Call for reliable LL and DC workers...

[chalsall]

Originally Posted by Dubslow

Do we still need to PM you to gain access to these tools?

If you are already a "GPU to 72" worker who has actually done work, probably no.

Interestingly, a certain person tried to torpedo our efforts.

So our AI agent's have their heuristics on edge...

If you're told you're not yet authorised or have already taken too much work, please just PM me and I will apologize for my 'bots for being too aggressive.

[Dubslow]

Well, I haven't tried yet, just curious. Won't try 'til late Sunday night at the earliest.

Originally Posted by chalsall

Interestingly, a certain person tried to torpedo our efforts.

What happened there?

[chalsall]

Originally Posted by Dubslow

What happened there?

It would be inappropriate to go into details.

But I think it is more than appropriate to say that it happened.

If under oath and during cross examination, I think six very public GIMPSers would be comfortable speaking about this.

Or, perhaps, only five out of six....

[chalsall]

Originally Posted by davieddy

PS I'm not going "Off Topic" am I?

Yes. You are. As usual....

You are what is known in the industry as "The Distractor".

I have to admit you are very good at your job....

[garo]

davieddy,
I have sent you a warning by PM but I am also making this public. Stop being a nuisance and wasting other people's time. I will not tolerate any more threadcrapping.

[akruppa]

Moved davieddy's garbage to the usual thread.

[Christenson]

Originally Posted by chalsall

It would be inappropriate to go into details.

But I think it is more than appropriate to say that it happened.

If under oath and during cross examination, I think six very public GIMPSers would be comfortable speaking about this.

Or, perhaps, only five out of six....

Turned in 2 factors today, got another week's worth of work...with a bit of luck, I'll get back to the high-end GPU I had awhile back and be able to set it to work...though the omens are not favorable...I had to go in to work today, the boss didn't understand why I didn't know details I had committed to his e-mail a week or two ago, and he found me resting at home instead of marching up a mountain.

[kladner]

I just dropped one mfaktc instance and dedicated the freed CPU core to low DC LL. While three mfaktc workers do produce higher throughput than two, it is getting into diminishing returns territory. I decided that the CPU core could make a better contribution for time spent. There seems to be plenty of TF getting done, and I still have two workers in that field.

EDIT: Quote Chistenson: "Turned in 2 factors today....."

After a dry spell going back to 11/23/2011, I finally hit a factor in the last day's running.

[Christenson]

@Kladner...I've been having a wet spell, instead...something charmed about those exponents, as if there were errors when the exponents were TF'ed to low levels.

Optimum, of course, is the amount of LL-equivalent testing you can knock out per unit time...counting 2 LL tests per factor found (at a value of 4x your LL-D at 25M) at 50M. DCs have value...and you can ask CUDALucas to do a DC on your GPU as well..

So your move seems quite reasonable. I have a GTX480 that does one thread of mfaktc and the other of CUDALucas. The other three cores do P64 work, generally P-1.

[kladner]

Originally Posted by Christenson

Optimum, of course, is the amount of LL-equivalent testing you can knock out per unit time...counting 2 LL tests per factor found (at a value of 4x your LL-D at 25M) at 50M. DCs have value...and you can ask CUDALucas to do a DC on your GPU as well...

I actually had not thought of using CL for DC, though that would take a bite out of my TF throughput. I'll experiment with it and see how it works out.

I admit to having a bit of trouble following the numerous discussions of relative benefits of different kinds of work. You might call this move a bit of diversification of the "portfolio". It does seem that those discussions frequently mention the need for DC, especially on a CPU. There have been some doubts expressed regarding CUDALucas' reliability.

Of course, I will continue to monitor the talk and adjust things if a different (perception of) consensus should develop.

EDIT: So my current line-up is:
CPU1: 1 1st time LL, 2 P-1, 1 Low DC, and 2 cores feeding mfaktc
CPU2: 2 P-1

[LaurV]

Originally Posted by kladner

I admit to having a bit of trouble following the numerous discussions of relative benefits of different kinds of work.

The thinks are very simple. But still dependent of the "wave" you chose to work, and the CPU/GPU you have.

1. Assuming you have a "somehow old" computer, and no GPU, the most profitable for the project is that you do P-1.

2. For a good/new computer without GPU, you still can do, in this order of importance from highest to lowest: P-1, DC, first-time-LL. Never do TF. The argument is that TF is wasting time on CPU, as an average GPU is still a couple of laps faster then a top CPU. P-1 is still most important, as it can not be (yet) done by GPU's.

3. For an average-to-high computer with an average GPU, it depends of the front you chose to work on.

• At LL front, you do TF. Joining GPU-2-72 is a good idea. With TF at LL front, you find a factor (therefore clearing one exponent) every 3-4 days, which is faster then LL-ing, even for high-end GPU's (which would need 12-14 days to clear one exponent by doing two LL's).
• If you chose to work at DC front, the things are arguable.You can do TF or DC, depending on how fast CudaLucas runs on your GPU compared with mfaktc.

4. For a Fermi or other high-end GPU (in this case I assume you also have a high end CPU to feed the GPU, otherwise you wasted the money on that GPU), also, it depends of the front you want to workout.

• Same as above, at LL front you will find a factor (clear one exponent) every 2-3-4 days on average, so doing TF is better than doing LL, for which you would need over 10 days to clear an exponent (two LL's plus some P-1).
• At DC front, things changes. Your GPU will be able to clear one exponent by DC-ing it, every 24-30 hours. For TF at DC front, you would still need few days to find a factor. Therefore, the best way to go is DC. This has also the advantage that CudaLucas is using your CPU much less then the 2-3 copies of mfaktc that would run if you try to do TF and max the GPU. So your CPU has more "freedom" to workout some P-1 assignments form GPU-2-72 or directly from PrimeNet.

Things could (and should) change in time, depending on the progress on hardware/software and depending on how the two fronts will advance. But now, this is the best way to go.

edit: I give a beer to anyone who put this in a nice Visio diagram