hyc@math.lsa.umich.edu (Howard Chu) (10/28/89)
Just for my own curiosity, I downloaded the QuickST 1.6 demo and checked its speed with QuickIndex 1.5 (geeze. Sorry Darek, I've used this QuickIndex a number of times and still haven't registered with you. A check is on its way tho, honest!). So, some numbers from my Mega 4 with TOS 1.4 in EPROMs, in case anyone else is curious... Abbreviations: A - no accelerators B - BLiTTER only T - Turbo ST 1.6 only TB - Turbo ST 1.6 + BLiTTER Q - QuickST 1.6 demo only QB - QuickST 1.6 demo + BLiTTER For monochrome resolution: A B T TB Q QB BIOS text 91 99 298 298 307 307 BIOS string 86 91 1210 1210 1081 1081 BIOS scroll 98 129 136 142 132 132 GEM draw 100 135 211 233 158 177 ____________________________________________________________ For medium resolution: A B T TB Q QB BIOS text 92 98 320 320 246 246 BIOS string 87 90 1364 1364 1373 1373 BIOS scroll 98 126 132 138 129 129 GEM draw 99 130 175 211 115 156 ____________________________________________________________ For low resolution: A B T TB Q QB BIOS text 94 106 415 416 221 221 BIOS string 89 97 851 856 443 443 BIOS scroll 98 125 133 133 128 128 GEM draw 99 149 182 235 101 151 Dunno why column A doesn't just say "100" all the way down the line. Oh well. Also interesting to note that when the blitter makes a difference with Turbo-ST, the difference depends on order of activation. If you turn on the blitter first, and then TurboST, you'll get about 6% more speed than vice versa. (Turn TurboST on first, then the blitter...) [This quirk isn't visible above, obviously.] Kind of interesting... Makes ya wonder how you could put up with text operations on an ST just out of the box. Things really do look real real slow, until you use one of these add-ons. Pretty hard to judge which one to use, though. I'm kinda bummed, myself. Neither of these two programs work with non-standard screen sizes, so my Overscan switch is idling in the normal position. Oh well. QuickST1.4 worked OK with Overscan. I wonder why the restriction now? Also TurboST doesn't seem to like PinHead. My system freezes when PinHead is in my AUTO folder, if TurboST is loaded. I s'pose I wanna comment on the numbers themselves, since I've gone to all the trouble of typing them in. Interesting to note that the BIOS text operations (Bconout, Cconws) done by both programs completely ignore the blitter. Also interesting that TurboST gains a few percent in scrolling when the blitter is active, but QuickST ignores it there as well. (Interesting, but not a very significant difference, to be sure.) Also odd how the different resolutions alter the speedups. Like, in monochrome, the blitter accelerates GEM draws by 35% all by itself. But, it only speeds up both TurboST and QuickST by about 20%. Then, in medium, the blitter accounts for only 30% by itself, but speeds up both TurboST and QuickST by about 40%. Funky Stuff there... Hm. Does anyone else out there really care about this stuff? I'm pretty curious to see some numbers for the various hardware accelerators now... -- -=- PrayerMail: Send 100Mbits to holyghost@father.son[127.0.0.1] and You Too can have a Personal Electronic Relationship with God!
w-darekm@microsoft.UUCP (Darek Mihocka) (10/31/89)
In <10087@stag.math.lsa.umich.edu> hyc@math.lsa.umich.edu (Howard Chu) writes: >Just for my own curiosity, I downloaded the QuickST 1.6 demo and >checked its speed with QuickIndex 1.5. Howard, the Quick ST 1.6 demo was not intended to be Quick Indexed. It is slower than the actual release version of Quick ST 1.6, and so the numbers are misleading. The purpose of the demo was to satisfy all the people who said they'd register if the Quick ST 1.4 wasn't so buggy. Well, it's no longer buggy, so I should be seeing a few more shareware donations, right? <grin> The demo is still faster than all previous versions of Quick ST and that other program, other than version 1.6 of the other program. The release version is a tossup, because it is faster than that other program in some cases, and slower in others. On average, pretty even, and half the size of that other program. Quick ST even does pretty well with that SPEEDTST program they provide with that other program. I do not mention the name of that other program cuz it is overpriced and its advertizing is misleading, and it doesn't need any more free publicity. ... long list of benchmarks omitted. ... >Dunno why column A doesn't just say "100" all the way down the line. You must have had some other TSR or desk accessory installed, like GDOS or Hotwire. Any program that hooks the TRAP vectors will reduce the speed. ... >Kind of interesting... Makes ya wonder how you could put up with >text operations on an ST just out of the box. Things really do look >real real slow, until you use one of these add-ons. Pretty hard to >judge which one to use, though. > The blitter is inefficient at plotting text. There is too much setup work to do to plot even a single character. >I'm kinda bummed, myself. Neither of these two programs work with >non-standard screen sizes, so my Overscan switch is idling in the >normal position. > Quick ST automatically supported Hyperscreen up to version 1.5. Since I only knew of two other Hyperscreen users here in North America, I did not include it in the demo. I was in contact with the author of overscan, and I suggested that the best place to put Quick ST support for Hyperscreen would be in Overscan itself, since both programs run from the AUTO folder, and so could be merged. Since Quick ST source code was posted, as well as Overscan source code, I'm sure some industrious person out there will merge it sooner or later. Quick ST can still support custom screen resolutions, since the code is resolution independent. All I need is your desired screen resolution and your shareware check <grin> and you'll have a Hyperscreen version. Since you're on usenet, custom versions can be sent out same day. We don't force you to mail in a disk and $5 every time there's a new file like with that other program. >I s'pose I wanna comment on the numbers themselves, since I've >gone to all the trouble of typing them in. Interesting to note >that the BIOS text operations (Bconout, Cconws) done by both >programs completely ignore the blitter. Already explained. ... >TurboST gains a few percent in scrolling when the blitter is >active, but QuickST ignores it there as well. (Interesting, but >not a very significant difference, to be sure.) Also odd how the >different resolutions alter the speedups. Like, in monochrome, >the blitter accelerates GEM draws by 35% all by itself. But, it >only speeds up both TurboST and QuickST by about 20%. Then, in >medium, the blitter accounts for only 30% by itself, but speeds >up both TurboST and QuickST by about 40%. Funky Stuff there... Quick ST doesn't use the blitter for scrolling. If you can tell the difference with and without a blitter, you should be a baseball umpire or something. Quick ST owes its small size partly due to the fact that it doesn't waste code trying to optimize something to gain an extra 1% or 2% in speed. The down side is that Quick ST will score lower on benchmarks. But it is silly to worry about a few percentage points. Your eye will not notice the difference, and so why why waste code on minor optmizations. For example, say your car is badly tuned, and the tires are a bit flat. You take the car into Quick Mechanics and have it tuned. Suddenly it's a lot faster and fuel efficient. You notice that. Then you add some air to your tires (which Turbo Mechanics charge you $50 to do). You probably won't notice a difference until you actually measure your exact milage, and then it'll be something small. You see the point I'm making. ... >Hm. Does anyone else out there really care about this stuff? >I'm pretty curious to see some numbers for the various hardware >accelerators now... Hardware accelerator results were compiled by a few different people a few months ago. From the results I saw, and from my own benchmarking, the Turbo 16 board by Jim Allen is by far the best accelerator. And you don't need Quick Index to see it. The other accelerators, like CMI, JRI, Datafree, etc, aren't worth the paper they're advertized on, unless you're into spending $100 - $300 for a 10%-20% speed increase in most programs. The Jim Allen board will give you about 50%-60%, possibly more during number crunching. Definately a good buy. No, he didn't pay me to say this. I saw his board and tried it and it's great. Like I mentioned, Quick ST source code is available on Genie and Compuserve and was probably somewhere on here a few months ago. If you have any questions about it, either look at the source code or ask me or Ignac. You don't need to guess. ------------------------------------------------------------------------------ Darek Mihocka ST Xformer II CIS: 73657,2714 Box 2624, Station B Quick Utilities GEnie: DAREKM Kitchener, Ontario MegaBlit SSG SPX DELPHI: DAREKM N2H 6N2 Shareware, not Vaporware BIX: darekm Canada Quick ST 1.6 is out! CheapNet: ...!uw-beaver!microsoft!w-darekm (519)-747-0386 A mind is a terrible thing to waste, so JUST SAY NO TO TOS. Opinions expressed are my own and not those of anyone not named Darek Mihocka. ------------------------------------------------------------------------------