InvestorsHub Logo
icon url

HailMary

12/18/03 2:32 PM

#21122 RE: chipguy #21116

Everybody, please stop using this number, it is utter BS.

I don't know why you are replying to me with a quote I didn't make. I never claimed 400. I just illustrated the concept of out of order execution.

P4 is 128 GPRs (integer), right?
icon url

HailMary

12/18/03 3:32 PM

#21136 RE: chipguy #21116

That gives a total of 52 GPRs (36 rename + 16 architected).

BTW - I think this figure is wrong. I don't have copies of MPR handy to look at. Can you double check what it says (or did you already do that)?

Instructions in flight is listed as 72 for Opteron, which implies more rename registers than you quote.

I'm sure you have more knowledge in this area than me (I'm more software-ish), so I will concede to your expertise here. I keep up with this area because software/hardware face similar problems in optimization and tradeoffs.
icon url

Tenchu

12/19/03 1:51 AM

#21193 RE: chipguy #21116

Chipguy, Microprocessor Report states the number of GPR rename regs in Opteron is 36.

400 seems way too much, but 36 seems way too small. Could you double-check the figures? Also, know that MPR sometimes gets their numbers wrong.

Tenchu