In article <1994Nov17.093153.34185@miavx1>, [log in to unmask]
(Kent Covert) writes:
> In article <1994Nov16.145923.34092@miavx1>, [log in to unmask]
(Aaron Porter) writes:
>> Kent Covert ([log in to unmask]) wrote:
>> : That's a fair question...
>>
>> : Model:                          DEC 4000-710 AXP
>> : CPU:                            One DECchip 21064 / 190 Mhz
>> : Caches:                         8KB Instruction Cache, 8 KB Data Cache,
>> :                                   4 MB processor cache
>> : TPS:                            300
>> : SPECint92:                      93.8
>> : SPECrate92:                     188.4
>> : LINPACK 1000x1000 (DP MFLOPS):  145.6
>> : Memory:                         256 MB
>> : Hard Disks:                     Two 2GB SCSI-2, Six 1GB Fast SCSI
>> : Total disk space:               10 GB
>> : CD-ROM Drives:                  5
>> : OS:                             OpenVMS 6.1 AXP
>> : Anything else that I've forgotten?
>
>>      Eeek.  No wonder it sometimes lags.  At least the memory is of
>> the right magnatude, but 190 Mhz isn't really all that fast nowadays.  I
>> shudder to think what it was like before!
>
> Be cautious with this information.  Mhz are probably the WORST comparison
> of processor speed.  You cannot use Mhz to compare the Alpha chip to a
> Intel X86 chip.  Even in the Intel chips, you can't use Mhz to compare
> speeds.  For example, a 33 Mhz 486 is a lot faster than a 33 Mhz 386 due to
> the internal structure of the chip.
 
Yes, this is a bad comparison. To be a little more detailed, the 486 is about
twice the speed of a 386 at the same clock speed. Comparing across CPU brands
is
practically impossible.
 
> In relation to the Alpha chips, the chip is a RISC based processor.  The
> chip has 190 Million cycles/second, but it can execute up to 4 instructions
> simultaneously in each of those cycles!  It's a 64-bit processor compared
> to a 32-bit X86 chip (which goes into 16-bit mode when using a "real"-mode
> application (I think)).  In addition, the chip has built-in pipelining - a
> feature that used to be available only with vector processors.
 
Actually, so do the latest PC chips - the Pentium has some pipelining abilities
(though this is hardly an adequate comparison). But, it is in RISC chips that
these types of features are mostly found.
 
> DEC just released a new Alpha chip called the DECchip 21164 that runs at
> 300 Mhz and issues 1.2 billion instructions per second.
 
This is truly one *awsome* chip. It is the worl'ds fastest microprocessor.
Running at either 266 or 300 MHz, it puts forth 330 SPECint92 and 500 SPECfp92.
It is the first microprocessor to run at over 1 billion instructions per second
- 1.2 BIPS to be exact (for comparison, a 25 MHz Motorola 68040 puts out a
little over 29 MIPS, or 29 million instructions per second).
 
> If you want to compare performance, use something like SPECint92 and
> SPECfp92 or TPS.  These are "supposedly" a better comparison between
> processors.  By the way, the label above that says SPECrate92 should have
> said SPECfp92...I mis-typed it.  SPECrate is a completely different
> performance specification that I didn't list.
 
To be honest, MIPS (or BIPS) is not a good measurement to use either...
 
> Another interesting comparision is that the Alpha chip only has 1.68
> million transistors.  The latest Pentiums have 3.1 million transistors.
> The PowerPC chip has 2.8 million transistors.  I don't know how this
> impacts performace or heat dissipation or anything else...it was just an
> interesting statistic.
 
This is interesting. With a Pentium, a good portion of those 2.8 million is in
the what basically amounts to an on-board 486, for the purpose of backward
compatibility. With a PowerPC 601 (can't remember right off hand how many
transistors, I think something like 2.1 million - ??), most are used to
implement the caches, which are larger than the 21064's. A PowerPC 620 holds 7
million (225 SPECint92 and 300 SPECfp92 @133 MHz), the new DEC Alpha 21164
contains a whopping 9.3 million. Obviously, the number of transistors does not
really effect speed directly; it just depends on what they are used for.
However, it does greatly effect heat dissipation and energy requirements. This
is why the PPC 601 is much cooler and smaller than the Pentiums (prior to the
P54C, whih incorporates extensive power management features). Be aware,
however, that when chips like the PPC 620 or Alpha 21164 come up, we are
talking about workstations, servers, and mainframe type applications. Under
these conditions, things like die size, heat dissipation, and energy usage are
not near as important as in the PC market.
 
Regards,
Bob
--
+------------------------------------------------------+
|  Robert E. Williams, Jr.                             |
|  Enterprise Software                                 |
|  2006 State Route 380                                |
|  Wilmington, Ohio  45177-9241                        |
|  (513) 382-8232                                      |
|                                                      |
|  E-mail: [log in to unmask]              |
+------------------------------------------------------+
|   Those who are patient in the trivial things in     |
|   life and control themselves will one day have the  |
|   same mastery in great and important things.        |
|      --Hapkido Master Bong Soo Han                   |
+------------------------------------------------------+