home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!cbmvax!bagate!dsinc!ub!zaphod.mps.ohio-state.edu!saimiri.primate.wisc.edu!caen!sol.ctr.columbia.edu!The-Star.honeywell.com!umn.edu!msus1.msus.edu!msus1.msus.edu!news
- From: lkoop@TIGGER.STCLOUD.MSUS.EDU (LaMonte Koop)
- Newsgroups: comp.sys.amiga.hardware
- Subject: Re: Chip RAM extremely slow with GVP A530
- Message-ID: <1993Jan27.214640.2232@msus1.msus.edu>
- Date: 28 Jan 93 03:46:39 GMT
- References: <93024.073149PATROL@ACI.CVUT.CS> <1993Jan27.010254.2209@msus1.msus.edu>,<1993Jan27.194717.13182@news.cs.indiana.edu>
- Reply-To: lkoop@TIGGER.STCLOUD.MSUS.EDU
- Organization: SCS GP/Engineering Cluster
- Lines: 60
- Nntp-Posting-Host: tigger.stcloud.msus.edu
-
- In article <1993Jan27.194717.13182@news.cs.indiana.edu>, "Stefan G. Berg" <sgberg@avocado.ucs.indiana.edu> writes:
- >lkoop@TIGGER.STCLOUD.MSUS.EDU (LaMonte Koop) writes:
- >
- >>In article <93024.073149PATROL@ACI.CVUT.CS>, Pavel Troller <PATROL@ACI.CVUT.CS> writes:
- >>The 68020, 68030, and 68040 always prefetch instructions using 32-bit read
- >>cycles. Since the A500 motherboard bus is a 16-bit port to the 68030 on your
- >>accelerator, essentially 2 memory read cycles must be accomplished for every
- >>instruction prefetch done. Right off the bat this gives you a 2x performance
- >>hit against a 68030 doing prefetches from a full 32-bit bus port. If you add
- >>to this the fact that, as you indicated, the reads are done asynchronously to
- >>a bus running a ~7.15 MHz, and the fact that you also have to deal with bus
- >>contention from the custom chip DMA on the CHIP RAM bus, you can easily see
- >>where performance drops into the ground in the case with the instruction
- >>cache disabled. Performance figures less than a system running the stock
- >>68000 in this situation are not unexpected.
- >
- >That makes me wonder if this has to do with my serial port problems. I have a
- >68030-33 without any 32 bit RAM (I wait to install it until I get the fan I
- >ordered), but with 1MB Chip and 8MB Fast. Using my Supra modem at full speed
- >works really great since I have the 030. The system doesn't slow down a whole
- >lot even at full speed and my cps rate has gone up considerably. However! When
- >I write something to a PC disk with Cross DOS or my module player in the
- >background switches it's modules I have a pretty good chance of getting
- >transmission errors. Could it be that the 030 is being really wasteful with
- >the Chip RAM bus and even though the CPU isn't slowed down I get problem with
- >my serial port due to an overloaded chip bus? Oh... interlace and high color
- >screens are also a pretty bad idea when doing file transmissions. A friend
- >of mine with a Supra modem, but "only" an AdSpeed (14MHZ 68000) doesn't seem
- >to have any of these problems although my computer outperforms his in terms of
- >benchmarking software.
-
- Well, this sounds like a a symptom of the effects of running off of a
- 16-bit ported memory arrangement alone. It looks like you are getting a lot
- of your problems from the latency encountered by the 68030 whenever it has
- to hit the CHIP RAM bus. That, coupled with the fact that a lot of the
- conditions you mention cause the contents of the instruction/data caches to
- be replaced can result in some of the problems associated with the 68030
- running bus cycles from a slower, asynchronous 16-bit ported resource.
- Word of warning: many 'module players' are not real kind to the system in
- terms of multitasking, and will exibit the problem you mention in most
- setups.
-
- >Let me know if there is some truth to this. How could I fix the problem? I
- >guess the 32 Bit Fast RAM won't help in this case. It might even make it worse,
- >because access to Chip RAM will be more intense (less time will be spend in
- >executing programs in the faster 32 Bit RAM). BTW, my KS ROM is mapped in the
- >512KB SRAM I have on my accelerator board.
-
- This paragraph doesn't make any sense. Adding a 32-bit ported memory
- resource to your system will of course help things along in terms of
- performance by reducing the latency in memory cycles. Accesses to the CHIP
- BUS will of course remain the same as before, but by having a 32-bit ported
- memory area, code/data fetches from that memory will be significantly faster,
- reducing the overall execution time.
-
- ----------------------------------------
- LaMonte Koop -- SCSU Electrical/Computer Engineering
- Internet: lkoop@tigger.stcloud.msus.edu -OR- f00012@kanga.stcloud.msus.edu
- "You mean you want MORE lights on this thing???"
- ---------------------------------------------------------------------------
-