Re: Nocona [Intel 64-bit cpu timing]
Date: 25 Aug 2005 18:42:37 -0700
Paul Rubin wrote:
> Yeah, but right now the max freq of the P4 is such that it can run
> more mips than the max freq'd PM. That's at a horrendous cost in power
> consumption, but there are users who don't care about that.
I don't quite follow you. Sure the P4 can run at a higher clockrate
than the PM. So what? It's instructions per cycle [IPC] count is much
> > For most purposes you spend 95% of the time within a small pocket of
> > memory.
> However, 95% of users don't use these high-end P4's. They are
> targeted among other things at apps like big database servers, which
> are precisely the 5% of purposes that are extremely ram intensive.
So strap an 8GiB/sec bus on the PM and call it a day :-). A read from
the P4 L1 takes 3 cycles, at 3.8Ghz that's [say SSE2 reads]
19.327GiB/sec max. If your main memory peaks at 4GiB/sec then what's
the point? Furthermore if your storage peaks at <500MiB/sec then
you're really wasting power.
> > If you want a shell ... all you have to do is ask.
> I may take you up on that for experimental/testing purposes, but by
> "hosted" I mean running cpu-hogging production apps 24/7 using tons of
> internet bandwidth. I don't think you're in a position to offer that
Well on the Intel boxes you can hog time. On the AMD64 you'll have to
let me know before you run anything intense. As for the bandwidth...
nope, just a cable modem. All three boxes have 1GiB of memory so
they're not exactly a "super computer cluster". But if you want to
benchmark algorithms or whatever they're perfectly fine.
If your intention is to test something like a web application then your
SOL as I don't run web servers locally at all.
Anyways, let me know if your interested via private email and I'll set
you up with an account. Fairly simple standing rules: Don't fill the
HD and don't make the machine blow up. :-)
> > Good call. Wait for a dual-core turion? hehehehe..
> Nah, for a laptop I don't much care about dual core.