Calculating dual core performance (GF/s)

0

I'm trying to figure out how to calculate performance (measured in GFLOPS) My thought was I would use GFLOPS = cores x clocks x (FLOPs/cycle), my issue right now is that one of my cores runs at a different speed than the other:

Core 1: ARM Cortex M4 - 167 MHz
Core 2: ARM Cortex A5 - 450 MHz

The formula I found (above) is assuming two cores running at the same speed. Is there an adjustment to figuring out GF/s on a dual core system where they're running at different speeds? Or does it make more sense to calculate each separately and then average the two?

Mike

Posted 2012-12-04T14:26:08.897

Reputation: 849

Answers

2

The reason the formula tells you to multiply by the number of cores is that you are actually performing addition multiple times assuming the core speeds are the same. If that is not the case, just separate the calculations and add the results:

[167 x (Flops/cycle)]+[450 x (Flops/cycle)]= GFlops

That's all there is to it.

Everett

Posted 2012-12-04T14:26:08.897

Reputation: 5 425

Gottcha, I was thinking that too, but I wasn't sure if that was oversimplifying it... considering extra IPC requirements between cores or shared resources or anything – Mike – 2012-12-04T14:41:38.220

Nope, that's really minor in the grand scale of a GFlop ;) But, good thinking, important not to miss stuff. However, keep in mind this a perfect world calculation too. Typically that is "understood" when you use it. Always expect to max out 5% beneath that calculation (a rule of thumb I've used YMMV). – Everett – 2012-12-04T14:44:26.977

Fun fact about GFlops. There is a Cray Super Computer in one of the scenes in Sneakers. Did you know that the current model iPad has the same GFlops rating as the SuperComputer used in that movie? – Everett – 2012-12-04T14:50:17.887

Asked: 2012-12-04T14:26:08.897

Viewed: 144 times

Active: 2012-12-04T14:37:52.277