ObsolYte! Sparc Linux BogoMips Page

Am I Wrong?

On the home page of ObsolYte!, I once claimed that the SparcStation IPX with the Weitek Powerup is as powerful as a Pentium 75 because I read its Bogomips value to be similar. The IPX page still lists it this way. However, a few weeks ago, I recieved this as an email from Timothy A. Seufert.
From your web page, about the sparcstation IPX - 

"BogoMips = 79, which makes it about as fast as a Pentium-75 PC."

Nope.  BogoMips was named "Bogo"Mips for a reason; it's totally bogus as
a benchmark.  It's nothing more than a timer loop -- the kernel runs a
null loop which does nothing but count down to 0.  It calibrates this
loop against the real time clock to see how fast it executes.  From that
point onwards the calibrated loop is used as a way for the kernel to do
timing delay loops; for example a device driver might want to delay for
10 microseconds while hardware does something.

The reported BogoMips figure is just how many times the delay loop can
be executed in one second.  So it really tells you nothing about how
fast the processor is for useful work.

From SPECint92 figures, the Weitek 40/80 MHz SPARC processor is
somewhere between a 486-33 and a 486DX2-66.  It is slightly better than
a 486DX2-66 at SPECfp92 however.
Desperate to prove myself right, I started doing some research. The first thing I learned is that, bogomips might have one meant something -- BEFORE the Pentium. For example, before the Pentium, the bogomips value of a 386 and a 486 was vastly different, even at the same clock speed. However, Intel made a huge leap with even the early Pentiums - at the same or similar clock speed (60Mhz Pentium Versus a 66Mhz 486) the Pentium outperforms the 486. I was begining to think that maybe I should have said that the Weitek was as powerful as a 75Mhz 486!

Egg On My Face

And yes, bogomips are pretty bogus indeed - for the most part, I've noticed that, particularly on the Sparc Linux implementation thereof, the bogomips value is almost always the same or very close to the Mhz rating of the CPU.

From the BOGOmips Mini-How-To:
System                BogoMips  Reporter
   ------INTEL SYSTEM--------------------
386DX/40                  8.06  Bill G. Bohling 
486DX/33                 16.77  Donald Lewis 
486DX/40                 19.8   Jose Calhariz  
486SLC2/66               19.02  Harry Mangalam 
486DX2/66                33.55  Donald Lewis 
486DX4/75                37.47  G Asmundarson 
486DX4/100               49.66  Pete Cervasio 
Pentium/60               23.96  Joost Helberg  
Pentium/75               30.22  Zoran Marjanski 
Pentium/166              66.36  T. Endo 
Pentium/200              79.69  Timm Gleason 
   ------SPARC SYSTEM--------------------
Sparc SLC/20 S1          19.86  Simon Karpen 
Sparc sun4c/25           24.88  Paolo 
Sparc IPX 4c             39.83  Paul D. Robertson 
Sparc SS20/50            49.86  Gary A. Donahue 
Sparc SS5                69.84  Richard D. Davis 
Sparc SS4/110           109.77  Chris Sylvain 
Now, looking at this, one might just say that my little IPX is as powerful as a Pentium-200, but we'd all know that's not true. Also, I can guess from looking at the Sparc systems that the bogomips matches the CPU speed. The listed SS5 is a 70Mhz version, I'll bet, while the SS4 at 110Mhz has a 109.77 bogomips value. What a shock. So, the evidence is piling up that bogomips don't mean a damn thing.

So let's look at something that might measure performance, the SPECint92 rating.
From: SPEC list - 1997/06/11
System            CPU        ClkMHz  Cache      SPECint SPECfp  Info  Source
Name              (NUMx)Type ext/in  Ext+I/D      92      92    Date  Obtained
================= ========== ======= ========== ======= ======= ===== =========
   ------INTEL SYSTEM--------------------
Compaq Deskpro    80486DX2   25/50   256+8        25.7    12.2  Mar93 SPEC news
Compaq Deskpro    80486DX2   33/66   256+8        32.2    16.0  Mar93 SPEC news
Compaq DeskproXL  Pentium    66      256+8/8      65.1    63.6  Sep93 SPEC news
Intel Xpress      Pentium    50/75   512+8/8      89.1    68.5  Mar95 www.intel
Intel Xpress      Pentium    60/90   512+8/8     106.5    81.4  Mar95 www.intel
Intel Xpress      Pentium    60/90   1M+8/8      110.1    84.4  Mar95 www.intel
Intel Xpress      Pentium    66/100  512+8/8     118.1    89.9  Mar95 www.intel
   ------SPARC SYSTEM--------------------
Sun SS/IPC        FJMB86902  25      64           13.8    11.1  Nov92 Sunflash
Sun SS/IPX        FJMB86903  40      64           21.8    21.5  Nov92 Sunflash
Sun SS2/PowerUp   WeitekPwUP 40/80   16/8         32.2    31.1  Jun93 c.sun.an
Sun SS5/110       MicroSP2   110     16/8         78.6    65.3  May95 SunIntro
   ------MIPS SYSTEM--------------------
SGI IndyPC        R4000      50/100  8/8          34      35    Jul93 SGI anno
SGI IndyPC        R4600      50/100  16/16        62.8    49.9  May94 SGI anno
From this, we can see that a SS2 with a Weitek 80Mhz PowerUp is NOWHERE NEAR a Pentium 75. As posted by Timothy Seufert, the machine is comprable in terms of raw CPU horsepower to a 486/66, although it's much faster in floating point operations. The SparcStation 5 is much closer to CPU performance of a Pentium 75. Also note that the Weitek PowerUp gives almost as much CPU performance as an Indy R4000.

However, I am only using the thing as a webserver, and to do that you don't need a killer CPU, just a lot of RAM and a fast HD. Still, you should consider the IPX to be a 486/66 with 1MB of CPU cache, built-in SCSI, built-in Ethernet, and, with a few extra parts, can have as much as 128MB of RAM. That's a pretty elYte 486! However, should my traffic ever get the point that my machine has to start doing extra work just to keep up, then it'll be time to think about something with a little more horsepower.

Want to make a case for or against this evidence? E-mail me and we'll try and figure it all out!