RIP DCTF.

[Mark Rose]

227 days.

[Mark Rose]

223 days.

So we're getting at most 4 days per day now. After we level off in a week we'll have about 200 days of work.

[Xyzzy]

FWIW, we only submit results on Sundays.

[Mark Rose]

At some point on Saturday my work will be without power while the building power is upgraded. Since my machines require a password to decrypt on boot, and I won't be back until Monday, I'll be short about 2 THz-d this weekend.

[Mark Rose]

And the power shutdown was rescheduled for a later time. Woohoo. Though I guess the panic is over here now anyway.

[petrw1]

103,792 GhzDays DCTF in the last week. (Can someone tell me when the days change over...maybe I need to capture these before 0:00 GMT)
60 different contributors
207 Factors found
12,036 P1/LL/DC work saved

28 contributors currently have assignments
8,625 Assignments out.

206.7721 estimated days to completion (down 32 days in the last week)
July 15, 2015

[chalsall]

Originally Posted by Mark Rose

223 days. So we're getting at most 4 days per day now. After we level off in a week we'll have about 200 days of work.

A quick back-of-the-envelope calculation suggests we'll end up finishing all the DCTF work in about 253 days from now; i.e. late August or early Septermber of 2015.

This (very rough) estimate is based on the trending of the DCTF processing being done. It could (obviously) change siginificatnly if the firepower changes.

[chalsall]

Originally Posted by VictordeHolland

Those are different for the different GPU architectures and whether you prefer TF or CudaLucas.

To 72bit
GTX980 @38M 71.92bits http://www.mersenne.ca/cudalucas.php?model=566
GTX580 @40M 71.96bits http://www.mersenne.ca/cudalucas.php?model=12
GTX680 @43M 72.04bits http://www.mersenne.ca/cudalucas.php?model=467
GTX780 @49M 71.93bits http://www.mersenne.ca/cudalucas.php?model=492

So I would personally say from 40M on to 72bits and @50M to73bit, as the GTX580 is still a very popular card for TF. People with a GTX780 and who do mind their efficiency could return to LLTF or run CudaLucas instead.

Thank you for pointing this out, and sorry for the delay in responding. I can still barely type thanks to "the chicken"...

It would be trivial for me to increase the bit-level tarket from 40M and up to be 72 instead of 71, and from 50M to 73 -- I will do so, and we can see what our estimated completion days looks like. Please keep in mind this will siginificantly increase the completion date, but I think it Makes Sense.

Does anyone think this is a bad idea? It could be argued that for some cards, we should have been going to 72 at 39 and possibly even lower. Anything over 71.5 is net profitable.

This actually brings up a thought... It would take a while to implement (it's not hard, but not trivial either), but I could add another pulldown menu in the "Get Assignments" form which let's the requester inform GPU72 what class of GPU they have. E.g.: "GeForce GTX 580", etc al.

Thoughts, comments, alternatives?

[petrw1]

Originally Posted by chalsall

Thank you for pointing this out, and sorry for the delay in responding. I can still barely type thanks to "the chicken"...

It would be trivial for me to increase the bit-level tarket from 40M and up to be 72 instead of 71, and from 50M to 73 -- I will do so, and we can see what our estimated completion days looks like. Please keep in mind this will siginificantly increase the completion date, but I think it Makes Sense.

Does anyone think this is a bad idea? It could be argued that for some cards, we should have been going to 72 at 39 and possibly even lower. Anything over 71.5 is net profitable.

This actually brings up a thought... It would take a while to implement (it's not hard, but not trivial either), but I could add another pulldown menu in the "Get Assignments" form which let's the requester inform GPU72 what class of GPU they have. E.g.: "GeForce GTX 580", etc al.

Thoughts, comments, alternatives?

Pulldown sounds like a cool idea. Could I assume that for us using MisFit it will be implemented there too?

With all the focus on DCTF I notice a lot less progress in LLTF. If you increase the DCTF Bit levels and hence duration do you foresee us starting to fall behind the need in LLTF?

I suspect you will soon be bumping the LLTF leading edge to 75 bits.

[chalsall]

Originally Posted by petrw1

Pulldown sounds like a cool idea. Could I assume that for us using MisFit it will be implemented there too?

Of course!

Originally Posted by petrw1

With all the focus on DCTF I notice a lot less progress in LLTF. If you increase the DCTF Bit levels and hence duration do you foresee us starting to fall behind the need in LLTF? I suspect you will soon be bumping the LLTF leading edge to 75 bits.

That raises an interesting and important point. Thank you for bringing it up!

Right now, even with the much lower LLTFing happening, we're still FAR ahead for all the "Categories", with the excepetion of the Cat2 range. We're still comfortably ahead there, but not by the huge margins of the others.

HOWEVER, this is only when considering going to 74 is "optimal" -- in fact we should be going to 75 for many of the LL ranges.

What do people think? Should we aim to go "all the way" in both the DC and LL ranges, and see how we fair? We could always release at "only" 74 if there was a crunch.

[petrw1]

Originally Posted by chalsall

Of course!



That raises an interesting and important point. Thank you for bringing it up!

Right now, even with the much lower LLTFing happening, we're still FAR ahead for all the "Categories", with the excepetion of the Cat2 range. We're still comfortably ahead there, but not by the huge margins of the others.

HOWEVER, this is only when considering going to 74 is "optimal" -- in fact we should be going to 75 for many of the LL ranges.

What do people think? Should we aim to go "all the way" in both the DC and LL ranges, and see how we fair? We could always release at "only" 74 if there was a crunch.

My vote it TF to "optimal ranges" in both and use the "escape hatch" when necessary.
That way many/most/all (if we're lucky) will get optimal TF.

BTW I have a premonition there will be a few more GPUs under the Christmas Tree this year.


PS: Does anyone have personal experience to tell me if a 500W PS is enough for a i5-2500 and a GTX-970 GPU? Neither OC'd. Nothing but GIMPS on it.