to terminate prematurely with a driver crash and recovery. I know I'm on beta OS, with beta drivers, but other units are way more stable.

...All I did was open remote desktop connection, and when it started redrawing slowly and hard drive started working, I knew what has happened.

In Windows, there often is a connection between remote desktop connections and GPU crashes. I'm not sure if that applies to your beta OS with beta drivers with whatever you use for remote desktop connections, but consider the possibility that there's a relationship on your configuration, too.

[extrasalty]

In Windows, there often is a connection between remote desktop connections and GPU crashes. I'm not sure if that applies to your beta OS with beta drivers with whatever you use for remote desktop connections, but consider the possibility that there's a relationship on your configuration, too.

...there is a trap every step of the way. I should start using VNC more often.

[dempaSD]

I should start using VNC more often.

Or logmein.

[geokilla]

Maybe I should put -advmethods back up....

Even though this is a public release, I didn't get any 5905 or 5906 projects. Getting loads of 511 point WUs though.

[kiore]

I have had 4 of these so far without problems, yes they do jump the PPD all over the place (using FAHmon last 3 frames) Have my 9800GT's at 1800 on the shaders and the temps stay below 60C (with mega air cooling).
Nice change from the infamous 575x (511 pointers) faster cooler more stable on my cards.
kiore.

[tonic]

These WUs are great! Keep 'em coming.

[MichaelB]

Question for PG
Core 14 obviously returns good results and is much easier on the hardware temps. What can't it do that core 11 can? If it is the same I feel we should finish up all the Core 11 projects and move to Core 14 for all the new ones. Am I missing something?

[ihaque]

Core 11 and Core 14 do some of the same science, but there isn't a huge amount of overlap. They're not drop-in replacements for each other.