home *** CD-ROM | disk | FTP | other *** search
- Xref: sparky comp.os.ms-windows.misc:4421 comp.sys.ibm.pc.hardware:29154
- Newsgroups: comp.os.ms-windows.misc,comp.sys.ibm.pc.hardware
- Path: sparky!uunet!stanford.edu!CSD-NewsHost.Stanford.EDU!Xenon.Stanford.EDU!dhess
- From: dhess@Xenon.Stanford.EDU (Drew William Hess)
- Subject: Re: ATI Ultra Pro vs. #9GXe
- Message-ID: <dhess.721159474@Xenon.Stanford.EDU>
- Keywords: ati ultra pro #9gxe number nine video accelerators
- Sender: news@CSD-NewsHost.Stanford.EDU
- Organization: CS Department, Stanford University, California, USA
- References: <dhess.721001807@Xenon.Stanford.EDU> <1992Nov6.163141.29490@seas.gwu.edu>
- Distribution: comp
- Date: 7 Nov 92 18:04:34 GMT
- Lines: 53
-
- prall@seas.gwu.edu (David C. Prall) writes:
-
- >In article <dhess.721001807@Xenon.Stanford.EDU> dhess@Xenon.Stanford.EDU (Drew William Hess) writes:
- >>
- >>What really caught my attention was a graph that shows the ATI Graphics
- >>Ultra Pro clocking in at 25.5M WINMARKs (which is what ATI advertises for it),
- >>and the #9GXe getting 40.5M WINMARKs! I find this number amazing.
- >>
- >>I looked hard at the ad to try to find some sort of "fine print" that would
- >>show that Number Nine had done some creative benchmarking to achieve their
- >>claimed 40.5M WINMARKs, but I found no such caveats. The only benchmark
- >>qualifications given were "WINBENCH 2.5 on 486/ISA at 1024 x 768 at 72 Hz."
- >>
-
- > I called Number Nine to check on this claim. They said that the board
- >would not be shipping until the first of 93. Also, that these were preliminary
- >figures for a pre-production board. It also does 64K colors at 1024x768.
-
- Since seeing this ad, I've read somewhere (sorry, can't remember where) that
- accelerator manufacturers are beginning to design their accelerators with
- WINMARKs in mind. In other words, they're building special hardware into
- the accelerator that recognizes the Winbench suite and artificially enhances
- performance. Generally, these are probably enhancements that you couldn't
- use in actual applications; their sole purpose is to bolster their benchmark
- claims.
-
- I'm not sure whether or not this is true, or even possible, but if it is
- this is a very dangerous trend. Things like WINMARKs are probably not good
- indicators of actual performance anyways, but for now it's all the consumer
- has to rate performance of video boards for Windows. If board designers
- start "fudging" their WINMARK claims using specialized hardware that can
- only be used to speed up Winbench, and nothing else, then we'll no longer
- have a believable benchmark. This trend would be akin to the trend in
- compilers on workstations, where some vendors have gone so far as to
- enable the compiler to recognize benchmarks like Dhrystone and perform
- special optimizations for that benchmark alone. Enhancing performance for
- toy benchmarks doesn't buy the consumer anything....
-
- Maybe we should write PC Magazine and others who use the Winbench suite
- to warn them of this possibility. Hopefully they'll develop a benchmark
- that reflects real system performance, say by taking a couple of
- applications and timing how long it takes to perform everyday operations.
- Even if the hardware vendors are not skewing their WINMARK results,
- a benchmark like this one would give the consumer a more accurate picture
- anyways.
-
- > Hope this helps,
- > David Prall
- > prall@sparko.seas.gwu.edu
-
- -dwh-
- dhess@cs.stanford.edu
-
-