In an effort to determine what the true performance benefit of 64-bit computing is, TheJemReport designed this benchmarking project using FreeBSD. The article can be found here. The article compares the performance of three test cases (Intel P4 Prescott 3.2E and Athlon64 3200+ in 32-bit mode and 64-bit mode) using stopwatch tests, openssl, and two synthetic benchmarks from the Ports system.
The 64-bit Question: AMD64 vs. i386
Submitted by Jem Matzan 2004-03-15 Hardware 13 Comments
The bouncing icon startup notification is pretty. I wish Gnome had that. I found KDE 3.2 to be fast, but still KDE. The annoying thing was that even though there were options everywhere, I couldn’t configure virtual desktops to my liking (3×3) and KWin did not seem to have any keybindings for geometric desktop switches out of the box, I had to set them myself. Why not use the same keybindings metacity and OpenBox do, they seem to make a lot of sense.
oops, wrong story
I think you may have the wrong thread. This was a nice review of how the latest processors match up. It will be interesting to see the benchmarks when the optimazations come out for the prescott and the new AMD64.
does any one know any links for how to program in amd64 assembler? not even tutorials but an overview of the ops and whatnot? Is it RISCy in AMD64 mode?
Start on this page here…
…I’m too lazy to dig any deeper for you than that right now.
Get the programming manuals from AMD at the link. They cover ALL assembly language programming for the AMD64.
Then get YASM at the link to do AMD64 assembly language. NASM hasn’t been updated for AMD64. You can also use gas if you are a gcc person.
but not unexpected. However, the reason why the AMD64 was most often faster is not because of magical 64-bit goodness, but because in 64-bit mode it makes use of twice the number of registers. A simular test on something like a UltraSparc, as shown on osnews.com in the past, usually shows 32-bit to be slightly faster. If I ever get an AMD64 box I’ll likely use FreeBSD. 🙂 Ohh, kinda off-topic but does anyone know if Intels 64-bit prosumer (non-Itanium) P4 solution also, like the AMD64, uses twice the number of registers in 64bit mode? I would hope so.
me too. i’m currently using FreeBSD on IA32. if i get a AMD64 , sure i’ll using FreeBSD.
Yup, the Intel version of x86-64 has the same number of registers (16). The operation sets should be (mostly) compatible, so they will probable have a common subset.
Intel CEO Craig Barrett said: “For the most part, (software) will run on both systems” http://www.theregister.co.uk/content/61/35628.html
The number of registers should mean better performance, and it’s about bloody time the x86 get more than 8 registers.
Also, those 16 registers are wider. Keen programmers and compiler writers can split those registers.. so that is double if you are using 32bit values.
(Complete guess, is this statement correct?)
I doubt this will be the case as AFAIK you can’t address the second part of those 64bit registers separately but instead have to shift in order to get at the second 32bit.
But it could be the case that a shift is still faster than a memory access.
And maybe also that might stop some parallism too..
But it still might be a neat trick. Not sure if current compilers use subregisters (AL, AH, whatnot) anyway.
Not worth it, unless you can ensure that the lower 32-bit does not affect the upper 32-bit
Example: 1 + 2 = 3
6 + 8 = 14
However: 16 + 28 = 44 (and not 3,14) Basically, the lower half carried over and altered the results of the upper half of the calculation.
You can do math to figure out whether the lower half (6,8) could have carried over and altered the upper half, but by then, you’re doing extra operations and pretty much negated any sort of benefit.