Feuer wrote:
> I will soon buy a computer. Could someone give me an idea what issues
> there are buying x86 vs. Alpha, particularly WRT Linux? I particularly
> desire information regarding
>
> price$$$$$$$
The best way to do this is to go to the vendors. There's a huge vendor
list at http://www.alphalinux.org/
> hardware reliability
Haven't had any problems here.
> kernel support--stability on platforms, device driver availability,
> availability/quality of support
The kernel in my experience is outstanding. However, XFree86 has not been
quite so strong. The good news is that behaviors which crashed 3.3.2
(remote Netscape drag'n'drop, xdvi zoom, remote WordPerfect Save As) don't
seem to crash 3.3.3.
> development support--libraries, compilers (how well does gcc handle x86
> vs. Alpha, etc.)
Mostly good. Some programs don't compile -O2 with egcs 1.1.1, so you have
to set file-specific C flags. The big hangup is that a LOT of big codes
cast between pointers and ints, which are of different size on 64-bit
processors, and unlike DecUnix compilers, gcc/egcs doesn't have a
workaround. This and other pointer size assumptions break a lot of codes.
> commercial/binary programs---have programs like Netscrape, Corel
> WordPrefect, Mathemagica, etc. been ported to Linux/Alpha?
The only one I know of is Applixware, which has received rave reviews on
[EMAIL PROTECTED] (a somewhat more active list than this, btw). As for
the rest, it will probably take them quite a while because of 64-bit
issues.
> I am also somewhat curious about the characteristics of the Alpha
> processor, such as: number of registers and their types, instruction set
> characteristics, cache, etc. etc.
The only one I can speak directly to is cache, which will depend on the
type of processor/board you get. The commodity 164SX board has 16K L1
cache on-chip and 1 MB L2 off chip. The 164LX and 164UX both have 16K L1
and 96K L2 on-chip with very low latency, and 2-8 MB off chip. The L1
cache is split evenly between instructions and data.
By the way, what will you use this for? For floating point calculations
and perhaps web/file serving, Alpha rules- optimized matrix routines run at
500-750 MFlop/s on my 533 MHz 164LX. For other tasks, get a high-end
Intel- there's way more commercial software, open source is usually stable
there first (e.g. the GNOME desktop at version 0.99 is an enormous CPU hog
on alpha for some mysterious reason), and the integer performance is on par
with Alpha at a similar price.
Hope this helps,
-Adam `Cold Fusion' Powell, IV http://www.ctcms.nist.gov/~powell/ ____
USDoC, National Institute of Standards & Technology (NIST) |\ ||< |
Center for Theoretical and Computational Materials Science | \||_> |
