SI
SI
discoversearch

We've detected that you're using an ad content blocking browser plug-in or feature. Ads provide a critical source of revenue to the continued operation of Silicon Investor.  We ask that you disable ad blocking while on Silicon Investor in the best interests of our community.  If you are not using an ad blocker but are still receiving this message, make sure your browser's tracking protection is set to the 'standard' level.
Politics : Formerly About Advanced Micro Devices -- Ignore unavailable to you. Want to Upgrade?


To: John Evans who wrote (123066)8/25/2000 11:49:47 AM
From: Ali Chen  Read Replies (2) | Respond to of 1584949
 
<..average number of dependencies in twenty instructions of x86 code>

I believe there were some academic studies about
average paralellism capabilities of x86 codes,
and it was found that the ILP is about 3.5-4,
largerly due to data dependencies. I also believe that
the register renaming technique somewhat relaxes
this, but not much. So I would guess that for
a 20 stages there will be 3 unresolved data dependencies
at least.

Regards,
- Ali



To: John Evans who wrote (123066)8/25/2000 1:07:29 PM
From: Petz  Respond to of 1584949
 
Very cool post, John. Welcome to the thread.

From what I understand, Intel added a lot of stuff to SSE2 so that considerable efficiencies can be gained by recoding FPU code to use the new instruction set rather than the archaic x86 FPU instructions. That worries me, all Intel has to do is get the 3 most-benchmarked games and DirectX to be rewritten using the new instructions and it will make the Intel FPU look as good as the Athlon FPU.

Intel recognized three things:
1)The Athlon FPU is superior and couldn't be beat anyway
2)By effectively changing the FPU instruction set, the Intel FPU can be made to look as good as the Athlon FPU. This pushes the cost of developing a good hardware FPU on the software industry.
3)Intel has control of CPU benchmarks so they will either be integer-oriented or use SSE2 instructions.

Petz