On Friday 25 July 2003 23:09, H�vard Wall wrote:
> AMD XP/MP: 128KB L1 cache (64KB instructions, 64KB data)
>            +512KB L2 cache
>            =640KB total cache
>
> Intel Pentium 4 <= 2GHz:
>               8KB L1 data cache
>            + 12KB L1 instruction cache
>            +256KB L2 cache
>            =276KB total cache
>
> Intel Pentium 4 >= 2GHz:
>               8KB L1 data cache
>            + 12KB L1 instruction cache
>            +512KB L2 cache
>            =532KB total cache
>
> Actually, P4 have 12K micro-ops cache, not 12KB, whatever that means in
> practice.
> Of course, AMD Duron and Intel Celerons are cheaper products with less
> chache.
>
So that means Athlons have a relatively LARGE cache? As I (think) I said, I'm 
not sure about the Pentiums and am just going by what I remember. Although, 
come to think of it, a 64KB block of code without jumps, etc sounds pretty 
huge. Either way, that doesn't change the fact that I still seem to get a 
fairly large performance gain with -O2.

To be certain my original CFLAGS was "-march=athlon-xp -O3 -mmmx -msse -m3dnow 
-mfpmath=sse -pipe" which I changed to "-march=athlon-xp -O2 -mfpmath=sse 
-fomit-frame-pointer -pipe". Come to think of it the performance gain may be 
due to the omit-frame-pointer flag.  Comments?

Jason


--
[EMAIL PROTECTED] mailing list

Reply via email to