93.5 GHzDays for a 51M LLtest

[davieddy]

Is this a triumph for the latest software?
FFT size?

The going rate has been well over 100 GHz days.

David

[petrw1]

Originally Posted by davieddy

Is this a triumph for the latest software?
FFT size?

The going rate has been well over 100 GHz days.

David

I too was over 100 GDs for 51M tests until v26 improvements, tuning and new FFT sizes and subsequent reduction in run times.

So if a test that was 105 GDs with v25 and a 3072 FFT and took a month a some core - and now with v26 and a 2644 FFT and takes 27 days it makes sense the GDs should be proportionally less.

[davieddy]

Originally Posted by petrw1

I too was over 100 GDs for 51M tests until v26 improvements, tuning and new FFT sizes and subsequent reduction in run times.

So if a test that was 105 GDs with v25 and a 3072 FFT and took a month a some core - and now with v26 and a 2644 FFT and takes 27 days it makes sense the GDs should be proportionally less.

Hmmm 2644 = 4*661
Is that size right?

You must mean 2688 shirley.

[KingKurly]

Originally Posted by davieddy

Hmmm 2644 = 4*661
Is that size right?

You must mean 2688 shirley.

Strange, my test of M51732287 seems to be using a 2800K FFT... I assume that exponent qualifies as a 51M LL test? :) Or am I totally misunderstanding?

I am using 26.4 build 1.

[davieddy]

Originally Posted by KingKurly

Strange, my test of M51732287 seems to be using a 2800K FFT... I assume that exponent qualifies as a 51M LL test? :) Or am I totally misunderstanding?

I am using 26.4 build 1.

Well, if the benchmark page is up to date,
2560K is good up to 49.10M
3072K is good up to 58.52M

So 2800 (2*2*2*2*5*5*7)K is in the right ballpark.
I am testing 51M on v25.11, which uses 3072K.

David