GHz days converting?

I have read that the old GHz day was calculated with a P90 day in mind. While they new GHz day is calculated as one day of a theoretical Core Duo 2.4 Ghz core.

Did you convert the old P90 GHz days producer list when you migrated to the new GHz standard?
Will you convert next time you use another processor as a standard for GHz days?

Thank you in advance.

[S485122]

Originally Posted by Unregistered

Did you convert the old P90 GHz days producer list when you migrated to the new GHz standard ?

Yes, the conversion was done.

Originally Posted by Unregistered

Will you convert next time you use another processor as a standard for GHz days ?

Probably, yes.

Jacob

[cheesehead]

There is a way to end these pesky conversions, of course.

Ask me to explain about 6-7 months from now.

[lfm]

Originally Posted by cheesehead

There is a way to end these pesky conversions, of course.

Ya, I think a unit of the FLOP (floating point operation) would be better understood and need less explaining. Probably as a petaflop or ectaflop.

Of course some people would still get confused with FLOPS (flop per sec(FLOPPS?)). Speed as opposed to work.

Also the non floating point trial factoring work would still be controversial I suppose.

[cheesehead]

Originally Posted by lfm

Ya, I think a unit of the FLOP (floating point operation) would be better understood and need less explaining.

No, I'm referring to a light-hearted but practical proposal I made several years ago, for a substantially different basis of measurement, related to thermodynamics rather than a specific CPU. Many forum veterans probably remember it (some are trying to forget), but it was before your time.

Of course some people would still get confused with FLOPS (flop per sec(FLOPPS?)). Speed as opposed to work.

Of course, speed has to be multiplied by the elapsed time during which the speed was maintained, in order to denote an amount of work accomplished.

Also the non floating point trial factoring work would still be controversial I suppose.

With my proposed system, that controversy would sink out of the group's attention, IMO.

[MatWur-S530113]

Originally Posted by Unregistered

I have read that the old GHz day was calculated with a P90 day in mind. While they new GHz day is calculated as one day of a theoretical Core Duo 2.4 Ghz core.

Did you convert the old P90 GHz days producer list when you migrated to the new GHz standard?
Will you convert next time you use another processor as a standard for GHz days?

Thank you in advance.

A theoretical Core Duo 2.4 GHz produces 2.4 GHzdays/day, both cores produce 4.8 GHzdays/day (theoretical... ;)

[R.D. Silverman]

Originally Posted by MatWur-S530113

A theoretical Core Duo 2.4 GHz produces 2.4 GHzdays/day, both cores produce 4.8 GHzdays/day (theoretical... ;)

These measures are meaningless. Read:

R. Silverman
Exposing the Mythical MIPS Year, IEEE Computer

[retina]

Originally Posted by R.D. Silverman

These measures are meaningless. Read:

R. Silverman
Exposing the Mythical MIPS Year, IEEE Computer

And for those too lazy to search:

http://academic.csuohio.edu/yuc/perf...tical_MIPS.pdf

[cheesehead]

Originally Posted by R.D. Silverman

These measures are meaningless. Read:

R. Silverman
Exposing the Mythical MIPS Year, IEEE Computer

I'm surprised by your dismissal. Are you, perhaps, referring only to a concern that the GIMPS unit "GHz day", because of its name, could be easily confused with more ambitious units (mis-)used outside of GIMPS? In that case, it's a matter of nomenclature or jargon, not meaninglessness.

From the article's subtitle:

MIPS can be a useful measure of processor speed, and MIPS years can be a useful measure of work.

You specify conditions (rules) for proper use, but GIMPS's particular version of "GHz day" meets those conditions, it seems to me.

But it should come with a set of rules. As long as you use it as a relative, rather than absolute, measure, and as long as you make the comparisons by running similar types of programs on similar types of architectures, MIPS is a useful measure.

It seems to me that GIMPS is using its particular version of "GHz-day" in a manner that satisfies your conditions for usefulness.

It's used only as a relative measure within the project, for comparing credits given for GIMPS results, or for measuring project progress in a relative, internal, way, not as some absolute standard valid outside this project. GIMPS doesn't represent it to the rest of the world as being an absolute measure.

It's calibrated by running only one piece of software on one particular CPU (at a time, anyway, depending on the founder's hardware inventory).

Historically, however, people have used MIPS without following these rules.

But isn't GIMPS appropriately following your rules?

Later on, the article declares that "The MIPS year measurement has four basic problems", but none of them applies to the way GIMPS uses its particular version of "GHz day" internally!

- -

Perhaps we should give the unit a more restrictive name, such as "GIMPS official-progress-day", to avoid confusion with the mythical measures you rightly decry.

[davieddy]

Originally Posted by retina

And for those too lazy to search:

http://academic.csuohio.edu/yuc/perf...tical_MIPS.pdf

For those too lazy/sensible to wade through that,
note that it seems that 1 GHzday/day equates to 2 "Actual" GFLOP/s
or 2 FLOP/clockcycle, according to the Primenet summary.

BTW What is "Potential" TFLOP/s ?????

David

And for that matter, to what does "work unit" refer?

[lfm]

Originally Posted by davieddy

BTW What is "Potential" TFLOP/s ?????

thats what they'd get if all the assigned work was actually completed in the time estimated. There are lots of jobs started but then abandoned.

And for that matter, to what does "work unit" refer?

A work unit is one LL test or one TF run or whatever. Like one line in the worktodo.txt file.