Test FFT Speed, LV7.1 vs. LV8.20

[Gary Rubin]

If I remember correctly, there was supposed to be a speed enhancement in some of the math functions between LV7.1.1 and LV8.x. I don't currently have access to LV8.x. Could someone please run the attached VI in both 7.1 and 8.x and let me know the speed results?

Thanks,

Gary

[French LVer]

Tested on an Intel Pentium M 1.7 GHz 512 Mo.

For 10 FFT

=> LV 7.1 : 126.5 to 128.2 ms

=> LV 8.2 : 143.6 to 145.0 ms

Might be better on a real computer

[Gary Rubin]

QUOTE(French LVer @ Mar 15 2007, 11:11 AM)

Tested on an Intel Pentium M 1.7 GHz 512 Mo.

For 10 FFT

=> LV 7.1 : 126.5 to 128.2 ms

=> LV 8.2 : 143.6 to 145.0 ms

Might be better on a real computer

Thank you very much!

I have to say that you have managed to raise more questions than answers...

My results are:

LV7.1.1 2.4GHz P4 , 2GB RAM, lots of other crap open: 249 ms. EDIT: WinXP Pro SP2.

LV7.1.1 2.4GHz Dual Xeon, 1GB RAM: 274 ms. EDIT: Win2k SP3.

Questions:

Why is yours twice as fast as mine?

Why is your LV7.1 faster than LV8.2?

Why is my P4 with lots of stuff in the background faster than my Xeon?

Anyone have any thoughts?

[AnalogKid2DigitalMan]

LV 8.2 1st run 166, 2nd run 106, 3rd through 5th runs 96-97. LV freshly opened with your code

Running P4 Duo Core 2.8GHz, 1G RAM (500M free), Win XP Pro SP2.

It was odd that the first run was so slow, 2nd run was faster, subsequent runs faitly repeatable. Maybe an initial memory allocation?

Sorry, only 8.2 available to me.

[Eugen Graf]

LV 8.0.1

AMD Athlon 64 3200+ 2.01 GHz 1 GB RAM XP+SP2

at 10 repeats have durations

from 127 to 130 ms

Eugen

[Gary Rubin]

Thanks to those who ran this for me. I guess I haven't yet confirmed whether LV8.x FFT is faster than the LV7.1.1 version, but I did notice something interesting.

I've attached a new vi that runs the one I posted above and plots runtime as function of FFT length. On my system, runtime starts to increase more-or-less linearly after 128kpoints.

I'm guessing, therefore, that below 128kpoints most of the time is spent doing the FFT, and I can see the logN part of the execution time. Above 128kpoints FFT, Labview is spending more time doing memory allocation than actually performing the calculations, so I'm mostly seeing a linear increase in time it takes to do the buffer allocation.

All you guys with the faster times than me probably have faster memory, or at least a faster memory bus.

Does that sound reasonable?

Gary

[AnalogKid2DigitalMan]

Gary:

I ran your original vi on my older Dell Precision 370 2.8GHz single core (circa 8-2004), 512M RAM (185M free) LV8.2 Win XP SP2. 81-82msec consistently (FASTER)

The original test I ran was on a Precision 380 circa 5-2005

I do not know RAM specifics, but using AIDA32 I did memory Read/Write tests.

Pecision 370: Read 5274MB/sec, Write 2081MB/sec

Precision 380: Read 5578MB/sec, Write 2152 MB/sec.

Do not know why the slower 370 memory executes faster.

[Gary Rubin]

QUOTE(AnalogKid2DigitalMan @ Mar 15 2007, 02:28 PM)

Do not know why the slower 370 memory executes faster.

Thanks. Quite the headscratcher...

Your signature seems quite fitting at the moment.

Gary