Path: utzoo!attcan!uunet!snorkelwacker!bloom-beacon!bu.edu!bu-cs!mirror!prism!rob
From: rob@prism.TMC.COM
Newsgroups: comp.sys.ibm.pc
Subject: Re: Speed in MIPS
Message-ID: <206900157@prism>
Date: 12 Jan 90 19:07:00 GMT
References: <111607@<1990Jan10>
Lines: 42
Nf-ID: #R:<1990Jan10:-11160700:prism:206900157:000:1823
Nf-From: prism.TMC.COM!rob    Jan 12 14:07:00 1990


>I am interested in finding out the speed of the original PC (4.7 Mhz),
>286 (10/12/16 Mhz), 386 (SX 16/25/33 Mhz), 486 (? Mhz), and PS/2 in terms
>of MIPS.  Does anyone have the information??

   With the disclaimers that a) MIPS ratings don't mean much by themselves, 
and b) there's disagreement over how to measure them, these are figures that 
are in the ballpark:

PC (4.77)		0.3
AT (6)			1
AT (8)			1.4
AT (16)			2.9
386 (16)		3.3
386 (25)		5.4
386 (33)		7.1
486 (25)		14

   You can extrapolate/interpolate from these numbers. Different machines 
running the same CPU at the same speed can show differences due to memory 
architecture and bus structure. Adding a cache (for example) can boost 
performance by 20% or so. 

   This table takes that into account to some extent. For example, running 
16 bit code, a 286-16 and 386-16 will offer about the same performance. But, 
since 386 machines often have better memory architectures, they tend to be 
faster. Still, there are some 16Mhz 286 machines that are faster than some 
16Mhz 386 machines running DOS programs.

   The 386SX can produce the same number of MIPS as a 386, but its narrower 
bus means that it will typically run more slowly at a given speed. Depending 
on the task, this penalty could be negligible or severe.

   The PS/2's can be figured by checking what CPU / clock speed combination 
they use. IBM's claims aside, the PS/2's are no faster, clock per clock, 
than competing machines.

   The number for the 486 is an estimate based on Intel's figure for the
'average' instruction execution time and on 386 vs. 486 benchmarks I've
seen. The 486 is very sensitive to the nature of the test code, and small, 
'loopy' programs like benchmarks can exaggerate its performance. I'd be 
interested in any independently measured ratings for the 486.