...
any thoughts from GPUGRID side on my suggestion above ?
Good idea - bad idea ?
Feasible - not feasible ?

still no reaction whatsoever from the GPUGRID team - why so?

is my idea/suggestion that bad that it's not even worth being considered to be implemented?

the Pascals dont's work with the NVIDIA Inspector

Wrong, worked just fine with my GTX1060 and nwo with my GTX1070.

is my idea/suggestion that bad that it's not even worth being considered to be implemented?

Well.. the current project team is not the most active in the forums, so I suspect they didn't even notice your request. Personally I agree that it's strange to alost not use the short runs queue at all and have long runs which differ by more than a factor of 2 in the runtime. That's clearly two different WU categories, so if they don't want to introduce a 3rd tier they coulöd at least use the 2 which they already have.

MrS

---

Scanning for our furry friends since Jan 2002

the Pascals dont's work with the NVIDIA Inspector

Wrong, worked just fine with my GTX1060 and nwo with my GTX1070.

hm, this sounds interesting. A few months ago, I read somewhere here in the forum that the NVIDIA Inspector does not work with Pascals.
Maybe that guy had an older version in use - which version is your's?

is my idea/suggestion that bad that it's not even worth being considered to be implemented?

Well.. the current project team is not the most active in the forums, so I suspect they didn't even notice your request.

This is the problem with this project, in general.
As I already said, with a project of this magnitude, a certain amount of "customer care" is a vital component. They need to receive and to understand the feedback from the crunchers. If this doesn't happen, much will get worse.

Personally I agree that it's strange to alost not use the short runs queue at all and have long runs which differ by more than a factor of 2 in the runtime. That's clearly two different WU categories, so if they don't want to introduce a 3rd tier they could at least use the 2 which they already have.

For some of the recent tasks, like the "ADRIA_FOLDGREED90_crystal_ss_contacts_100_ubiquitin", a GTX750Ti under Windows10 is clearly overtaxed. It can take up to 3 days to get such a monstreous task finished (with several stops inbetween, caused by one if the bugs of the current crunching software).
So, it would definitely be important to implement a 3rd tier for such types of task.
And I am pretty sure that many crunchers are still using one or more GTX750Ti.

I'm using NV Inspector 1.9.7.8. On afterthought: amybe he meant a certain feature was not working, or not working as intended. Setting the CUDA memory clock for Maxwell and Pascal is a bit weird.

BTW: on topic.. has there been any change notification from 9.15 to 9.18? Maybe they fixed some bug or added a feature, which resulted in more work being needed?

MrS

---

Scanning for our furry friends since Jan 2002

BTW: on topic.. has there been any change notification from 9.15 to 9.18? Maybe they fixed some bug or added a feature, which resulted in more work being needed?

I am really out of my depths here, and I apologize to Richard Haselgrove if I am referencing him out of context. But there is an interesting discussion on the BOINC forum between him and boboviz (among others) that touches on some of these issues.
http://boinc.berkeley.edu/dev/forum_thread.php?id=11647&postid=79043#79043

I think improvement in all areas would not hurt, but getting it done with the present limitations on both BOINC and GPUGrid (not to mention the Nvidia drivers) is another matter.

But there is an interesting discussion on the BOINC forum between him and boboviz (among others) that touches on some of these issues.
http://boinc.berkeley.edu/dev/forum_thread.php?id=11647&postid=79043#79043

I just read an interesting statement in the discussion there:

"gpugrid admins don't listen their volunteers. :-P"

which for me is best proof that I am not the only one who feels this.

No, I don't read it that way at all. They don't have the people (particularly with the BOINC expertise) to fix it. But on the other hand, why not go back to the old app for Windows? Maybe that would provide more output, or maybe just more confusion at this point.

They don't have the people (particularly with the BOINC expertise) to fix it.

I guess it's not a matter of BOINC.
Fact is, that the new crunching software 918.80 was put together in a hurry and obviously not testet well enough; hence, it's buggy.

So, in order to eliminate the bugs, it would just need to re-work the software including thorough testing, and then releasing it.

which for me is best proof that I am not the only one who feels this.

No doubt about that. However, I see it this way: they're mainly scientist interested in doing good science. The number crunching is an essential part of it, but it's neither the only nor the most important one. For more important is what simulations they run and to analyse what the results actually mean. Apparently they're quite busy doing just that, as the project ran out of work several times in the last months.

I understand this as "We're limited by manpower to setup and process simulations and results. Having even more GPUs available or making them a few % faster wouldn't help us much."

Mind you, I'm not talking officially for the project and don't endorse an almost complete lack of communication. But it's something I keep in mind and lets me lay back relaxed, thinking that overall we're doing quite well (giving them a bit more power than they can handle). If they would have to wait for results we may see them more often in the forums, chatting and pushing for performance improvements and new users.

MrS

---

Scanning for our furry friends since Jan 2002

they're mainly scientist interested in doing good science.

yes, this is perfectly right. I am not saying at all that it's the job of the the scientists to take care of the infrastructure. This needs to be done by someone specialized in this. However, as it seems, they person in charge of this is available only very rarely, if at all.

an almost complete lack of communication

for sure rather frustrating for many crunchers, believe me.
I am reading in the forum quite much, and too often I see people bringing up the same kind of problem or question, over and over. However, no reaction whatsoever.

Having said that, I can only repeat what one of the admins from another project under BOINC has published recently:
"Of course have happy volunteers is very important for the health of a project so it is something that should be addressed"
So let's hope that at some time the GPUGRID people will realize this.