From: dd@mips.com (David DiGiacomo) Newsgroups: alt.sys.sun Subject: Re: wanted: any info on Sun RoadRunner Message-ID: <8533@spim.mips.COM> Date: 19 Sep 91 18:26:38 GMT References: <18353@life.ai.mit.edu> <7827@autodesk.COM> Sender: news@mips.COM Lines: 27 Nntp-Posting-Host: slack.mips.com Originator: dd@slack.mips.com In article <7827@autodesk.COM> marc@Autodesk.COM (Marc Ramsey) writes: >As one of the 386i developers, and as the one who should most directly >take the blame for the "PAINFULLY" slow sunview implementation, I'd like >to make a few comments. Every benchmark we ran indicated that the >overall performance of sunview on the 386i exceeded that on the Sun3, >and was comparable to the 4/110 (the SPARCstation came later). One of >the main problems was that the Sun3 MMU provided far superior performance >to the internal 386 mechanism. This meant that things ran great as long >as you didn't have to page, but would bog down somewhat if you did. The >other big problem was the "clever" engineering trick of putting the cache >on the memory boards, which meant that all frame buffer accesses were >totally uncached. 1. Frame buffer accesses are totally uncached on *every* Sun workstation. 2. The 386i looked bad to users because the 3/60 and 4/110 color frame buffers had an overlay plane, but the 386i color frame buffers did not. I believe that the performance tests done by Sun ECD did not use the overlay plane, which was arguably "fair", but not representative of typical user experience. 3. There were some minor problems with the benchmarks (at least the ones that I got to look at) which tended to handicap the non-386i machines. I don't think this was as significant as the overlay plane issue though. -- David DiGiacomo, MIPS Computer Systems, Sunnyvale, CA dd@mips.com