home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!usc!chaph.usc.edu!news
- From: baffoni@aludra.usc.edu (Juxtaposer)
- Newsgroups: comp.sys.atari.st
- Subject: Re: 68040 accelerator board?
- Date: 7 Sep 1992 22:03:29 -0700
- Organization: University of Southern California, Los Angeles, CA
- Lines: 52
- Message-ID: <laod11INN47r@aludra.usc.edu>
- References: <9209031931.AA15117@photon.magnus.acs.ohio-state.edu> <lafqubINN6fn@aludra.usc.edu> <H.NB0mn1ZPbEM@fredrik.atari.no>
- NNTP-Posting-Host: aludra.usc.edu
-
- In article <H.NB0mn1ZPbEM@fredrik.atari.no> jornmoe@fredrik.atari.no writes:
- >> From the descriptions so far, it is a stand alone (a tower?) from
- >>the main Mega ST (only so far) that houses a 66MHz 68040 (way over-driven
- >>IMHO, put perhaps bearable if it has some INDUSTRIAL STRENGTH heat sinks
- >>and one heck of a fan - maybe liquid cooled :). Perhaps more busses (I hope)
- >>and memory. Supposedly runs at about 24MIPs. I wonder if it uses a cache
- >>(outside of the onboard of the '040)? Would really Haul A** with a decent
- >>(ie: 256k+) cache...
- >>
- >This make no sence! The 25Mhz version of the 040 does 20MIPS. If a cklock
- >frequency more than two times higher only yields 25% increase, I'd be
- >realy disapointed.
-
- Hmm. What source are you quoting for the 20mip@25MHz? Motorola?
- If so, don't forget that they don't take into consideration real-world
- problems like a) slow internal memory, b) OS overhead/ineffieciency, c)
- interrupts, etc. Motorola uses an optomised test program (that does nothing
- but execute no-ops or something similarly useless - maybe it just increments
- a counter ... in a register ....) for those figures.
-
- However, if your MIP reading is a real-world figure, then yes, I would
- say that is sad performance. Even the Intel DX2 chips reported on real systems
- were found to have a 50-70% increase over base system speed (more cache the
- better). This could mean that the Mega '040 thing is really inefficient and
- probably has no cache (outside of the onchip 8 (16?)k. Then again, it is
- probably having to deal with slow (120ns) memory of the MegaST which is
- further limited by a time-slice arrangement with the video. If that is the
- case (that it doesn't have its own memory), then I am REALLY impressed that
- they could get 24MIPS out of it.
-
- Remember, it still has to use/communicate with the ST bus/architecture,
- and there is likely to be a WHOLE lot of time eaten for I/O between busses.
-
- >No, the thing is that this processor actually runs at 33MHz, but it has a
- >onboard clock-doubler wich make the internal clock 66MHz. This is the same
- >trick as done on the Intel DXII processors!
-
- A thousand pardons for not specifying that it used a slower 33MHz
- bus. BTW, if it is using a 33MHz bus, it is not using the MegaST's - if so,
- it's bus must a) exist, b) have more than a CPU hooked to it, c) not reside
- in the MegaST. So the question becomes: What is on the bus with the CPU?
- However, like I said, there should be more than a measely 25% improvement
- (unless I am not taking something into consideration that I should).
-
- > _______________________________________________________
- > / Joern F. Moe / All above is my own personal opinion! \
- >/ Oslo, Norway / Any lack of opinion above is also mine! \
- >-----------------------------------------------------------
-
- -Mike
-
-
-