Prime95 29.4 PRP oddity noted

[tServo]

I noticed something odd when doing a PRP of an exponent 87,xxx,xxx with the latest version of Prime95 ( 29.4 ).
It starts going predictably and chugs along saying the ETA is 3 days 6 hours or so then suddenly the ETA jumps to a huge number like 70 days and change.
This happens after 4 hours to 12 hours or so ( I ran a few tests ). However, the ms/iter does not change but stays in the same ballpark it has been.
My thought is that it got errors and started using a bigger FFT size but it seems the ms/iter would have changed.

My machine:
Processor is i7-6950x 10 core slightly overclocked.
I use 1 worker giving it 8 cores.
Nothing else is going on in the machine.
Windoze 10.
The processor supports 4 channel memory and the machine has 4 channel 2400, 32GB memory.

[irowiki]

If you just let it run does it finish in the 3 days?

[kriesel]

Originally Posted by tServo

I noticed something odd when doing a PRP of an exponent 87,xxx,xxx with the latest version of Prime95 ( 29.4 ).
It starts going predictably and chugs along saying the ETA is 3 days 6 hours or so then suddenly the ETA jumps to a huge number like 70 days and change.
This happens after 4 hours to 12 hours or so ( I ran a few tests ). However, the ms/iter does not change but stays in the same ballpark it has been.
My thought is that it got errors and started using a bigger FFT size but it seems the ms/iter would have changed.

My machine:
Processor is i7-6950x 10 core slightly overclocked.
I use 1 worker giving it 8 cores.
Nothing else is going on in the machine.
Windoze 10.
The processor supports 4 channel memory and the machine has 4 channel 2400, 32GB memory.

Check the results file. Are there Gerbicz check errors? If so it will back up perhaps a large number of iterations and recalculate. The time per iteration does not change, but the rate of progress per calendar day is recalculated.

[henryzz]

Could it be during the Gerbicz error checks?

[tServo]

Originally Posted by irowiki

If you just let it run does it finish in the 3 days?

Good idea.I shall let it run to the 3.5 days mark.

Also, I checked the results file and nothing there to indicate a problem.

[irowiki]

Originally Posted by tServo

Good idea.I shall let it run to the 3.5 days mark.

Also, I checked the results file and nothing there to indicate a problem.

My FX-8350 after a while goes to "infinite" time remaining but still finishes "on time".

[tServo]

Originally Posted by irowiki

My FX-8350 after a while goes to "infinite" time remaining but still finishes "on time".

Thanks.
The more I think about, it looks like it just be a trivial fault in computing time to go,
just like you experience.
Anyway, I'll know soon enough as its running now.

[kriesel]

What are the power settings on the system?

[tServo]

Originally Posted by kriesel

What are the power settings on the system?

Power settings are:

Turn off display after 15 minutes

Turn off system: never

Power setting: High performance


The PRP test finished in the very first eta timne as irowiki suggested,
so the eta computation goes wonky at some point.

[irowiki]

This is what mine does on LL tests, for fun:

[Sep 24 23:22] Iteration: 84060000 / 86542459 [97.13%], ms/iter: -1.#QO, ETA: -2147483648:-2147483648:-2147483648

[ET_]

Originally Posted by tServo

Power settings are:

Turn off display after 15 minutes

Turn off system: never

Power setting: High performance


The PRP test finished in the very first eta timne as irowiki suggested,
so the eta computation goes wonky at some point.

Could it be something related to the way Windoze configures time? I mean a call to a ntp server gone awry, a bad battery backup, a failing EPROM to maintain the time of day...