RE: Buying a new computer...

Colin L. Hildinger (colin@ionet.net)
Thu, 24 Apr 97 12:22:09 -0500


On Thu, 24 Apr 1997 09:17:48 +0200, Johan Hartzenberg wrote:

>
>Hi Howard.
>
>Well, I can't remember seeing any MKEPS (M Keys Per Second :) ratings
>for the MMX machines, but PPro is way faster than normal Pentiums as far
>as MKEPS is considdered. Compared to the MMX, the PPro have a better
>underlying design, better microcode instruction parsing, better
>out-of-order execution, better branch prediction, more pipelines, Extra
>steps in each pipeline (if I remember correctly), etc. You're not just
>buying for DES Chal, so buy the machine with the more power.

I'd buy the PPro, but the PPro won't be faster for DES Chal. My
brother's PPro 180 does about the same # of keys/s as my normal P166
(of course I have a good 430HX based MB, I don't know about his...).

>>>that takes advantage of the SIMD architecture, it is probably
>>>faster on
>>>MMX. However, if you don't use the MMX part, which is faster?
>>>The MMX
>>>chips does have a larger cache than the Pentium Pro.
>P/Pro have got 1 Chache for L1 and L2, and it's either 256 Kb or 512 Kb
>at the moment. More over, it's built into the CPU.
>
>A bigger cache is more complex to manage effectively, and I don;t know
>how well the P/Pro does this. That would be the difference between
>rating this on-CPU cache as L1 or L2.

Uh, the PPro has a L1 and L2 cache, that's how.

>L2 cache generally fetch larger blocks of data at a time, but does so
>slower. L1 Fetch smaller blocks of data, mostly instructions and byte
>values, but does so much faster. L1 Cache runs very close to the CPU so
>that tehre is almost no delay in getting anything from L1 cache into the
>CPU registers. I also don't know what sort of delays there is in moving
>data between the CPU registeres and the Cache.
>
>I sort of suspect that Intel managed to put something together that is
>faster than the average taken between L1 and L2. Sorta like you have a
>Large but slow L1 cache and no L2 cache, or else a very fast L2 cache,
>and no L1 cache. Maybe our Intel engineers can help out here. This is
>why I'm buying me a P/Pro - to see how it fits together!!!

PPro has a L1 cache and a separate, but still in the die, L2 cache.
The L2 cache runs at the Clock speed of the CPU, most external L2
caches do not.

>> Our few MMX Machines tend to be really fast, even for things
>>that don't use the MMX extensions. This is primarily because of
>>the huge (for a Pentium) on-chip cache. If money is an object,
>>go with the MMX. If it isn't (and you don't mind having your
>>office get warmer) , you probably want a 4 Proc PPro 200.
>
>Oi - what's a 4-CPU M/b cost? But I agree fully with Adam here. I
>think buy the fastest machine you can afford. You can add Memory,
>Bigger Hard Drives, faster Display cards, better network adapters,
>bigger screens, etc. Later when need arise (Tho it's sometimes
>difficult to convince your boss that you need a bigger screen, the
>others is a matter of statistics)

OK, I'm drooling at the thought of a 4 PPro Warp SMP machine...

>>>assembly version of the key cracker for MMX yet, but it will
>>>give me
>>>at least some indication. Does anyone know the difference
>>>between the
>>>two?
>> I'm not sure if it's going to make much difference. Around
>>here, it looks as though the main speed increase comes with the
>>raw clock speed. Our PPro 166 machines aren't doing keys much
>>faster then our Pentium 166 machines. I don't think that the
>
>Now I wonder why I can't find any P/Pro 166s sold anywhere.... But I
>have to admit this sounds odd. Are you using the right clients on the
>right machines?

Yes, it doesn't suprise me at all based on my experience. I don't know
if the PPro code isn't optimized as well or if it's just simple, and
thus that closely related to clock speed...

>>Cache is going to make a whole lot of difference (although I
>>could be wrong--we have not run the client on the MMX ones yet).
>>I'll look into this.
>
>>>my Linux box. Is it still possible to set up a guest account
>>>on Win
>>>95? I suspect that this is impossible, but then again, I have
>>>never
>>>used Win 95 [ Yes, it's 1997 now, but I'm still proud of that
>>>fact
>>>that I haven't used Win 95 :) ]. If it is possible, I would
>>>not mind
>>>setting up a guest account at all.
>
>You can install some Remote Shell Daemon thingy on Windows95 which
>allows it's own special client to execute commands on your machine.
>It's not even close to Telnet-ing to that W95 machine, but you can run
>commands on other machines using this. I've never had a case where it
>was worth the trouble of doing this, but you can find a Remote-Shell
>Client and Daemon pair on The http://Windows95.com site - Well, there is
>at least 3 different ones. I can't remember where for sure, but I think
>it was under networking stuff/miscelaneous. It wasn't under Graphics.
>:) :)

It probably won't help do what he wants to do, which is let them jump
on his machine and compile and optimize a program.

Colin L. Hildinger

Games Editor - OS/2 e-Zine!
http://www.os2ezine.com/

The Ultimate OS/2 Gaming Page
http://www.ionet.net/~colin/games.html

The Official Unofficial AWE32 and OS/2 Warp Page
http://www.ionet.net/~colin/awe32.html