Return to BSD News archive
Newsgroups: comp.unix.bsd Path: sserve!manuel!munnari.oz.au!spool.mu.edu!uwm.edu!zaphod.mps.ohio-state.edu!rpi!newsserver.pixel.kodak.com!psinntp!psinntp!l5next!scotty From: scotty@gagetalker.com (Scott Turner) Subject: Re: public S3 design (Was: Free software and ... Diamond products) Message-ID: <1992Oct3.055409.17914@gagetalker.com> Sender: scotty@gagetalker.com Organization: L5 Computing References: <BvGoLF.8nn@pix.com> Date: Sat, 3 Oct 1992 05:54:09 GMT Lines: 57 In article <BvGoLF.8nn@pix.com> stripes@pix.com (Josh Osborne) writes: |> In article <BvCyt0.JG4@immd4.informatik.uni-erlangen.de> Torsten Duwe (duwe@informatik.uni-erlangen.de) writes: |> [...] |> >Doesn't performance depend solely on the main (bus/memory/whatever) clock ? |> >... would be a bad thing if drawing speed was closely coupled to the pixel |> >clock. |> |> With normal DRAM (i.e., not VRAM, not duel ported RAM) if the shifter needs |> to output 65 Mega Pixels per second (assuming 8bits per pixel) 65 Mbyte/sec |> of bandwidth is used to feed the shifter, which is 65Mb/sec you can't use to |> transfer data to the DRAM. |> |> [...] |> >In my opinion local bus is unnecessary if you have an 'intelligent' SVGA to |> >give complex commands to. Only memory to screen bitblt is an issue but that |> >should be solved using more video RAM - hope the '928 is able to address more |> >than 1 Meg. |> |> Well please optimise copy plane of pixmaps with xor on and have enough RAM |> that all of xtanks pixmaps can be resident (I think there is over 1M of them |> once rotated out, if not I know there is over 500K of them). [...] |> |> >[...] |> >>(4) S3 corportation provides a software development kit |> >Yup! They're very DOS-infected but nice people. |> |> I thought they had a windows bias, the 911 isn't very fast unaccel'ed |> (isn't as fast as, say, the ET4000), and most DOS programs arn't accel'ed. |> Not that windows is all theat diffrent from DOS... Interesting discussion guys, but you're both missing the big advantages to using something like local buss or an S3/8514 card. The single biggest thing you need is unbanked access to the video memory. Next is mouse in hardware. Then comes video refresh rate vs VRAM/DRAM. Not having to look over your shoulder to make sure you haven't switched banks or drawn on the cursor make a huge difference in performance. The S3 chip has an on board mouse and having an 8514 interface it has unbanked access to the video memory. But massive bandwidth to the memory doesn't always pay off. I've benchmarked the ATI Graphics Ultra against the ATI GraphicsVantage. The Ultra has over twice the bandwidth of the Vantage and yet is only 20% faster in the windows benchmarks. I did a BLIT port of X11R4 to SysV386 3.2 and using the 82786's on board mouse hardware almost doubled the speed of the server (using the onboard drawing engine also gave it a BIG kick in the pants, too bad the damn thing died, sigh.) Western digital was shopping around a chip earlier this year that was designed specifically to speed up X. It pratically ate X straight (like the 82786) and thus might be an interesting chip to play around with. Scotty