Error Using PrimeNet Get Manual GPU Assignments
 

In the past I've been using Primenet's "Get Manual Assignments" page for trial factoring work that I've thrown at my GPU. I've never had a problem with getting assignments in this way or returning results. Refer [url]http://www.mersenne.org/manual_assignment/[/url]

In these forums I recently stumbled on an alternate assignment page for assignments, targeted at GPU work: [url]http://www.mersenne.org/manual_gpu_assignment/[/url]. I've attempted to use this assignment page, trying several permutations of filling out fields and submitting a request but I keep getting the following error:

Error code: 40
Error text: No assignment available for GPU trial factoring, cpu_id: 15100, user_id = 65

I'm certain that I'm logged in. Can anyone offer any suggestions on what I can do to fix the problem or should I not be using this method of getting assignments?

All of those Assignments are currently held by GPU72 for getting assigned there. This 'official' GPU TF page is for eventually replacing the extern assignment page, but there was no effort yet to fully implement this. If you want to get better assignments for your GPU, head over to GPU72.com.

I'd prefer to choose my assignments direct and not through a proxy/3rd party and what they believe the better assignments are. I misunderstood the purpose of the GPU page which I thought it was available for all that had gpu's and wished to use them for TF'ing for the project and not just for those going through gpu72. Thanks for clearing that up, I'll stick with the standard assignment page.

you can "choose your assignments direct and not through a proxy/3rd party and what they believe the better assignments are" at gpu72 assignment page too. and you report the results directly on primenet, under your name, which will be credited with finding the new prime. especially if you find the new prime by TF on a gpu...

[QUOTE=tmorrow;434436]
I'm certain that I'm logged in. Can anyone offer any suggestions on what I can do to fix the problem or should I not be using this method of getting assignments?[/QUOTE]

Try higher exponents, (i.e. up in the 900M->999M range) and that manual gpu page should work. Of course, that won't help eliminate soon-to-be-checked candidates, but it might help in a decade or so.
Alternatively move to a higher bit level. Have a look at [url]http://www.mersenne.ca/status/[/url] for where things are at the moment.

[QUOTE=LaurV;434457]you can "choose your assignments direct and not through a proxy/3rd party and what they believe the better assignments are" at gpu72 assignment page too. and you report the results directly on primenet, under your name, which will be credited with finding the new prime. especially if you find the new prime by TF on a gpu...[/QUOTE]

Thanks for the info. I had a look sometime back at the GPU72 page. It seemed to me that I have to sign up with an additional username, password etc before I can get assignments, thus introducing a layer of irritation into the works. Furthermore it would seem that since they don't ask for my primenet username that any assignments they give me must be done with a level of indirection, i.e. they must be getting assignments from primenet under a user and then that users assigments are reallocated to me. This does not appear to be direct to me.

I would prefer to simply authenticate to primenet and say "have gpu, can therefore do more factoring than a cpu in the same time, hit we me with some exponent choices and a higher bit level options". I appreciate now that this still can't be done. I guess GPU72 have set things up to be as helpful and efficient as possible with handing out GPU work to those interested. I'll give them another look after my next round of work has been completed.

You have to fiddle with both the work type, the range, and the pledge level, to make the manual_gpu_assignment page work.

[QUOTE=Mark Rose;434469]You have to fiddle with both the work type, the range, and the pledge level, to make the manual_gpu_assignment page work.[/QUOTE]

Yeah, I can't be of any help with that page unfortunately. Since that page isn't officially supported (well, it's not linked from anywhere), I've never given it much thought and don't really know how it all works.

Thanks Madpoo, all is fine from my perspective. I can use the official page to get assignments manually (and in bulk) for gpu work or I can create an account with gpu72 and possible get some choice on work that they've obtained from primenet in some manner.

As a future feature request though, the project should seriously consider how to best take advantage gpu's on client machines. Perhaps this discussion has already taken place and I just missed it. Perhaps gpu72 is just the first step in the process. Just the same I expect there are a lot of idle, untapped gpu's out there that could be put to good use by participants that aren't up to trolling through these forum threads to find out how to do it. If the prime95 client could be extended to gpu detection and assigning suitable work for the gpu, it would be a boost to the project. And I do appreciate the difficulty involved.

[QUOTE=tmorrow;434528]Thanks Madpoo, all is fine from my perspective. I can use the official page to get assignments manually (and in bulk) for gpu work or I can create an account with gpu72 and possible get some choice on work that they've obtained from primenet in some manner.

As a future feature request though, the project should seriously consider how to best take advantage gpu's on client machines. Perhaps this discussion has already taken place and I just missed it. Perhaps gpu72 is just the first step in the process. Just the same I expect there are a lot of idle, untapped gpu's out there that could be put to good use by participants that aren't up to trolling through these forum threads to find out how to do it. If the prime95 client could be extended to gpu detection and assigning suitable work for the gpu, it would be a boost to the project. And I do appreciate the difficulty involved.[/QUOTE]

In theory, the authors of the GPU programs (CudaLucas, mfakt*, etc) could use the published Primenet API to get assignments and return results but it would be up to them to implement. I'm sure George, Scott or even myself (to the limit that I understand it) would be able to assist.

Right now there's also Misfit, besides GPU72, but again we're getting into an area I don't know much about since I don't have a fancy GPU or anything. Maybe I really should consider getting an older 580 or something so I can tinker with GPU based TF and LL and then maybe I'd see the pain points and do something about it. :smile:

It works. You MUST enter the optional exponent range (and probably the optional bit level). They seem to be no longer optional.

NUMBER OF ASSIGNMENTS default 10
WORK RANGE What makes sense
_OPTIONAL_ EXPONENT RANGE 75000000 to 100000000
WORK PREFERENCE - lowest exponents
_OPTIONAL_ BIT LEVEL - 75

Apparently the script defaults to the active ranges at the time the script was created (and it presumably worked without this range being entered in the past), but now ALL DCTF has been COMPLETED, and the active range for LLTF is about 82M. So the Optional Exponent Range is no longer optional and the optional bit level may very well not be optional anymore.

These figures are no longer accurate
The currently active range for double-check TF is 32000000 to 45794298.
The currently active range for first-time LL TF is 52000000 to 77505668.
The currently active range for 100 million digit TF is 332192809 to 332432937.