Re: Nocona [Intel 64-bit cpu timing]
tomstdenis_at_gmail.com
Date: 08/26/05
- Next message: tomstdenis_at_gmail.com: "Re: The Chinese MD5 attack"
- Previous message: Paul Rubin: "Re: Nocona [Intel 64-bit cpu timing]"
- In reply to: Paul Rubin: "Re: Nocona [Intel 64-bit cpu timing]"
- Next in thread: Paul Rubin: "Re: Nocona [Intel 64-bit cpu timing]"
- Reply: Paul Rubin: "Re: Nocona [Intel 64-bit cpu timing]"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ] [ attachment ]
Date: 25 Aug 2005 17:25:54 -0700
Paul Rubin wrote:
> Again, that's just the particular implementation. It's going to take
> a while to deploy much 64-bit software in the corposphere, and by the
> time it's out there, newer CPU's with more address bits will also be
> available.
My point is there shouldn't be a market for this core. It's clearly
sub-par. And this isn't the sort of thing "minor tweaks" are going to
fix. Face it, ALU performance still counts. Sure the P4 design has a
decent FPU but that's not the be-all of computing [and even then the
AMD64 design can hold its own].
And frankly who cares about having gobs of cache. I can read system
memory at 4.5GiB/sec [memtest86+ reported speed]. That's thanks to the
128-bit memory interface which the P4 designs lack [they use
interleaving to fire DDR memory at QDR rates]
> As for P4 vs PM, the P4 is far less power efficient, but that's ok for
> servers, where plenty of power and cooling are available. The P4 is
> faster than the PM because its simpler pipeline stages are capable of
> much higher clock rates than the PM.
This is hard to judge though, you need two ratios the frequency and the
cycle counts. If you have 2x the frequency but take 3x the cycles
you're not ahead and you're just burning energy to do "little really
fast!"
As in the case of the Nocona vs. AMD64. Even with a clockrate of
600Mhz faster it's still slower [walltime] at doing most tasks
[compiling, AES, RSA, etc] then the AMD64.
The Pentium M last I tested was close to the AMD32 [Athlon-XP] in
performance. IIRC a 1024-bit exptmod took roughly half the cycles of
my Prescott when on a Pentium M. The Pentium M can hit 2Ghz and the P4
tops out at 3.8Ghz, that means you have <2x clock boost and >2x the
cycles.
The P4 [sad to say] makes a good server processor because it's good at
two primary tasks
1. Data shuffling
2. FPU processing [which doesn't help servers...]
So if you need to do string operations or what not the P4 is probably
faster... however if you need to do any crunching then you're better
off with just about any other processor.
Heck, a 90$ Sempron can beat a P4 at RSA computations and take half the
power while doing it ;-)
> By the way, what AMD64 laptop are you using? Does it have a Turion
> cpu? I'm sort of watching the market in those, but all the reasonably
> priced ones I know of have graphics chips with secret instruction
> sets, so can't run non-closed-source X servers.
Really? I thought HP had some with nvidia cards which should work fine
with xorg.
Anyways, my laptop is a two year old Presario 2100 [I think...] it's a
32-bit Athlon-XP-M 2400+ [512KB L2 cache, 1.8Ghz clock]. My other
boxes are a AMD64 4200+ Dual-core [2x512KB L2, 2.2Ghz], P4 Prescott
[1MB L2, 3.2Ghz] and the new one is a Nocona 820 [2x1MB L2, 2.8Ghz,
EMT64].
Of all of them the AMD64 box has the highest memory bandwidth at over
4.5GiB. The Prescott tops out [guestimating] at probably around
2-3GiB/sec tops, the nocona is probably in the 3-4 range [it's using
PC4300 memory in "interleaved" mode] and the AMD32 pathetically tops
out at <500MiB/sec.
All the boxes run Gentoo Linux :-)
Last I read, the Turion was wedged between the mobile AMD64 and the
Pentium M in terms of power consumption. However, unless they crippled
it the power EFFICIENCY should be higher than the Pentium M.
Tom
- Next message: tomstdenis_at_gmail.com: "Re: The Chinese MD5 attack"
- Previous message: Paul Rubin: "Re: Nocona [Intel 64-bit cpu timing]"
- In reply to: Paul Rubin: "Re: Nocona [Intel 64-bit cpu timing]"
- Next in thread: Paul Rubin: "Re: Nocona [Intel 64-bit cpu timing]"
- Reply: Paul Rubin: "Re: Nocona [Intel 64-bit cpu timing]"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ] [ attachment ]
Relevant Pages
|
