jbritt@shumv1.ncsu.edu (Joe Britt) (05/02/90)
dce@icky.Sony.COM (David Elliott) sez: >Joe, would it be possible for you to do a short explanation of how >accelerators work without saying anything proprietary? For example, >what areas have to be handled specially because of the difference >between the motherboard speed and the accelerator speed? How much >does cacheing help, and how can one write code that will take best >advantage of the accelerator cache? So be it! As most folks know, accelerators replace the existing processor in a machine with a better/faster one. Some accelerators are fancier than others, having features other than simple increased clock frequency. And then some accelerators don't increase the clock speed, but just add a cache (like the Dove Racer and Orchid MacSprint). So what's the scoop? Well, the main problem is that the motherboard of whatever the accelerator is connected to has been designed to run at the old speed (big surprise, huh? :-). If the accelerator is of the faster processor variety, then the speedier CPU spends a lot of time waiting on the slower motherboard peripherals and memory. The only place where you get improved performance is from CPU intensive operations. The processor is executing instructions faster, but it still takes just as long to get them in and out. Internal caches (like on the 68020 and '030) help a lot here. Some designers add faster RAM to the accelerator, and leave it as an option to ignore the motherboard RAM. With the faster RAM, instructions and data now flow in and out of the processor faster (as well as execute faster) and a significant speed improvement follows. The Dove SE MaraThon '020 accelerators do this. In addition to faster RAM, you might want to add "wider" RAM. When you try to slap a processor with a 32-bit data bus onto a motherboard with only a 16-bit data bus (like plugging a 680x0 onto a 68000), you lose some (a lot) of possible horsepower. All Mac programs spend a great deal of time in ROM, which is slow and only 16 bits wide. The Dove SE MaraThon '020 boards alleviate this problem by pulling ROM into faster, wider RAM on startup. (as an option) Problems..:-( Naturally, since you are sticking something foreign into your Mac, the possibility of problems comes about. The guy building the accelerator also runs into more than a couple of problems...problems the user should never have to worry about. Just about all of them are timing problems, mainly due to assumptions the code in the ROMs make about how long it takes to do certain things. In some cases, the ROMs try to get around this problem (thoughtful of them :-) by measuring the speed at startup and setting some nice timing constants that other programs can use. Problems arise when you go to Warp 7 a little later, though. The ROM has set up the constants for one speed, then your Acme XP-7 Super-Stealth 600MHz accelerator comes along and makes them totally invalid. All this means it's Muckin' Time, as the engineer tries to figure out wha'happen... Cache cards take a different approach- don't increase the clock rate, just keep the most recently used chunks of code/data in some very fast RAM. Most programs tend to be in loops a lot, and if you can fit the loops into some really fast RAM, you'll see a definite speed improvement. The 68030 has 256-byte instruction and data caches built in, and they _do_ make a difference. Any very CPU intensive code that fits in the caches wil cook right along. It's important to note that Dove isn't marketing the Plus/SE '030 as an accelerator. I think that the ads aren't explaining this well enough...The main idea behind it was to give the user a PMMU and (soon!) a 68881/2. We sent the guy at Connectix one of the boards, and he got Virtual running on it. It's pretty neato to see a Plus that thinks it has over 10 megs of RAM! Multifinder Mecca... Well, I don't know if this was too simplistic or not. Let me know, and let know if you want to know any more. I do have to be careful what I say though, or the legal eagles will come down and swoop me away... Dave Elliott continues: >Oh, wait. You're at NC State. Maybe you can tell us how to accelerate >a cow as well? (Sorry, my "Heels" took over my brain for a second ;-) Sheesh. Actually, cows are more difficult to accelerate than Macs. But then again, a cow _weighs_ more than a Mac right? :-b <in the background: ...so if she weighs more than a duck...she's a witch!...> -- +----------------------------++---------------------------------------+ | jbritt@shumv1.ncsu.edu || Give me a free Mac II. Please? | +----------------------------++---------------------------------------+
rkm@PacBell.COM (Richard Mossman) (05/03/90)
>dce@icky.Sony.COM (David Elliott) sez: >It's important to note that Dove isn't marketing the Plus/SE '030 as an >accelerator. I think that the ads aren't explaining this well >enough...The main idea behind it was to give the user a PMMU and (soon!) >a 68881/2. We sent the guy at Connectix one of the boards, and he got >Virtual running on it. It's pretty neato to see a Plus that thinks it >has over 10 megs of RAM! Multifinder Mecca... Am I being naive or does this give me (a poor worker bee) the opportunity to upgrade my Plus to an '030, run Connectix (therby loading up oodles of SNDs, DAs, Fonts, etc.); clip (tape, solder, glue) a TPD interface to the machine AND save gobs of $$$. Or, like I asked, am I just being naive? Of course, I will then be looking forward to being able to pop in Sys 7 and keep on Computing. -- Richard K. Mossman {bellcore,sun,ames,pyramid}!pacbell!rkm 415/823-0974 ========================================================================= The worst day skiing ... is always made better by discount lift tickets.