Date: Fri, 1 May 92 13:57:05 MDT From: sharmony@nova.ta52.lanl.gov (Stephen C. Harmony) Subject: [*] PowerCache performance in LC Dear Moderators: I think the following report on the DayStar PowerCache performance in LC's would be a useful addition to the info-mac/reports directory. The following explanation of how the PowerCache minimizes the effect of the LC's 16-bit data path was written by Steve Tuttle, the Online Resources Mgr for DayStar Digital. I will submit an expanded report to the info-mac archives as soon as Steve forwards the results of the benchmark tests on the unenhanced LC and IIci, but the results as presented give a good indication of the performance difference between a PowerCache-enhanced LC and IIci. "Altho it would appear intuitive that due to the 16 bit path on the LC you are going to be forced to run slower, it is not necessarily the case. You see, all Macs have some sort of delays inherent in their design: The II,IIx,IIcx, & SE/30 all have 120ns RAM and require 2 wait states to be added to the memory cycle in order to ensure the memory is stable enough before the processor accesses it. On the IIci they go to 80ns RAM & 1 wait state. So all we're really talking about here is just another built in delay in the Mac's architecture that keeps the processor from running as fast as it would normally be able to run. "Our solution? Static RAM Cache. What we do is take a 32K cache of 25ns static RAM and build it into our accelerator boards. What happens is that all cacheable data and instructions that are normally read out of memory and loaded into the processor are now intercepted and loaded into cache - the processor always works from our cache. What this means is that the processor can run at almost zero wait states because for all intents and purposes, it is running with 25ns memory. "Is it effective? Extremely so. We are able to maintain a cache hit rate of 80-90%, meaning that only 1 time in 5 to 1 time in 10 do we have to go out and make a slow memory cycle. Over 80% of the time we are working out of cache. You can see the dramatic difference caching makes by turning the cache off on a PowerCache - the performance boost, even of a 50MHz board, drops off to about a 30% increase. Why, simply because the added processor speed means nothing if it has to sit there idly waiting on memory to give it more information to process. "This method works on any sort of bottleneck to the processor. We are sold on caching as being a fundamental method of performance enhancement - look at our entire product line! We have our FastCaches & ComboCache, simple cache boards for roughly 20% boosts, our line of PowerCaches that add cpu acceleration to caching for boosts up to over 300%, our SCSI PowerCard, which uses caching in addition to increased thruput to speed hard disks, and our RAM PowerCard, which is a large RAM Disk for caching of your hard disk & VM scratch files. Simply put, whenever you can replace something slow with something faster (especially when you can pretend like you're doing it but actually using far less memory!), you have an advantage. "Altho I shy away from standard benchmarks, as we prefer real world benchmarks, here are the numbers comparing a 50MHz LC to a 50MHz IIci in Speedo 3.06. I have given the summary numbers for each category here. Both were running on 8 bit internal video (the CI gets an additional 15% boost if you use a video card), 50MHz 882 FPUs, the CI had 8MB, & the LC 10MB RAM. They ran off the same Quantum 105 hard disk, running Sys 7.0.1 w/ TuneUp, in 32 bit mode, and had After Dark and QuickMail running init-wise. PowerCache 50 LC PowerCache 50 IIci % Faster Than LC ---------------- ------------------ ---------------- P.R. Rating: 11.88 12.91 9 Bench Avg: 20.82 18.22 *14 (LC is faster) FPU Avg: 4.05 4.05 0 Color Avg: 2.30 3.96 *72 * The LC was faster on cpu intensive activities because the CI has to dedicate more cpu activity to video than the LC when running on internal video. Conversely, it blows the LC away in the video category because it is handled in this manner (vs the LC having to work thru 16 bit data paths). If you add a video card, the P.R. & bench ratings improve, but video comes down to about 21% better than an LC. "Hope this answers some of the questions out there. If you still doubt that it can be true, give your local dealer a call, get an LC PowerCache and test it for yourself. If you are not completely satisfied, return it within 30 days for a full refund (thru participating dealers). "As I have said before on CIS, I had an LC running at SF MW with a 50MHz PowerCache running 50MB of VM and doing full renderings in Infini-D, and it was just screaming along! You will not believe it's running on an LC. It's _that_ fast!" Steve Tuttle DayStar Digital voice: 404-967-2077 fax: 404-967-3018 CIS: 75300,1544 AppleLink: DAYSTAR.MKT Steve can be reached in the CompuServe Macintosh Vendor Forum A (go macaven). DayStar Digital messages are in Section 6 of that forum. As a new PowerCache owner (40 MHz, no FPU), I am impressed with the acceleration. Applications load faster, and games like MS Flight Simulator and Vette that depend on animation are much more playable. The acceleration doesn't make the vehicles run faster (that would be undesirable); the framing rate of the animations goes up noticeably. Before, response to control input was jerky and so I tended to overcontrol. Now I can control the plane or car much more smoothly. I can even do a power-induced oversteer in the Vette without crashing, impossible before the PowerCache because of the lower framing rate. Steve Harmony internet: sharmony@lanl.gov CIS: 71101,404