From: Peter Arremann loony@loonybin.org

It is a valid benchmark though :-) compile speed is a actually a good measure for any integer app that is small enough to run in large cache... Image processing, oil companies for their simulations, cad... they all act very similar to compile benchmark - if a compile is twice as fast, a software image rendering is usually twice as fast too :-)

Now hold on. Did I just here you state that the ALU is used for CAD, and most scientific and engineering applications?

Now I'll grant you that _some_ image processing _can_ be 32/64-bit integer transforms. But most high-quality, 64-bit precise transforms use the FPU. Not even Intel's SSE (on the P3/P4) is good when you need such precision.

In fact, I don't know how many times I've caught errors because someone threw the "-O3" switch on "-march=p4". You do _not_ want to use SSE for scientific computing -- it's for when you want performance in exchange for loss of precision (let alone accuracy).

If you run databases or so, then yes, your comment is valid.

Thank you.

Fair enough :-)

And people say *I* go off on tangents.

No offense, you were already in 2 posts before I even responded. If you look at other threads, I'm typically responding "OT" after 2-3 people have done so, and I largely see an inaccuracy, or I see a correction very much necessary.

[ My posts on that thread _prior_ had just been to inform people that yes, there _were_ Xeon and Itanium systems with hot-swap capability. ]

That's all I was trying to do in this case. Not "prove you wrong."

-- Bryan J. Smith mailto:b.j.smith@ieee.org