freeman@spar.UUCP (Jay Freeman) (01/07/86)
[] A while back, there was some net discussion about NECs new processors that are pin-compatible upgrades of the Intel 8088 and 8086. I recently bought an NEC V20 -- the 8088 upgrade -- and installed it in place of my home computer's 8088. My machine is a Micromint MPX-16, which was a Steve Ciarcia project in Byte for November 1982 through January 1983. In hardware features, it roughly resembles a "disunbundled" IBM PC. Mine has two DSDD 5.25-inch floppy disc drives, an 8087, 896K DRAM, a serial terminal (no bit-mapped display), and runs CP/M-86. The MPX-16 runs at a nominal 4.7 MHz clock. As cheap insurance, I bought the 8 MHz version of the new processor. (My system clock remains 4.7 MHz.) With the NEC V20 installed in place of the 8088, everything seems to run. The machine boots, the various CP/M-86 utilities do their respective things, word processors operate, and my C compiler (Mark DeSmet's v. 2.41) functions. I don't use the 8087 much, but a few tests of transcendental functions, via a "desk calculator" program, give the correct results. (This program accesses the 8087 through the 8087 support library that comes with DeSmet's compiler.) I have played with the V20's 8080-emulation mode via the debugger. A few short routines work as advertised. I have not yet gotten around to trying any of the 80186 instructions that the V20 supports. Knock on wood, I have seen no mysterious crashes or failures that might be attributable to the new processor. (I write lots of system code, so I see plenty of garden-variety crashes and failures.) The V20 is observably faster in many applications, particularly in a word processor I have, that uses block moves extensively to shuffle data in text buffers. The V20 is rather more than twice as fast as the 8088 on block moves -- 8 clocks per byte versus 17 -- and it shows. My system uses half a meg of its on-bus RAM as a memory disc, whose driver uses the block move instruction to read and write the "disc": Thus any program that either uses the memory disc or is stored there, will be sped up somewhat. The NEC chip cost about $30. 8088s presently run between $5 and $10. The rest of my system is probably worth between $1K and $2K. For a one percent cost increase, the improved performance of the V20 seems well worth having. I hope the legal controversy between Intel and NEC is settled in a way that makes the V20 available. It would still be worth it if the cost were increased several dollars by a license agreement. If there is anything special anyone would like me to try, send mail. I have no way to port code or source from my net address to home, so I probably will ignore any long test programs I get. -- Jay Reynolds Freeman (Schlumberger Palo Alto Research)(canonical disclaimer)
davidsen@steinmetz.UUCP (01/09/86)
I have two V20's which I got just to try them out. In an XT there is no really
visible change in performance (measurable, but not really much, 15%). On my
S100 system running 8MHz no wait state (static memory), interpreted BASIC
seems to run a good bit faster. I had to change some software timing loops.
I compiled some C programs using the 80186 option, and that *does* make a
diference! I think the big change is "push immediate", which eases the
register allocation, as well as being a bit faster. I have not yet gotten the
CP/M emulator program, since I have a baby blue card to use.
The real bonus is if you get down into assembler and use the extra
instructions. I have been thinking about a postprocessor for assembler source
out of C, which would identify some operations and use special instructions
instead. This would still leave me with portable source code, just another
compile step for the V20 machines.
--
-bill davidsen
seismo!rochester!steinmetz!--\
/ \
ihnp4! unirot ------------->--- crdos1!davidsen
\ /
chinet! ---------------------/
"It seemed like a good idea at the time..."hart@cp1.UUCP (rod hart) (01/11/86)
$30 seems alittle high to me. I only paid $17 for my 8mhz v30. Using Norton's sysinfo my at&t 6300 jumped from a 1.9 to a 3.9 in- dex. Kind of hard to beleive, but it is TRUE! -- =========================================================================== Signed by: Rod Hart (WA3MEZ) Minicomputer Technical Support District Chesapeake & Potomac Tel. Co. A Bell Atlantic Company Silver Spring, Md. UUCP: aplvax!cp1!hart - aplcen!cp1!hart - umcp-cs!cp1!hart - gamma!cp1!hart INTERNET: umcp-cs!aplvax!cp1!hart@SEISMO.CSS.GOV ===========================================================================
ray@othervax.UUCP (Raymond D. Dunn) (01/13/86)
In article <1133@cp1.UUCP> hart@cp1.UUCP (Rod Hart) writes: >..[using NEC v30].. Norton's sysinfo >my at&t 6300 jumped from a 1.9 to a 3.9 in- >dex. Kind of hard to beleive, but it is TRUE! Please note, and tell all your friends, that Norton's SI test results show NO relationship to the actual overall machine performance. The SI command is a simple timing test loop which contains and is heavily weighted by, a multiply instruction. Thus the 'good' result with the V20, and little effect on the result from memory wait states etc. Lets try and kill this one stone dead before manufacturers start designing their machines to produce 'better' SI results! Norton where are you? Can you do us all a service and publish a realistic system performance test?? Ray Dunn. ..philabs!micomvax!othervax!ray
glen@intelca.UUCP (Glen Shires) (01/15/86)
> In article <1133@cp1.UUCP> hart@cp1.UUCP (Rod Hart) writes: > > >..[using NEC v30].. Norton's sysinfo > >my at&t 6300 jumped from a 1.9 to a 3.9 in- > >dex. Kind of hard to beleive, but it is TRUE! > > Please note, and tell all your friends, that Norton's SI test results > show NO relationship to the actual overall machine performance. The > SI command is a simple timing test loop which contains and is heavily > weighted by, a multiply instruction. Thus the 'good' result with > the V20, and little effect on the result from memory wait states etc. > > Lets try and kill this one stone dead before manufacturers start > designing their machines to produce 'better' SI results! > > Norton where are you? Can you do us all a service and publish a > realistic system performance test?? > > Ray Dunn. ..philabs!micomvax!othervax!ray I agree SI is weighted, but what is a realistic system performance test? One application may be spreadsheet recalculation which is numerics intensive. Another may be a compile with multiple libraries which is disk intensive (Or a compile with RAMDISK which is string-operation intensive). Maybe the best test is to simply time your favorite application. -- ^ ^ Glen Shires, Intel, Santa Clara, Ca. O O Usenet: {ucbvax!amd,pur-ee,hplabs}!intelca!glen > ARPA: "amd!intelca!glen"@BERKELEY \-/ --- stay mellow