GIMPS HITS {insert number here} COMPUTERS

[Prime95]

Originally Posted by jinydu

So why the bias against trial factoring?

There is no bias against trial factoring. You are credited, to the best of our ability, with the CPU time you actually used.

In v5, we'll probably revisit the formula to bring it up-to-date.

[garo]

Ah! So when the Primenet server was first designed, computers were much better at TF relative to LL testing than they are now. Seems obvious since George has improved the LL testing code so many times whereas TF has only undergone a few revisions.

[jinydu]

I haven't seen this in a long time (if ever, I don't quite remember):

The virtual machine's sustained throughput* is currently 20305 billion floating point operations per second (gigaflops), or 1686.8 CPU years (Pentium 90Mhz) computing time per day. For the testing of Mersenne numbers, this is equivalent to 725 Cray T916 supercomputers, or 362.5 of Cray's most powerful T932 supercomputers, at peak power. As such, PrimeNet ranks among the most powerful computers in the world. (*Measured in calibrated P5 90Mhz, 32.98 MFLOP units: 25658999 FPO / 0.778s using 256k FFT.)

[ninjabill]

Hi there, what would be the output of a P4 in flops / mhz? Am I being deeply dim, and the answer is 1,000,000, or is the calculation a little more in depth?

[jinydu]

The answer is not 1,000,000. In reality, the answer depends on how fast the other parts of your computer run.

I'm sure there are others in this forum who could answer that question much better than I can, so I'll leave it to them.

[axn]

Originally Posted by ninjabill

Hi there, what would be the output of a P4 in flops / mhz? Am I being deeply dim, and the answer is 1,000,000, or is the calculation a little more in depth?

P4 (and AMD Athlon/Athlon 64) can sustain 2 FLOP / cycle. However, this is the theoretical peak output.

[ninjabill]

So is it possible to calcualte how many P90 cpu years/day throughput you get from a P4 running at n Mhz?

regards

geoff (maths numpty)

[axn]

Originally Posted by ninjabill

So is it possible to calcualte how many P90 cpu years/day throughput you get from a P4 running at n Mhz?

If only it were that simple

The actual thruput will be affected by

• L2 Cache size
• L2 Cache latency
• FSB speed
• Memory speed
• Exponent being tested (FFT size used)
• etc...

Also, performance doesn't scale linearly with MHz.

OTOH, given an exponent and a particular machine, you can quickly calculate the P90 CPU years/day easily.

Just run a benchmark for the exponent and note down the iteration time. Then go to this page and figure out the corresponding iteration time for a PII-400.
Then your machine's P90 CPU yrs/day = <PII-400 iter time> * 5.5 / <Your iter time> / 365
Simple, isn't it?

[ninjabill]

Thanks for that, I seem to be getting about 97 P90 years/year in a P4 3.0 (530 with HT disabled) does this value sound about right?

thanks

geoff

[jinydu]

Originally Posted by axn1

Also, performance doesn't scale linearly with MHz.

It doesn't? Is the second derivative positive or negative?

[axn]

Originally Posted by ninjabill

Thanks for that, I seem to be getting about 97 P90 years/year in a P4 3.0 (530 with HT disabled) does this value sound about right?

Sounds about right.

Originally Posted by jinydu

It doesn't? Is the second derivative positive or negative?

If we plot efficiency (Actual thruput/theoretical thruput) as a function of frequency, we'll see that efficiency drops as frequency goes higher. Does that answer your question? Linear scaling would mean that efficiency would remain constant irrespective of frequency.

PS:- All of this assumes that all other things (FSB, cache, etc...) remain the same.