Re: Kernel support for specific cpus
On Tue, May 15, 2001 at 08:30:27AM -0400, Raul Miller wrote:
> Todd Finney pointed me at a site documenting some kernel benchmarks.
>
> These benchmarks show that on a pentium III, lmbench doesn't gain any
> improvements over the performance of kernels which didn't offer the
> pentium III optimizations:
>
> http://euclid.nmu.edu/~benchmark/
That's an incorrect conclusion if I read the report right:
: The format of the configuration has changed slightly from kernel to
: kernel, so we chose the options that were most similar. As we compiled
: new kernels we saw new options that were not available before. Since
: we are using a fairly new computer and some old kernels, we could
: not always choose the exact options for our hardware. For example,
: we did not have the choice of a Pentium III processor until kernel
: version 2.4.0-test4. Before that, we chose the closest processor: the
: Pentium Pro. For the most part, we were able to get the configurations
: nearly identical. In configuring we also tried to have as few modules as
: possible, especially for the kernels that were compiled on our secondary
: computer with gcc v. egcs 2.91.66. For those kernels we did not want
: any modules. We then transferred the image file to our primary computer
: to boot from. In the end, we had nineteen kernels, all configured as
: closely as possible.
So what they're saying is that they switched from PPro to PIII at
2.4.0-test4. In reality, there is no difference whatsoever between the
two choices which explains why no difference was observed in their
performance.
This has no bearing on the difference in performance between a kernel
optimised for PPro and one optimised for 386 running ona PIII.
--
Debian GNU/Linux 2.2 is out! ( http://www.debian.org/ )
Email: Herbert Xu ~{PmV>HI~} <herbert@gondor.apana.org.au>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt
Reply to: