Glucas results on dual G5

[Paulie]

I think the G5 would be a factoring monster with it's dual 64bit FPU's.

Welcome Apple to the 21st century! :) :) :)

[Aillas]

I really don't know too much about FFT, but I remember a press release from Apple about working with Genentech on FFT and I found this on the Apple web site:

<http://developer.apple.com/hardware/ve/acgresearch.html>

They have developped a FFT library that uses the Velocity Engine (Altivec) unit of the G4.

Does this library could be used for mersenne?
Is it more efficient than GLucas or MLucas one?

Ludovic

[GP2]

Originally posted by Aillas
<http://developer.apple.com/hardware/ve/acgresearch.html>

From the linked page:
"Please direct inquiries about ACG research to Richard Crandall."

Does this library could be used for mersenne?
Is it more efficient than GLucas or MLucas one?

Maybe someone should "direct an inquiry"

Then again, the linked-to paper dates from 2000.
http://developer.apple.com/hardware/ve/pdf/g4fft.pdf
It's for the G4... old news presumably.

[ewmayer]

Originally posted by Aillas
I really don't know too much about FFT, but I remember a press release from Apple about working with Genentech on FFT and I found this on the Apple web site:

<http://developer.apple.com/hardware/ve/acgresearch.html>

They have developped a FFT library that uses the Velocity Engine (Altivec) unit of the G4.

Does this library could be used for mersenne?
Is it more efficient than GLucas or MLucas one?

Ludovic

The Altivec FFT is strictly single-precision, which the only floating-point precision the AVec is capable of. Useless for LL testing, I'm afraid.

[Dresdenboy]

As we know, Altivec is not useful for LL testing and this disadvantage is now even greater for the G5/PPC970 since it's Altivec throughput is lower than on a G4.

But something else got better -> the G5 can do 4 double precision Ops/cycle using FMAC on the 2 FPUs. So a dual 2GHz G5 Mac could theoretically achieve 16 Gflops in double precision (if my info is correct).

[Paulie]

Some good info:

http://developer.apple.com/technotes/tn/tn2087.html

Take advantage of the additional double-precision FPU

The G5 has two complete double-precision floating-point units, and each one offers better performance than the single floating-point unit in the G4. Having 2 FP scalar units can be viewed by software as 2-way double precision vectors. To make sure that you can get the best use of the additional FPU, schedule your code so that dependencies are minimized (via loop unrolling, software pipelining, etc.) so that no one FPU is the bottleneck in your code. Write your FP code so that it can run on both FPUs simultaneously -- each unit has 6 cycle execution latency, so SW should attempt to fill 12 pipeline slots. It may be simpler for SW to treat the CPU as having a single FPU with 12 cycle execution latency

[rogue]

I believe the Mlucas can also be compiled and run on a Mac. Could somone get some benchmarks for it on a G5?

--Mark