home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.arch
- Path: sparky!uunet!pgroup!lfm
- From: lfm@pgroup.com (Larry Meadows)
- Subject: Re: Call for Opinion: Viking or i860
- Message-ID: <Bs9E2o.3v8@pgroup.com>
- Date: Fri, 31 Jul 1992 15:08:48 GMT
- References: <5678@nosc.NOSC.MIL>
- Organization: The Portland Group, Portland, OR
- Lines: 29
-
- In article <5678@nosc.NOSC.MIL> wolfgang@sunspot.nosc.mil (Lewis E. Wolfgang) writes:
- > What do you think? Upgrade the cpu with Viking?
- >Or purchase an equal number of i860 based array processors?
-
- The sparc will certainly be substantially more versatile, as there are
- no longer any i860-based workstations. So if you buy an array
- processor, you will probably have to go through some substantial effort
- porting your code.
-
- The performance difference depends on how vectorizable your code is.
- If it is written in Fortran and has lots of nice vector loops you could
- see pretty high mflops; on the other hand, if it is written in C and
- spends tons of time chasing pointers, the i860-XR will not work as
- well, since it has such a small cache and there won't be opportunity to
- fill the FP pipelines.
-
- Remember that the i860 is not superscalar, and that the only way to get
- peak floating point performance is using explicit pipelined fp
- operations and these really difficult dual operation instructions; the
- only way to get integer + fp at the same time is in dual mode, which
- results in halving instruction cache bandwidth.
-
- If, however, you are willing to spend the time porting and tuning your
- code, and if you wait for an i860-XP based processor, possibly with
- external cache, then I'd expect that you would see substantially better
- performance than the Viking.
- --
- Larry Meadows The Portland Group
- lfm@pgroup.com
-