home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.sys.mac.hardware
- Path: sparky!uunet!nwnexus!kanefsky
- From: kanefsky@halcyon.com (Steve Kanefsky)
- Subject: Re: Daystar Powercache (need info...)
- Message-ID: <1992Sep14.152454.21271@nwnexus.WA.COM>
- Sender: sso@nwnexus.WA.COM (System Security Officer)
- Organization: The 23:00 News and Mail Service
- References: <1901vqINNkah@burr.cs.utexas.edu> <1992Sep13.214046.7300@nwnexus.WA.COM> <1913mpINNfvb@mohawk.cs.utexas.edu>
- Date: Mon, 14 Sep 1992 15:24:54 GMT
- Lines: 87
-
- In article <1913mpINNfvb@mohawk.cs.utexas.edu> newton@cs.utexas.edu (Peter Newton) writes:
- >> I expect that the 33mhz '040 PowerCache due out soon will beat the
- >> Quadra 700 and 900 on the Math and CPU related stuff, and be very
- >> close to (possibly even ahead of) the Quadra 950 in these areas. The
- >> 16-bit data path of the LC II has very little effect once you go to an
- >> external cache.
- >
- >I wonder. Just how much cache memory comes with with Daystar
- >PowerCaches? One problem with with small benchmarks like those built
- >into speedometer is that they tend to fit entirely within even fairly
- >modest caches. Things might be a bit different with production sized
- >apps, depending on their locality. It would be interesting to do some
- >timings with larger programs.
-
- Ah, a perfect opportunity to present my "real-world" benchmarks. Last
- night I decided to remove the PowerCache and do some benchmarks, after
- which I put it back in and ran the same benchmarks. I tried to be as
- careful as I could about not letting caches (either disk or RAM) confuse
- the test results. I always tried to flush the caches before running each
- test (e.g. before running an Excel test, I would open up Word).
-
- The results seemed to agree with the Speedometer benchmarks pretty well.
- Naturally, if Speedometer says that the CPU is three times faster and the
- disk is only 20% faster, than a task involving both disk and CPU will be
- somewhere between 20% and three times faster.
-
- (BTW, the answer to your question about the cache is that it's 32K
- direct-mapped)
-
- So, without further ado, here are the benchmarks I ran and the results:
-
-
- Boot time from Restart (with just the 32 sec 28 sec*
- Daystar PowerStation control panel)
-
-
- Boot time from Restart (with a dozen 52 40*
- or so extensions)
-
-
- Time to launch MS Word 5.0 24 12
-
-
- Time to launch MS Excel 4.0 29 16
-
-
- Time to launch MS Excel 4.0 (with several 117 50
- linked worksheets and add-ins, including
- the time to finish re-calc)
-
-
- Time to uncompress and display GIF file 32 12
- in Giffer 1.2 (503K after decompression)
-
-
- Time to de-binhex 303K .hqx file in DeHQX 43 22**
- (216K after de-binhex-ing)
-
-
- Time to compile Art Class demo in THINK C 1045 386
- 5.0 from scratch
-
-
- ----------
-
- * Note that the PowerCache boots with it's 32K cache off, eliminating a
- lot of the advantage until the PowerCentral control panel loads and turns
- on the cache. Loading PowerCentral first speeds up the rest of the boot
- process.
-
-
- ** It was very clear in the DeHQX test that disk access took most of the
- time. The program would just fly through a chunk of the decoding process,
- then sit there and wait while it wrote out to disk, then continue with the
- next chunk. I didn't get a chance to test this with a RAM disk, but I
- suspect the difference would have been much greater.
-
- I may do another round of benchmarks, so if anyone has any specific
- suggestions on what to test (you can send me sample documents if you
- wish), I'd be happy to try them out.
-
-
-
- Thanks,
- --
- Steve Kanefsky
-
-