Path: utzoo!utgpu!news-server.csri.toronto.edu!mailrus!uwm.edu!rpi!zaphod.mps.ohio-state.edu!uakari.primate.wisc.edu!ames!eos!shelby!portia.stanford.edu!dhinds From: dhinds@portia.Stanford.EDU (David Hinds) Newsgroups: comp.sys.ibm.pc Subject: Re: 486: Fantastic or Flop? Message-ID: <1990Jun4.215723.19339@portia.Stanford.EDU> Date: 4 Jun 90 21:57:23 GMT References: <1990Jun3.181830.2145@chinet.chi.il.us> Organization: AIR, Stanford Universit Lines: 41 In article <1990Jun3.181830.2145@chinet.chi.il.us> orac@chinet.chi.il.us (Paul Giovacchini) writes: >In article clong@topaz.rutgers.edu (Chris Long) writes: >> >>Has anyone else read the article in the June issue of PC World where >>33 MHz 386s are compared to 25 MHz 486s? The author conludes that >>there is no substantial difference in performance in most cases >>between the two. This shocked me, because I'd expect the 486 to >>be about twice as fast, on the average. > > Did the machine tested have an AT bus? It must have. Because > the AT bus can not do the "burst mode" of the 486, the performance > increase that you will get, as I have read in about 3 different magazines > is "negligible". BUT if you have either a MCA or an EISA bus, which can > both handle the burst mode, the performance is astronomical. > The bus type isn't directly relevant here, if CPU performance is the question. Memory installed on the system board is not accessed via the I/O bus, so a board can support burst mode accesses to main memory regardless of the I/O bus type. I would expect the overall performance gain for a 25MHz 486 to be only on the order of 5% with burst mode. Remember that this is only improving the performance of cache misses - sure, that helps, but the whole idea is to not go to main memory as often in the first place. The performance improvement will be more significant for 33MHz and 50MHz parts, because the CPU otherwise may have to wait around a lot for main memory. In the latest issue of BYTE, there is a review of the latest Cheetah 486-33 system. The system does not use an external cache, but uses 35ns DRAM's for main memory (!?!) to get near-zero wait states on internal cache misses. And the board has only an ISA bus, though with some tricky performance improvements. I think the important question to ask about the original posting is, for what software? If the 386 and 486 are running DOS code, then they are both taking major performance hits to begin with, and can't take advantage of the 32-bit-wide bus. I would guess that if the CPU is spending all of its time loading and storing data 8 or 16 bits at a time, even if these accesses are largely cached, they will bury differences in instruction execution times. For 32-bit code, execute time will start to become limiting, and the large advantages of the 486 should be much more apparent. -David Hinds dhinds@popserver.stanford.edu