emmo@moncam.co.uk (Dave Emmerson) (05/05/90)
In article <1584@mindlink.UUCP>, a575@mindlink.UUCP (Michael G. Henders) writes: > > Well, no-one seems to be prepared to comment on the Cyrix or IIT > math chips, so here's my $.02 worth... OK OK OK! I dug out the literature.. Cyrix Fastmath (TM) 83D87 Processor 5.5 MFLOP System Performance Reduced cycles/instruction [compared to what? emulation?] Most instructions accelerated by more than 5x Lowest power consumption Most accurate results - Transcendental functions accurate to more than 65 bits - 1.5 x 2^-66 Fully 80387 compatible - verified with > 24,000,000 tests. ...... Achieves 4 to 15 times greater performance than Intel 80387 by implementing its floating point primitive operations in hardware rather than in a microprogrammed sequencer. ....performs floating point at same speed as 80386 can perform integer additions. ... instructions ..are binary and function compatible with .. 80387. ... clock rates of 25MHz and 33MHz. The booklet lists a decent array of hardware and software products it has been tested with. All the above is extracted without permission, and I accept no reponsibility for any misrepresentation of this product. If you REALLY want to know, ask them directly. Their address Cyrix Corp, 1761 International Parkway, Richardson, TX 75081 (214) 234-8388 Dave E.
mmm@cup.portal.com (Mark Robert Thorson) (05/05/90)
Both the IIT and Cyrix chips offer improved performance. The Cyrix chip is the same price as Intel, and IIT is about 9% cheaper than Intel at the distributor level for their 387-compatible, 22% for their 287. There are divergences between Intel and either IIT or Cyrix. At a minimum, this applies to the stack frame following certain floating-point exceptions. But virtually all software ignores this, at least in the case of the Cyrix chip. The problems are more serious with the IIT chip, but for stuff like spreadsheets, this also isn't a problem. I've heard that some advanced CAD and simulation stuff has problems with IIT. Use of either Cyrix or IIT voids the warrantee of some computers. Cyrix has recently introduced a new revision of the chip, which is identified by the letter B following the date code. This fixes a problem with the handshake between the 386 and the chip in the 387 socket. This problem only occurs on a few pc's, such as the ones from Epson and NEC. When the problem occurs, the machine hangs. Unlike that problem with early 486's, no silent production of corrupted data occurs. Some reports of bargain prices on the Cyrix chip seem to be the result of clearing the old revision out of the distribution channels. People get pretty nervous when they hear their silicon is down-rev. The Weitek chip requires software to be compiled specifically for that chip. I can't think of a single commercial program which uses that chip. I thought AutoCAD was available for use with the Weitek chip, but I've been told that isn't true. (Perhaps you can correct me?) AMD makes 287's and is about to introduce a 80C287 clone. This is going to be the focus of a big fight between AMD and Intel, because AMD used Intel microcode in their design. AMD claims they have a right to, under an old cross-licensing agreement. The AMD 80C287 appears to be a stalking horse for a rumored 386 clone (that's right, a CPU). If the legal issue can be won for the C287, at least AMD will be in a position to break Intel's CPU monopoly. Intel takes a dim view of that.