Project 5902 - Deadline too short

[Michael H.W. Weber]

If the final submission deadline for project 5902 could be extended by only one day (currently set to 5 days; required would be 6 days at maximum), it would be possible to participate with notebooks having the Geforce 9200M GS graphics card installed. Else I would have to take this machine away from supporting FAH by GPU.

Michael.

[AZBrandon]

Can you overclock the shaders on the integrated graphics? It looks like stock clock is 1300mhz shader speed, which is way below what many of the chips in this series run. If you could bump it up to around what many 9800GT's run, say, 1566mhz that may only produce a little more heat (since it only has 8 shader cores to begin with compared to a 9800GT's 112) and produce enough speed to meet the deadline.

[Flathead74]

In cases like this, extending deadlines won't help anything.

If one cannot return a work unit before the preferred deadline, three days for the p5902,
then the work unit is reassigned to someone else to process.

This causes duplication of work and wasted resources,
and also causes the project to be slowed down while awaiting the WUs to be completed. <-- sad, not mad...

Even if you could overclock the shaders enough so that you meet the final deadlines,
the work is going to be reassigned because the preferred deadline has passed.

Perhaps that particular system could be put to better use by running one of the other clients.

[PeterA]

I'm currently running a 5903 that's estimated to be completed 8 hours after the 5 day deadline. I have a GeForece 8400 GS and don't know how to speed it up. I frustrating knowing that I won't get any credit for it.

[bruce]

With a GPU that slow, you'll probably get more credit simply by running a CPU client (one per CPU core) instead of the GPU client.

[Michael H.W. Weber]

Overclocking of my laptop GPU is not an option...

If one cannot return a work unit before the preferred deadline, three days for the p5902,
then the work unit is reassigned to someone else to process.

This causes duplication of work and wasted resources...

Well, to my knowledge duplicated WU processing is part of almost all distributed computing projects, anyway.

Perhaps that particular system could be put to better use by running one of the other clients.

That particular system does indeed run two FAH CPU clients in parallel to the GPU client and performs quite well (2,4 GHz Intel dual core). In fact, for me it is no problem taking off the FAH GPU client from this machine. Note that it was intended to support the FAH project at my private electricity cost which I can of course save for something else. Please also note that around 30% of the computers sold nowadays are laptops. Hence, setting deadlines too short most certainly will result in significant reduction of potential support for the FAH project.

I'm currently running a 5903 that's estimated to be completed 8 hours after the 5 day deadline.

That's the issue I was trying to put forward here: The deadlines of these WUs are only missed by a few hours, hence I think extending it could bring some additional support to the FAH project (if desired).

With a GPU that slow, you'll probably get more credit simply by running a CPU client (one per CPU core) instead of the GPU client.

Well, to my experience comparing single clients on a laptop there is no client giving more PPD than a GPU client. Two CPU clients taken together (one per core), however, indeed should give more credits than a single GPU client on that same machine. But most effective FAH support could be achieved with two CPU clients plus a single GPU client - all run in parallel, of course - and that was exactly what I was doing until I recently started to get these 5903 WUs...

Michael.

[X1900AIW]

Running a new 8300 mGPU was just a small contribution with one workunit in ~4 days (no 24/7 system), posted the PPD results here, when the new project started my mini cruncher was locked out because of exactly these missing hours it can´t finish the workunit.

I know that there are saver options to reach the deadline, like using faster GPUs I did before but none of these cards shows a better efficiency in PPD/W. Perhaps some day the servers can assign an adequate task to GPUs with less shaders (perhaps P5900 at the moment), it´s not a question to increase client TFLOPs but to increase the number of participating members and in this manner increase attention to the project. In fact F@H may not be dependent on more members ?

There are and will be enough members with no problem at all concerning the deadline (they are discussing PPD drops lost in minutes - PPDs we would be happy to get in days ), though every CUDA enabled card could be able to meet the requirements "to help science" except this minor point called deadline.

[Flathead74]

Michael H.W. Weber, good luck arguing for more time.
I tried it in the past, to no avail, and ended up disabling dual-Xeon systems instead.
I am not the enemy.

However, I do not believe that it is Pande Group's intention to process every work unit two times.

...Perhaps some day the servers can assign an adequate task to GPUs...

What a great idea, not only for GPUs, but for any processor that PG deems capable.

[PeterA]

As I feared:

Code: Select all

[14:03:54] Completed 91%
[15:27:36] Completed 92%
[15:27:36] Unit 7's deadline (April 5 14:47) has passed.
[15:27:36] Going to interrupt core and move on to next unit...
[15:27:37] Got kill signal -- issuing INTERRUPTED core shutdown
[15:27:37] 
[15:27:37] Folding@home Core Shutdown: INTERRUPTED
[15:27:39] CoreStatus = 66 (102)
[15:27:43] - Preparing to get new work unit...

Almost made it. It's now running a 5780. Same deadline, but at least it processes faster.

[MtM]

Overclocking of my laptop GPU is not an option...

If one cannot return a work unit before the preferred deadline, three days for the p5902,
then the work unit is reassigned to someone else to process.

This causes duplication of work and wasted resources...

Well, to my knowledge duplicated WU processing is part of almost all distributed computing projects, anyway.

Believe me, it's not the intention with f@h and you're infact introducing delay and adding nothing by running a client like that.

As I feared:

Code: Select all

[14:03:54] Completed 91%
[15:27:36] Completed 92%
[15:27:36] Unit 7's deadline (April 5 14:47) has passed.
[15:27:36] Going to interrupt core and move on to next unit...
[15:27:37] Got kill signal -- issuing INTERRUPTED core shutdown
[15:27:37] 
[15:27:37] Folding@home Core Shutdown: INTERRUPTED
[15:27:39] CoreStatus = 66 (102)
[15:27:43] - Preparing to get new work unit...

Almost made it. It's now running a 5780. Same deadline, but at least it processes faster.

The same goes for you. If you're taking on a wu, the server waits untill you either submit it or untill the deadline has passed before it get's submitted to another folder. You're introducing delays, and adding nothing.

[7im]

FAH does NOT send out multiple copies of the same work unit for verification like SETI and other projects do. FAH is the exception.

[ihaque]

Sorry, I updated the stats credit to compensate for the speed drop with 1.25, but neglected to do the same with the deadline time. We can tolerate an extra day on these projects; I've increased the deadline to 6 days. Given the log you posted, that should be more than enough.

[bruce]

Overclocking of my laptop GPU is not an option...

If one cannot return a work unit before the preferred deadline, three days for the p5902,
then the work unit is reassigned to someone else to process.

This causes duplication of work and wasted resources...

Well, to my knowledge duplicated WU processing is part of almost all distributed computing projects, anyway.

Duplicated WU processing is an inherent part of Seti@home and other such projects, but I believe you are over-generalizing when you say "almost all DC projects". FAH is specifically designed to eliminate as much of that wasted processing as is reasonably possible, which is why we expect people to complete all WUs by the Preferred Deadline.

[Michael H.W. Weber]

Duplicated WU processing is an inherent part of Seti@home and other such projects, but I believe you are over-generalizing when you say "almost all DC projects".

As I said, most DC projects do compute WUs redundantly although to my point of view it would of course be sufficient to re-compute only those that hold the (apparently) good results.

FAH is specifically designed to eliminate as much of that wasted processing as is reasonably possible, which is why we expect people to complete all WUs by the Preferred Deadline.

Well, it is good to know that concerning this issue, FAH once again seems superior to other projects.

Michael.

[bruce]

Duplicated WU processing is an inherent part of Seti@home and other such projects, but I believe you are over-generalizing when you say "almost all DC projects".

As I said, most DC projects do compute WUs redundantly although to my point of view it would of course be sufficient to re-compute only those that hold the (apparently) good results.

FAH accepts the good results when submitted and moves on to the next Gen of that trajectory.

FAH only reassigns those WUs which expire or which fail. You'd be surprised how many WUs fail on some (probably) overclocked machine and then complete successfully when reassigned. The real challenge is to figure out how to keep those excessively overclocked machines from pulling WUs off the server and ruining things for everybody.

NOTE: There's no way to know if the machines that turn in bad results are actually excessively overclocked or if they have some other sort of hardware problem, but it's a convenient way of speaking about them.