mersenneforum.org

mersenneforum.org (https://www.mersenneforum.org/index.php)
-   Information & Answers (https://www.mersenneforum.org/forumdisplay.php?f=38)
-   -   Gigahertz Day (https://www.mersenneforum.org/showthread.php?t=24240)

storm5510 2019-03-29 12:36

Gigahertz Day
 
Gigahertz day is a concept that still seems a bit muddy to me after all these years working on this project.

One gigahertz day is a CPU running at 1 GHz for 24 hours. So, if a person has a 3.8 GHz processor, for example, this would be 3.8 GHz/Days in 24 hours.

I know it is probably much more complex than this when applying it to assignments. Is what I have the basics of this?

M344587487 2019-03-29 12:52

It's 1GHz for 24 hours of a specific CPU that's ancient by today's standards, an early core 2 duo or something near there. As a yard stick for throughput it's not the best especially when you start trying to compare work of different types, but it's only meant as a rough guide.

Brian-E 2019-03-29 13:04

@storm5510 I think what you write sums it up exactly. At least that's my understanding too. I compare it in my mind with the "kilowatt hour" as a measure of consumption of electricity.

@M344587487 Indeed. But I was running a machine on GIMPS with a processor speed of about 1 GHz right up until it packed up in November 2017. By the end it was taking about 5-6 months (running about a third of the time - I didn't leave it on 24 hours a day) to do a double check in the 40-45 million exponent range which the project had reached by then. My machine was occasionally the subject of a certain amount of hilarity here.:smile:

Uncwilly 2019-03-29 13:36

AMD CPU's never lived up to the 1GHz days per day.
The oldsters :paul: will remember the P90 years as a measurement of work done.

ATH 2019-03-29 15:00

You can calculate what 1 Ghz-day is defined as in terms of FLOP (Floating Point operations):

[url]https://www.mersenne.org/primenet/[/url]

Right now in the "Aggregate Computing Power" last 24 hours:
498.551 TFLOP/sec = 498,551 GFLOP/sec but it is also equal to: 249275 GHz-days.

So 1 Ghzdays is defined as 498551/249275 = [B]2 GFLOP/sec for 24 hours[/B] which is:

2 GFLOP/s * 86400 sec = 172,800 GFLOP = 172.8 TFLOP

So [B]1 Ghz-days is 172.8*10[SUP]12[/SUP] Floating Point operations[/B].


But all the Ghz-days calculations are approximate, I think it is hard to determine the exact number of FLOP that LL, PRP, P-1, TF and ECM tests requires.

kriesel 2019-03-29 18:53

[QUOTE=M344587487;512115]It's 1GHz for 24 hours of a specific CPU that's ancient by today's standards, an early core 2 duo or something near there. As a yard stick for throughput it's not the best especially when you start trying to compare work of different types, but it's only meant as a rough guide.[/QUOTE]
I think it was based on one core of a particular core 2 duo model.
The GhzD/day unit of throughput rate is particularly ugly. My engineering professors would never have tolerated the uncanceled units. Ghz-equivalent might have flown; Gheq.

kriesel 2019-03-29 18:56

[QUOTE=ATH;512128]
But all the Ghz-days calculations are approximate, I think it is hard to determine the exact number of FLOP that LL, PRP, P-1, TF and ECM tests requires.[/QUOTE]Indeed. The number of cycles required is not determinate. It varies depending on the results of roundoff checks, Jacobi checks, Gerbicz checks, etc. for the same exponent and computation type. (continuing from the last save file...)

storm5510 2019-03-29 22:43

I forgot about the cores. :numbskull:

I remember the big deal over Y2K. I had a 1 GHz. P3 at the time in the place I worked. I thought I was in hog-heaven.

It seems like the FLOP value would be determined by the they type of work. Anything pure integer would be zero.

Thank you all for your replies. Perhaps somebody will come along in the future and find this useful to them.


:smile:


All times are UTC. The time now is 22:22.

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2021, Jelsoft Enterprises Ltd.