Short WUs:

If a prime is found on a Short WU, it will be a new world record for a GFN, replacing the one in my signature, will be no worse than the 13th largest prime of any type ever found, and would most likely be 11th or 12th overall. Short WUs have an exponent of 524288 (2^19).

The purpose of the Short WU project is to 1) Find a new, largest GFN, 2) Find a prime at N=534288 in BOINC, and 3) provide a BOINC GFN project where CPUs can particpate.

LONG WUs:

If a prime is found on a Long WU, it will be a much larger number. Except for some of the very first (and therefore smallest) of these WUs, any prime that is found will be the largest prime ever found, of any type. Of the approximately 110 thousand candidates to be searched, the first 289 are smaller than the current largest known prime, and all the rest would set a new record. The current leading edge of the Long WUs is over 14 million digits long, and is about a million digits longer than the current world record. Long WUs have an exponent of 4194304 (2^22).

The purpose of the Long WUs is simple, and is the driving force behind the effort to bring GFN searching to BOINC: We're looking for the largest prime number ever found.
____________
My lucky number is 75898⁵²⁴²⁸⁸+1

So, a little while ago, I started to put a bit of work into this subproject. I wanted to run the "world record" search, but keep getting computation errors during (whenever the task has to be paused, then resumed; but it happens randomly).
At any rate, I found it silly to attempt to run world record searches and have them keep failing on me after a long time.

As for the details on my GPU, I have heard of that issue before. It's interesting, though, because it's only ever failed on world record GFN searches (I would have to check my history, but I'm 95% sure about that). I've been running PPS sieves and short GFN units just fine. (Not to mention the use it has had in other projects).

As of now I lowered the memory clock to 1900mhz.

Thanks for the second clarification.

I did want to note that my card is (likely) less affected because the factory overclocking was rather low (It's Gigabyte brand, as oppose to EVGA). So that reason, alone, could be why I was still able to run some of the short tasks.

I'll try to get a world record task and see if it can run well. I just hope to not end up failing the task at a late percent (The farthest the card has run for is ~40%).

Indeed.
The previous clock setting was 2050mhz, so 1900mhz is a decently steep drop. (In fact, I think that's the reference clock speed)
Anyway, I'm completely one last PPS sieve and then I'm getting a GFN_WR. Wish me luck!

I assume, that this can help to identify some problems @start ??
Not sure..

Best luck .. parabol

@Alex
Gigabyte, as well as Asus and MSI seems to have better cooling on most of their GPUs than EVGA or reference cooling. Even though your temps are getting up to 75C on the GPU chip, the memory modules may run a little cooler then EVGA or reference because of this, which would allow less problems at OC levels than others. Backside of the card cooling also helps this, so if you have decent air movement across it that too helps keeping memory modules (and the whole card) slightly cooler as well. The long tasks use a lot more memory, and will run hotter, and that may be why your long WUs error out at some point on your 550 Ti.

Hope your able to complete your long WUs now with the lowered memory clock. Good luck.

NM*

____________
Largest Primes to Date:
As Double Checker: SR5 109208*5^1816285+1 Dgts-1,269,534
As Initial Finder: SR5 243944*5^1258576-1 Dgts-879,713

Another question:
From my calculations, it apparently will take 270 hours to complete the task.
What are some typical run-times for these? I'm wondering if it might be smarter for this project (as well as others) to mostly focus on PPS sieves instead.

Well, I will start off by saying that credit doesn't actually motivate me in any direction. What I meant by that was I wanted to compare my computation times to others who might have a more "apt" card for the tasks.

I regularly run PPS sieves and PPS llr tests. At the moment I've returned to dedicating my ATI GPU to multiple projects, including this one. I was literally comparing the "need" for GFN tasks to get done to my ability to complete them. Whether a prime is found or not, I wouldn't care if it's in my name, but instead simply that I helped in finding it.

Another question:
From my calculations, it apparently will take 270 hours to complete the task.
What are some typical run-times for these? I'm wondering if it might be smarter for this project (as well as others) to mostly focus on PPS sieves instead.

A word of caution about WU durations:

This early in the World Record project, there are substantial differences in WU durations depending on the value of b. Don't forget that they're not handed out in sequential order because of resends of errored tasks -- and there's lots of those! The run times of your WUs will go up and down and up and down -- sometimes by very significant amounts.

Right now, the smallest b still being worked on is 52 and the largest is 2376.

On my GTX 460, those WUs would take 109:35:05 and 215:34:58, respectively. That's quite a difference. So be careful when comparing one GPU to another based on how fast they did "a long WU".

The short WUs are more consistent in length, although they are slowly increasing. Comparing run times of the short WUs would give you more consistent comparisons.

In a couple of months, the durations of the long WUs won't vary so much, and using WU run times to compare hardware will be more reliable. But for right now it's probably better to run the benchmarks and compare those times.

There's a fairly easy to apply formula that can give you the expected run time based upon the benchmark results and the value of B. That's actually what the program does when it prints out the expected run time.

If you just want to know what the fastest GPU is, for us, that's the GTX 580, which is about 2.5 times as fast as a 550 Ti.
____________
My lucky number is 75898⁵²⁴²⁸⁸+1

I have the following question for you.

I have completed five GFN tasks thus far. Still no World Record tasks, although I have a couple of these tasks in progress right now.

They all run by means of CUDA.

I am guessing that these tasks are all LLR-based, even though they run by means of CUDA.

Yes, I have asked this question before. But the point I would like to make this time is that I am able to see that GFN also can be run as a sieving task directly, meaning it only returns factors (if any) when it finishes up.

Are such sieving tasks for GFN readily available and if so, would I be able to immediately see the difference between these two types of GFN while they are running?

Just an add-on question. How does this work out (both types) when it comes to 64-bit processing?

Just an add-on question. How does this work out (both types) when it comes to 64-bit processing?

The GeneferCUDA app is a 32 bit app. Since most of the processing is done on the GPU, it doesn't matter much if the CPU is running a 32 bit or 64 bit program. When I tested it, a 64-bit version of GeneferCUDA was slower than the 32 bit version.