vaso@mips.UUCP (Vaso Bovan) (12/12/87)
Norton's SI is getting to be a public embarrassment, especially when used to tout 8088 and 8086 machines containing V20 and V30 uPs. Often an ad will say "1.8 times IBM-PC speed" or "twice IBM-PC speed" "by N.I. SysInfo". It amounts to a fraud on the public to sell a V-series computer on the basis of its S.I. rating. I've spoken to the authors of two other benchmarks, the Landmark and the P.I. Both benchmarks show little improvement in performance for the V-series over 80XX series uPs. The Landmark benchmark assigns ZERO improvement, and the author has stated to me that most real-life applications will not show more than a 2% improvement. Incidentally, the V40 and V50 show slight (10%) improvement in throughput. The JC LIPS and the Packard-Bell VX88 are two interesting 8 MHz V40 machines. There is an article in the third Special IBM-PC BYTE Magazine, which discusses the role of speedups and performance improvements. In it the author gives a rough "rule of thumb" that the 8086 has 50% more throughput than the 8088, due to the larger data path. It appears also that straight clock-speed tweeks are strictly linear. Putting the two effects together, it appears that the 8 MHz PS/2, Model 30 is about 2.5 times the performance of the original IBM-PC. The new Ogivar System III machine, from Canada, which runs an 8086 at 10 MHz is about 3.1 times the performance of the IBM-PC. Now, considering Landmark's rule of thumb that a 2 MHz AT is approximately equal in performance to a PC, then it appears the Ogivar 10 MHz 8086 is faster than the original 6 MHz IBM-AT ! Incidentally, Norton wrote back to me to say that they had no intention of changing S.I. to remove the distortion in the V-Series performance figures. Their decision smacks of high-handed arrogance. Comments ? -Vaso
jallen@netxcom.UUCP (John Allen) (12/15/87)
In article <1098@quacky.UUCP> vaso@mips.UUCP (Vaso Bovan) writes: >Norton's SI is getting to be a public embarrassment, [....] > >I've spoken to the authors of two other benchmarks, the Landmark and the >P.I. Both benchmarks show little improvement in performance for the V-series >over 80XX series uPs. The Landmark benchmark assigns ZERO improvement, and ---- Is this any better than showing a LARGE difference? >the author has stated to me that most real-life applications will not show >more than a 2% improvement. In point of fact, the V20/V30 can be MUCH faster given any application which makes extensive use of 186 enhancements. When I upgraded my PC Clone from an 8088 to a V20, I was literally awed by how much faster Windows ran. PC-Paint also looked a lot faster. Whether the V20 gets better gas mileage for you or not depends on what type of driving you do. If you use multiply or divide a lot, it helps. >Incidentally, Norton wrote back to me to say that they had no intention of >changing S.I. to remove the distortion in the V-Series performance figures. >Their decision smacks of high-handed arrogance. Sysinfo may not be the best standard, but isn't it better to leave it consistent, than to make subsequent versions incompatible, and therefore incomparable? John Allen ========================================================================= NetExpress Communications, Inc. uunet!netxcom!jallen 1953 Gallows Road, Suite 300 (703) 749-2238 Vienna, Va., 22180 =========================================================================
joel@peora.ccur.com (Joel Upchurch) (12/16/87)
> In point of fact, the V20/V30 can be MUCH faster given any application > which makes extensive use of 186 enhancements. When I upgraded my PC Clone > from an 8088 to a V20, I was literally awed by how much faster Windows ran. > PC-Paint also looked a lot faster. > > Sysinfo may not be the best standard, but isn't it better to leave it > consistent, than to make subsequent versions incompatible, and therefore > incomparable? > > John Allen I noticed the speedup with Windows on the V20 also. I suspect that the improvement may be caused by a lot of shift and rotate type instructions in the display driver. The V20 runs these instructions about 2 to 3 times faster than the 8088. It makes Windows tolerable on a 8MHz clone. One point I haven't seen mentioned is that the Norton SI also distorts the results for 80286 based machines also, since the instructions that run faster on the V20 also run faster on 286. Thus these ridiculous double digit SI ratings on a lot of the 286 clones. For speeding up Windows you might be better off getting a faster display card like the new one from Thomson. -- Joel Upchurch @ CONCURRENT Computer Corporation (305-850-1040) Southern Development Center/2486 Sand Lake Road/ Orlando, Florida 32809 {decvax!ucf-cs, ihnp4!pesnta, vax135!petsd, akgua!codas}!peora!joel
wtm@neoucom.UUCP (Bill Mayhew) (12/17/87)
I heartly recommend the use of PC Magazine's PC Labs' _Bench_ IBM compatible computer test suite. I forgot just how we obtained our copy. I think it was gotten from one of the software offerings from the back of PC Magazine. Bench is a pretty good test suite and can be used to test quite a few aspects of system performance. CPU relevent feautures including instructions, FPU and regular as well as (E)EMS memory timings are tested. Hard disk and video apators can also be spec.'ed. The CPU oriented items tested are: 1. 128K NOP Loop 2. Do-nothing Loop 3. Integer Add Loop 4. Integer Multiply Loop 5. String Sort and Move 6. Prime Number Sieve The times are given as %of 4.77 MHz XT and 8.00 MHz AT Before I put a V-30 in my AT&T 6300, the integer Multiply was about 0.3 of an AT. After the V-30, the Mults went up to about 0.6 AT. String Sort and Move also showed noticable improvement. The other CPU operations were not so dramatic. I don't remember the exact figure difference for 8086 vs V30 as reported by si, version 4.0; it was something like the performance of the V30 was 2.5 times the 8086. Just watching the computer run, it was obvious that the Norton Index was over-rating the V30. --Bill