Same here - Athlon XP on old mobo. Jan On Thu, 2004-05-06 at 09:20, Steve Harris wrote:

Quoting Russell Hanaghan <hanaghan(a)starband.net>et>: I know it's probably heresy for a Linux user to use Intel, but in a value for money comparison THESE days, a P4 is money well spent. They run cooler, support SSE/SSE2 (the latter I've used to brilliant effect in an experimental anti-spam product which does wire-speed connection monitoring + diversion), and has higher memory + subsystem bandwidth with the FSB800 dual-channel chipsets (865PE/875). When you couple the Intel CPU with the (FREE!) Intel Compiler, you really have a winner hands-down. Rebuilding the OpenSSL library with ICC gave me a 35% performance advantage over the best equivalent settings in gcc-3.3 in the Pure C crypto cores, to name one small example. With the very small number of exceptions if you hand-pick the low-end Thoroughbred cores, run so bloody hot - they are little nuclear power plants. Yes, the new top-end Prescott P4s top them in heat dissipation, but those aren't necessary for your application. Quoting Dave Phillips <dlphilp(a)bright.net>et>: If you were tempted to go AMD, I'd go for the Athlon FX-53, but that's PRICEY. And overkill, I suspect. At least the Athlon64s have full SSE/SSE2 though. I suspect you're the victim of a bad build script, but there ARE genuine performance advantages to using a Pentium 4 if you build your own binaries. Motherboard: ============ The Asus P4C800-E Deluxe if money is no object, but the Asus P4P800 (or P4P800 Deluxe if he needs/wants onboard Firewire) will give you the same performance for a bit less money. The quality of these motherboards is EXCELLENT. Even though the P4P800 uses the i865PE chipset, it supports "PAT" for most of the same memory bus optimisations the more expensive i875 chipsets (in the P4C800) support. CPU: ==== P4-2.8Ghz-800 is probably the sweet spot (in North American markets) In the Land of Mordor, where Quality lies, =MB= -- A focus on Quality.

No, but it doesn't need to be if you are running code w/o reselling its binaries. The licence Intel mandates for the "Evaluation Licence" of Linux compilers is such that only the resale of the output of the compiler is limited, if you don't have a paid licence. Running the code on my own production servers is not prohibited by the Intel licence, for example. You aren't running "100% pure GPL" code if you use nvidia's graphics drivers for X11, or various companies' audio drivers. Even MySQL AB acknowledges using ICC is good for a significant boost in performance, when targetted appropriately. In vero, Qualitas, =MB= -- A focus on Quality.

Absolutely. They don't "collide" compiler-driver-wise. ICC installs into /opt in fact, by default. You can even build completely-static binaries so you can always be sure the faster Intel math libraries are being used. You can also static only the Intel mathlib override portions, and leave the app shared-linked against the "normal" glibc stuff. My kingdom for Quality, =MB= -- A focus on Quality.

On Thursday 06 May 2004 16:44, Malcolm Baldridge wrote: Does this imply that AMD no longer has a big advantage in floating point performance over intel? When I bought my last motherboard a few years ago, an Athlon 800Mhz, floating point was supposedly significantly faster on the AMD chips, and since I do a lot of realtime synthesis, going with AMD was a natural choice. Have things changed since then? Larry Troxler