Happy Me Thread

[rogue]

Originally Posted by kladner

Being Chicago, the pertinent question is "how many times will it be counted?".

[Dr Sardonicus]

Originally Posted by rogue

Being Chicago, the pertinent question is "how many times will it be counted?".

That would depend on which party's candidates he voted for.

[kladner]

The real question for me is, "Will the outcome reflect the votes actually cast, or will chicanery exclude hundreds of thousands of legitimate ballots?"

[Dr Sardonicus]

Originally Posted by kladner

The real question for me is, "Will the outcome reflect the votes actually cast, or will chicanery exclude hundreds of thousands of legitimate ballots?"

Why not both?

[Xyzzy]

Originally Posted by Xyzzy

We won the lottery and managed to place a 3080 (founder's edition) on order for delivery Monday.

We decided to keep the 3080. It was delivered yesterday.

[kruoli]

Do you now have both?

[Xyzzy]

No, just the 3080 for now.

We need to sell our 2060 Super before we get any more toys.

[kriesel]

Finally sorted out most of the last box downed by a storm weeks ago. So 24 cores and 1 of the 2 gpus have now rejoined the fray. No idea what's wrong with the other gpu; doesn't even drive a display, but claims to be running at full clock rate while accomplishing nothing. It was running headless at the time of the storm.

[petrw1]

Originally Posted by Xyzzy

We decided to keep the 3080. It was delivered yesterday.

Please let me know the mfaktc rates you get....in case I decide I need one.
P.S. Are you using the 2047 class version?
For my 2080Ti it was 10% faster.

[LaurV]

Originally Posted by petrw1

P.S. Are you using the 2047 class version?
For my 2080Ti it was 10% faster.

What's that 2047 clases???
Mfaktc is either 4620 or 420 classes.

One being faster than the other doesn't depend much of the hardware, but it highly depends on the exponent range and bitlevel. It comes to balancing the sieving with exponentiation. Therefore it can be tuned a bit also from the sieve upper limit. But not much. Lots of candidates make sieving faster (i.e. high bitlevel, low p), while not many candidates (lower bitlevel and high p) leads to slower sieving and longer exponentiation for each remaining candidate, therefore you must decrease the number of classes to have more candidates for each class in the sieve. Sieving is more efficient on large chunks. On small chunks there is a lot of overhead to determine initial starting point for each chunk, and sieving is less efficient, as it eliminates less candidates at each pass, so it needs more passes. Depend of the range you work, bitlevel, and card (in this order) you can chose one variant of mfaktc, or the other.

[kriesel]

2047 M GpuSieveSize. In a modified mfaktc.ini, for a modified mfaktc.exe:

# GPUSieveSize defines how big of a GPU sieve we use (in M bits).
#
# Minimum: GPUSieveSize=4
# Maximum: GPUSieveSize=128 unless running a 2047-limit modified mfaktc
#
# Default: GPUSieveSize=64

GPUSieveSize=2047

There are indications performance would be helped a little on the fastest gpus if the variable was change from the current limiting signed 32 bit to unsigned allowing up to 4095. Another influence on low bit levels or high exponent is IO time for logging the class completions, and for which speed is helped by ssd or ramdisk.