Slashdot Mirror
Athlon XP1900+ -- Faster Than A 2GHz P4?
doormat writes "AMD releases their AthlonXP 1900+ Processor today, thats 1.6GHz. And it seems like its enough to topple the P4-2.0GHz, even in Quake 3 Arena!! AMDMB has a review of it." Ian Bell points out an AMD press release on the new processor. I love watching my old Athlon get slower every day ...
One thing I've noticed over the last year or so when talking to non-techie friends/family is that many people with relatively little knowledge of IT as a whole are starting to realise that the processor speed, however it's being measured, is far less important than the vendors want them to think.
The end result of Intel and AMDs battle of "my processor's faster than your processor" seems to be that people are saying "I don't care" - as they realise that there 'obsolete' PII is actually perfectly capable of doing all the things they use their PC for and that only graphics people and the hardest of hardcore gamers actually need 1.5 to 2GHz.
Nice editorial work guys.
I saw a review comparing an Athlon 1800 and a 1900.
I didn't see a single thing in there that mentioned the P4 being outperformed or toppled.
Just unsupported speculation.
Hi.
Large matrix multiplies are in fact used as benchmarks by many sites, and are part of specFP, sysmark, and other benchmark suites, many of which you'll see quoted (though not specifically the mat mult part) in these reviews.
The problem is that matrix multiply only tests your floating point/simd unit and your i/o bandwidth. Not a very comprehensive test, and unless you actually plan to do large matrix multiplies, quite synthetic.
As for your web page serving idea, it's called specweb, and anyone who is catering to or buying in the web server market cares a lot about this benchmark. It is a more comprehensive test than just multiplying matricies, but still only targets certain aspects of the cpu (I/O again, cache size).
Games actually make a good addition to these benchmarks. A modern game engine can tax a cpu a great deal, and will use a mix of integer and floating point applications, plus put pressure on the memory subsystem. If the performance isn't limited by the graphics card, then you can use games quite effectively as CPU benchmarks.
It's funny that you mentioned "real". If you're running sicentific apps that multiply lots of matricies, then matirx multiply is "real". If you're running anandtech.com, web serving is "real". And, if you're a gamer, games are real and the performance you see in them is what is the "real" performance limited.
It almost sounds like you want the "max", as generated through some kind of synthetic test. As in the performance you'd never, ever get in a actual application that you'd want to run once you'd bought the system. Which was how it used to be done, and it sucked, so everyone stopped. Let's not suggest we go back, hm?
The enemies of Democracy are