New Assignment Server feedback/problem

[husq]

My GTX780 770 570 480 GPU, with or without parameters, all received 762X or 8018 WU

In any case don't get CORE17 WU

[PS3EdOlkkola]

@DutchForce: You are correct that I'm not using the most recent version of FAHClient (all systems currently running v7.2.9). My very hard-learned philosophy on software upgrades is, "If it ain't broke, don't upgrade, and if an upgrade is necessary, plan to spend 3x the estimated time to account for Murphy and his minions". In the intervening 18 hours since my last post, all my systems have recovered on their own and are now receiving work units, albeit all Core 15/16. I was tempted to upgrade all systems to the most recent FAHClient, but unless v7.2.9 has other issues necessitating an upgrade, I plan to stay on the current version. Consistency of driver software, utilities, operating systems and FAH across 11 systems 100% dedicated to FAH is the only way I can keep my sanity and not have maintenance issues dominate every spare moment of free time.

@bruce: I jumped the gun assuming it was the new AS. You are absolutely correct that performing a full root-cause diagnostic is the appropriate path to take before taking any drastic action. I should have recognized the inconsistency of recommending a roll-back knowing that doing so violates my own software upgrade/change philosophy. My apologies for the hypocrisy; I generally know not to let my emotions get the better of me, but kudos to you for clarifying the appropriate response. I take the smooth and consistent operation of all my systems very seriously, so when I believe all is OK on my end but my systems are not running properly, it causes a lot of consternation on my part.

@mods: I made an error recommending an outsource testing organization. It was not my intention to promote their services, but simply to provide Joe Coffland and his team some help knowing that they do good work. However, I still stand by my offer and if Joe would like to contact me through a private email on the Forum, I'll gladly provide whatever help, advice and financial support I can to assist him. I guess this morning is one full of apologies, and I owe you one as well. I should have know better not to promote a specific company here.

Gentlemen, thank you for your honest feedback.

[-alias-]

My only GPU server populated with 4 X GTX660Ti has been folding only core15 projects over the last 2-3 days so when it's finish the current projects I will close it down and not start it again before core15 is confirmed finally / gone / dead and buried . The server uses 560W and produces only 100K PPD when core15 is the only project it gets, and I have no use for the heat where it stands either, so my choice is easy. When it folds core17 projects it gives around 270k PPD, which is acceptable for me.

[Kjetil]

I did the same. Only running one win and one linux. One 980 and 750Ti and linux is running smp. Have 5 780Ti, 3 750Ti and 3 980+ i7 3770K@4200 on smp.

[7im]

There is nothing terribly broken with v7.2.9. But v7.4.4 added the ability to contact a second AS for GPU assignments, among several other fixes and improvements. Having that second option certainly expands the range of work units available.

[JimF]

So we're stuck with high-end GPUs running low-value Core15's until Joe finds the bug(s)…

I wouldn't equate low scores with low value. The point system for Core_15 just did not include the Quick Return Bonus, so their scientific value could be just as great as the Core_17s. On the other hand, they don't run the Core_15s very often, usually when they have nothing else to hand out, so I think we can infer from that that they aren't the highest-priority projects in the world. I will keep at least one Nvidia machine active for the duration, but might find something else for the others to do.

[Sailer]

I've been running into this same problem for over a week. To say the least, its the pits. I have a four GTX780 Ti cards and a Titan Black and all they get are core 15s. My PPD has dropped to less than half of what it used to be, with a total PPD dropping from the area of 980,000 down to around 400,000 and that was after adding an older 6th computer with a GTX560 Ti. While I know that its all for science, I might as well be running GTX560 Ti cards on the computers. Yes, I do run v7.4.4 on all the computers along with 340.52 drivers. Everything worked fine until about a week and a half ago. Is there any guesses as to when the core 17s will start getting assigned again? Right now I'm putting off buying a GTX980 until the core 17s start running again.

[Gary480six]

I hope that the F@H team is looking for feedback with the new AS code.

As reported here Dr. Pande said

With the newer NVIDIA drivers, it looks like Core17 works well on Maxwell. We’ve released Core17 to Maxwell on adv (“Advanced Methods” setting). If you’re having problems, you can set from adv back to the regular fah setting, allowing donors to opt out if they’re having problems. The latest drivers are recommended.

I have three different systems with GTX 750Ti (Maxwell) video cards, the 7.4.4 client, Windows 7, and the Nvidia 340.52 drivers.

Within the last 24 hours, all systems were crunching core 18 work. (YAY!!)

As per Dr Pande comments, I added the client-type advanced 'switch' to the GPU slot.

I then checked the log file to see that the line <client-type v='advanced'/> was shown.


When the core 18 work finished, I hoped to either get a new core 17 work unit or at the very least, another core 18 work unit.

In two cases, I was issued core 15 work and one got core 18 - none got core 17 work.

Anybody else applied the advanced 'switch' to a GTX 750Ti Folding system and then started getting core 17 work?

Or was this advanced 'switch' intended only for the new 900 series Maxwell cards?

I am aware that it could just be the random distribution of the Assignment Server... but before the new code was installed, my GTX 750Ti cards were getting almost exclusively core 17 work - for about 30 days. (and without the advanced 'switch')




I'm not sure where to leave this kind of feedback, so please move this if it would be more useful in a different subject group.

[VijayPande]

I'll forward this thread to Joseph so he can see that the issue still isn't resolved. For now, I strongly suggest you use the adv flag and a v7 client if you're not already.

My guess is that this is definitely an issue w/the new AS, since we have a ton of Core17 and Core18 WUs. This is issue #1 for us to address and we'll post an update when we know more.

[VijayPande]

Since it's the weekend, most of the staff is out, so I've made some changes myself just now to the AS. If that works, then we know the issue was an AS configuration issue (which I think I caught). If it doesn't work, then it may be an AS code bug that Joe will need to further investigate. Thanks for your understanding with this. Switching over to the new AS has had more growing pains than we expected.

[VijayPande]

PS It will take probably a few hours for us to know if this is working (enough time for enough new assignments to come in). I'll check back in an hour or two to see where we are.

[Breach]

I can confirm that I am now getting Core 17 WUs on both advanced and beta. Unfortunately, I have numerous failures (project 9406). Same with both core versions - 0x17 0.0.52 (advanced) and 0.0.55 (beta). [Edit: Worst of all it also affects affects client-type=FAH (Project 13001). I guess there was a reason that Maxwell was blacklisted on Core 17 and that reason is back]

Same with some core 18 WUs: 10473, 10472 (beta) (some project 1xxxx WUs fold find though)

18:38:45:WU02:FS01:0x17:ERROR:exception: Force RMSE error of 606.591 with threshold of 5

That's 970 GTX (Maxwell) with the latest nVidia drivers - 344.16.

At least Kepler owners will be happy...

[bruce]

I have three different systems with GTX 750Ti (Maxwell) video cards, the 7.4.4 client, Windows 7, and the Nvidia 340.52 drivers.

Upgrade to the latest drivers

(I've added that to the announcement.)

[Kjetil]

The last driver is 344.11. I have 3 750Ti and 3 980. All gets bad wu. PS 980/970 can use 344.16, but 750Ti can not.

[kookykrazee]

Is there a known issue with core 9406. Have gotten 4 so far, first 2 failed within 30m of starting and the other 2 are attempting to complete for about 30m. I read in another thread, there had been previous problems with 9406. I am using most recent version of F@H and have latest drivers for my 3x 750ti cards.