Plamen Tanovski <[EMAIL PROTECTED]> writes: > what is the best hardware nowadays for tetex? I need max > performance, since I have to compile 23M of TeX source very quick. > > My thoughts: > > 1. CPU > > I think faster is better. But how about 64bit v. 32bit?
Code compiled for Opteron tends to have more registers for allocation IIRC, and so tends to be faster in general. That's more or less an accidental side effect of the crippled 32bit architecture of Intel: 64bit should not buy you any advantage for TeX. > Is a RISC better then CISC? Hyperthreading, Multiprocessing would > not increase the speed of TeX, wouldn't they? Alpha CPUs are good at > floating point operations, but TeX doesn't perform any? TeX performs floating point operations, but very sparingly. I doubt you could even significantly measure a difference between an 80386 with arithmetic coprocessor and one with coprocessor emulation when running typical TeX code. TeX manages memory using unsorted linked free list, those tend to fragment memory. So it is an advantage if you have large cache memory and if cache misses are not too expensive. > 2. Compiler > > For Intel, Alpha etc. there are optimized compiler. Are there any > benefits, if I compile the teTeX sources with these? For integer performance, you can expect some improvements. However, with the latest gcc, they would probably not be overly exciting. > 3. OS > > Well, I prefer Linux, but are the BSDs, Solaris etc. better > regarding TeX performance? For typical TeX tasks, the operating system will only have marginal influence on performance. > 4. Storage > > Do I need SCSI? If you get into heavy swapping or have massive parallel compilation jobs, SCSI can be advantageous. Of course, for heavy swapping situations, more RAM can be much more advantageous. > Is ramdisk better? TeX basically processes its files sequentially. A RAM disk does not significantly speed up this access patterns, unless you are running massively parallel jobs. -- David Kastrup, Kriemhildstr. 15, 44793 Bochum
