VanLUG Email Archive
> On Sat, 17 Oct 1998, Dave Martindale wrote:
>
> > It's my impression that Pentium floating-point performance is still better
> > than the non-Intel competitors. My wife sometimes fires up simulation jobs
> > in Matlab that run for hours - mostly matrix arithmetic, probably all
> > floating point. That would be a good reason to care about FP performance.
Just to put in my $.02:
My IBM 120+ gets, using the byte benchmark, the following scores:
integer: 2.0 * a_pentium_90
fp: .7 * a_pentium_90
As you can see it's night and day. With the advent of 3d cards
I wonder if this advantage will disappear?
Of course this is when I had it o'clocked to 2x66MHz. It was
pretty stable when using int ops. However quake lasted about
2 minutes on average before a fp error occurred. Linux
didn't crash, though!
> As an example, a Alpha PC164SX motherboard, with a 533 MHz CPU and
> a meg of cache, offers a SPECint95 of 12.4. This is darn near the
> 12.7 SPECint98 of a 333 MHz Pentium II on an Intel DK440LX motherboard
> with 1/2 MB of cache. However, the Intel has a SPECfp95 of only
> 9.25, whereas the Alpha has a SPECfp95 of 16.1. It's the general
> case that, with an Intel and Alpha of similar integer performance,
> you'll get one and a half to twice the FP performance from the
> Alpha.
What about similar cost? :>
+-----------------------------------------------------------------------+
| R Garth Wood | <insert witty comment here> |
| | -R G Wood |
|
| |