Prime95 version 28.5 (deprecated, use 28.7)

[hensowi]

This is the message I get:
Optimizing for CPU architecture: Unknown Intel, L2 cache size: 256 KB, L3 cache size: 4MB

I am looking at the CPUSupportsXXX flags in undoc.txt now.

[Prime95]

In Options/CPU dialog box prime95 will tell you if it thinks AVS and FMA are supported.

Also, when you start or resume a test, it will tell you if an AVX or FMA FFT is being used.

[ewmayer]

Originally Posted by Prime95

From undoc.txt:

...The Program Normally Uses The Smaller Fft Size If The Average
 Roundoff Error Is Below A Value In 0.241 To 0.243 Range. ...

"Average Roundoff Error" defined how? Average over all (unnormalized, unrounded) convolution output words and multiple iterations, or the maximum Roundoff Error among the output words, averaged over multiple iterations?

[Prime95]

Originally Posted by ewmayer

"Average Roundoff Error" defined how? Average over all (unnormalized, unrounded) convolution output words and multiple iterations, or the maximum Roundoff Error among the output words, averaged over multiple iterations?

Start with random data.
Execute 1000 LL iterations getting the maximum roundoff error for each iteration.
Average the 1000 samples.

[ewmayer]

Originally Posted by Prime95

Start with random data.
Execute 1000 LL iterations getting the maximum roundoff error for each iteration.
Average the 1000 samples.

Thanks - that's what I call the "Avg Max Error" - I also use the "Max Max Error" in gauging roundoff impact of e.g. choice of a specific short-length DFT algorithm or FMA usage, although it's a much noisier metric, as a result of which I've switched to doing 10000-iteration test runs at each FFT length once I've found the best-FFT-params-to-use for the various lengths using 1000-iter timing runs.

Random data meaning random signed ints whose average bitsize (assuming conversion to a nonnegative-digit normalization) equals p/n? [p = M-exponent, n = real-signal array length.]

[fregames]

is there a way to run prime95 on many computers testing different sections of the same LL number?

[Dubslow]

Originally Posted by fregames

is there a way to run prime95 on many computers testing different sections of the same LL number?

No unfortunately. Each individual iteration is the direct product of the previous iteration. Therefore they must be done exactly in order.

[LaurV]

But of course, no one stops you to run P95 in as many computers you can, doing different LL tests...
The more the better...

[storm5510]

What is the purpose of these values and are they really necessary? I am entering them by hand from a mobole device. It gets tedious after a while.

[TheMawn]

Assignment ID. This number is actually quite important if you're interested in being credited with the work. This is an extra bit of safety used to make sure you work on the correct job. If the AID doesn't match the actual assignment, I believe Prime95 discards the assignment and goes to work on the next?

[Batalov]

Originally Posted by storm5510

I am entering them by hand from a mobole device.

That is definitely not the best way to enter them. Even mobile devices have a copy-paste equivalent.