13.
poster 22:
".. Like to see more Engineering apps, as my tutor doesn't believe me that Athlon64/FX/Opteron craps on Intel's P4 and Xeons in Engineering apps. (The raw number crunching without SSE2 or HT support)."
I absolutely agree. You can see this from this article's Mathematica as well as compiling benchmark (which can be indicative of how well an architecture handles branchy code -- used in a lot of academic apps like AI). In fact, just look at any SiSoft Sandra benchmark for the P4 and notice the NON-sse2 performance figures i.e. x87 floating point and they're absolutely terrible.
All said, this is why AMD seems more popular among the HPC and linux/unix crowd where cross platform code takes precedence over cpu specific optmizations. (intel's compiler can boost speed by 30% by doing scary optimizations like loop unrolling with sse2, and shifting a whole lotta instructions around to fit the netburst architecture and prevent stalling with that long pipeline, but the result is a binary that won't run on anything else of course)
It's ironic but fortunate that they're finally going back to the P3 based Pentium M for the future desktop. Also of note, is that Itanium 2 is like an opteron on steriods (ignoring the difference in ISA for a moment) Ace's hardware had a benchmark showing an Itainum 2 at just 1Ghz running PovRay ran twice as fast as a 2Ghz Opteron which in turn ran about 50% than a 3Ghz Xeon. Oh, and nominal implementations of Itaniums have just an 8 stage pipeline :)
[
Posted by: jv

|
Date: 06/19/04 12:41:48 PM]