tim@cit-vax.Caltech.Edu (Timothy L. Kay) (06/02/90)
I just ran my ray tracer on my 386 and a SparcStation 1. The 386 is 20 MHz with a 20 MHz 80387. The SparcStation 1 is not a 1+, so it runs at 20 MHz. I used gcc on the 386, and cc on the SS1. I found that the SS1 is about 8 times as fast as the 386. Does this fit with others' experiences, or did I mess up my benchmark? I would guess that a 486/25 is roughly 4 times as fast as my 386/20. Do people find this reasonable? I would guess that the SLC is about the same speed as the SS1. Is this right? This information is useful in light of the recent SLC vs. 386 discussion that has been going on here. If you don't need speed, you can probably get a 386 for less than the SLC. However, if you need speed, you need to compare the SLC against a 486. Then the SLC looks like a very good price. I'll probably keep running 386/ix on my 386/20, and add a SLC (which I can get for an educational price of about $3100. This way, I get speed and color. (I have a Hercules Graphics Station Card which gives me VGA and 24 bits/pixel.) (I hope that the X11 1.2 update will give me both 800x600 non-interlaced and 1024x768 interlaced. My monitor can't handle the 1024x768 ni, eventhough the card can.) Tim P.S. I just bought a 14" CTX monitor for about $350 which supports all the Hercules GSC modes except 1024x768 non-interlaced. I really like this monitor and the price is right, too. I can get the model number for anybody interested.
debra@alice.UUCP (Paul De Bra) (06/03/90)
In article <15105@cit-vax.Caltech.Edu> tim@cit-vax.UUCP (Timothy L. Kay) writes: >I just ran my ray tracer on my 386 and a SparcStation 1. The 386 is >20 MHz with a 20 MHz 80387. The SparcStation 1 is not a 1+, so it >runs at 20 MHz. I used gcc on the 386, and cc on the SS1. > >I found that the SS1 is about 8 times as fast as the 386. Does this >fit with others' experiences, or did I mess up my benchmark? > >I would guess that a 486/25 is roughly 4 times as fast as my 386/20. >Do people find this reasonable? I don't think the ray tracer (which I don't have) looks like a typical application, unless you happen to have a very slow machine. But maybe it's floating point? My experience with benchmarking systems is that a SparcStation (and other risc machines like the DecStation 3100 and the HP 9000/825...) does extremely well in small tests, but not in larger tests. Aside from typical benchmark programs my real test is to run a 230 page document through Latex. This takes a little under 3 minutes on a SparcStation and a Vax 8550 (around 7 mips) and a little over 3 minutes on an Everex Step 386/25 with a '387. This 386 box performs about 2x what a 386/16 does, and the 386/20 is somewhere in between, so I would guess that a SparcStation is less than 2 times faster than the 386/20 and a 486/25 should be about twice as fast as a SparcStation. Now, if you want to run X-windows for instance, and you only have VGA (in whatever form), then the Sparc may do a lot better because VGA is real slow. Paul. -- ------------------------------------------------------ |debra@research.att.com | uunet!research!debra | ------------------------------------------------------
richard@pegasus.com (Richard Foulk) (06/05/90)
> >I found that the SS1 is about 8 times as fast as the 386. Does this >fit with others' experiences, or did I mess up my benchmark? I've compared the SS1 to a Mylex 33MHz 386 (w/ISC 2.0.2) on a number of different tests and found that they varied widely. The 386 was noticeably faster on a number of things, and the SS1 was faster on others. The Mylex lacked a 387, but on purely integer things it was often 10% faster than the SS1. -- Richard Foulk richard@pegasus.com
amull@Morgan.COM (Andrew P. Mullhaupt) (06/07/90)
In article <1990Jun7.023359.12523@maverick.ksu.ksu.edu>, terry@eesun1.eece.ksu.edu (Terry Hull) writes: > debra@alice.UUCP (Paul De Bra) writes: > WHile LaTeX might be a typical application to run, it is hardly a good > indicator of raw CPU performance. LaTeX does far too much disk > I/O to be a CPU performance indicator. If you want to test overall > throughput, that is one thing, but CPU speed is another. > > One of our faculty members just ran a spice circuit analysis and found the > following: > MIPS machine was 2x faster than our campus dual processor Solborne. > Solborne was 2x faster than SparcStation I > SparcStation I was 4x faster than a cached 386/16 with a '386 installed. ^^^^^^ I should hope so. Running the benchmark without the 80386 installed could adversely affect the CPU performance... But seriously, folks: I have a 386/20 (64K cache + 80387) and an i486/25 (Cheetah Gold 425) and I can run stuff on a lot of UNIX boxes (Sparcstation-1, Solbourne Series 5, Sun 3/110, etc.). For CPU: the Solbourne Series 5 seems about 2x the SS-1. The i486 is a bit quicker than the Sparcstation-1 except in case of floating point intensive code, in which case the Sparcstation-1 is a bit faster. The 386/20 is about 35-40% the speed of the i486 for CPU, and so it is in the neighborhood of 50% of the SS-1. For I/O: The i486 is about 1.5 x the speed of the SS-1. The Solbourne is usually loaded so I can't give a good number for it based on code I've run. The 386/20 is about 50% of the speed of the SS-1. Moral of the story: the decision between the SS-1 and the i486 is based more on what software you want to run than on performance. The 80386/20 is a fine, but old machine. They were very well placed for price/performance for a while, (if you got one of the faster ones). Hardware cache (size+design) makes a big difference in the relative performance of 80386 systems. The ISA bus is not a millstone around the neck of a fast CPU if the bus is used to communicate with intelligent peripheral cards like the ESDI caching controllers, and not allowed to adversely affect memory access. It, (or the EISA and MCA buses) allow for a wide variety of expansion possibilities. The Sparcstation comes with a big monitor and an ethernet connector. This makes life simple if what you already have is a network, (e.g. of other Sun workstations). Later, Andrew Mullhaupt
srodawa@vela.acs.oakland.edu (Dr. Srodawa) (06/09/90)
I ran the byte unix benchmarks on a SparcStation-1 and a 386-16 with 64K cache, 4Mbyte main store running SCO Xenix '386 2.3.2. I too found the Sparc about 4 times faster. (80387 was installed). Interesting enough, in the simulated concurrent user load at the end of the benchmarks Xenix closed slightly on the Sparc as the concurrent processes increased. All in all, the 80386 and Xenix stood up pretty well to the challenge. (Yes, that was the Xenix product and not Unix). -- | Ronald J. Srodawa | Internet: srodawa@unix.secs.oakland.edu | | School of Engineering and CS | UUCP: srodawa@egrunix.UUCP | | Oakland University | Voice: (313) 370-2247 | | Rochester, Michigan 48309-4401 | |
pwilcox@paldn.UUCP (Peter McLeod Wilcox) (06/09/90)
In article <1990Jun7.023359.12523@maverick.ksu.ksu.edu>, terry@eesun1.eece.ksu.edu (Terry Hull) writes: > One of our faculty members just ran a spice circuit analysis and found the > following: > MIPS machine was 2x faster than our campus dual processor Solborne. > Solborne was 2x faster than SparcStation I > SparcStation I was 4x faster than a cached 386/16 with a '386 installed. Since both the spice run and the earlier ray tracing test are heavily dependent on floating point performance, rather than integer and general cpu speed, it would be interesting to run the tests on 386/486 machines with the Weitek FPU processors, in addition to the Intel FPU. The Weitek FPUs are claimed to give a factor of 10 improvement in the flops number. This would make the 386/486 more than competitive with the SS1 (if true :). PS: I am assuming, of course, that the ray tracer didn't use integer math. -- Pete Wilcox ...gatech!nanovx!techwood!paldn!pwilcox