- Joined
- Jun 20, 2001
- Location
- Vancouver, WA
This question was posed by Demont in the thread "Are you going to go 64-bit", and I believe it to be a good question to debate on. So here it is!
Here's my thoughs on this -
Yes, I think I would probably buy an Athlon-64. Why? Because there's nothing to loose (probably), and a little (to a lot) to gain.
What is there to loose by going 64-bit?? 'Legacy' 16 and 32 bit apps will run fine, with absolutely no performance hit caused by the 64-bit architecture. To the program, your brand new chip will seem exactly the same has it allways has. It will use the same registers it allways has, and will run identically internaly. It won't even need* a recompile. Any performance increase will be due wholly to improvements of the chip (as any generation would), and to clock speed/IPC.
Also, apps which are compiled for 64-bit need not make use of the astounding multi-terabyte memory limit to have any good done to them. With the Athlon-64, programmers and compilers get 8 new general purpose registers to work with, and so can keep more data directly on the core (not in L1/L2 cache, but directly on the CPU's 16 'variables'). This means less pushing and popping from the stack, and increased instruction speed. Results have shown that this increase is on the order of only 10%-30%, but what's wrong with that?? The Barton over the Throughbred dosen't have gigantic performance increases, so we shouldn't expect 64-bit alone to magically double our speed when that's not what 64-bit is a measure of (in fact, I'll bet 90% of those performance increases are due to core redesigns and the avalibility of the 8 new registers)....
OK, so I've listed a lot that we won't loose by going 64-bit, and even a little we will gain. But what will we loose?? Pretty much the only thing I can think of is money. No doubt, when this chip debuts, it's going to be pretty dang expensive. Probably somewhere near the Barton's debut price. But in time, that price will drop, and come down to more reasonable levels.
No performance hit from 64-bit technology for existing apps
+ Highly probable performance increase because of core redesign/clockspeed increase
+ Performance increase (though admittadly small) from recompile to 64-bit (even if >4GB memory isn't used)
-------------------------
Highly probable perfomance increase across the board which is greater than or equal to a new core alone, and all for the cost (probably) a new core.
* - actually, apps that run in real mode and virtual mode WILL require a recompile, however, real and virtual mode were made obsolete with the advent of the 386.
OK, your turn to spew forth!
JigPu
Here's my thoughs on this -
Yes, I think I would probably buy an Athlon-64. Why? Because there's nothing to loose (probably), and a little (to a lot) to gain.
What is there to loose by going 64-bit?? 'Legacy' 16 and 32 bit apps will run fine, with absolutely no performance hit caused by the 64-bit architecture. To the program, your brand new chip will seem exactly the same has it allways has. It will use the same registers it allways has, and will run identically internaly. It won't even need* a recompile. Any performance increase will be due wholly to improvements of the chip (as any generation would), and to clock speed/IPC.
Also, apps which are compiled for 64-bit need not make use of the astounding multi-terabyte memory limit to have any good done to them. With the Athlon-64, programmers and compilers get 8 new general purpose registers to work with, and so can keep more data directly on the core (not in L1/L2 cache, but directly on the CPU's 16 'variables'). This means less pushing and popping from the stack, and increased instruction speed. Results have shown that this increase is on the order of only 10%-30%, but what's wrong with that?? The Barton over the Throughbred dosen't have gigantic performance increases, so we shouldn't expect 64-bit alone to magically double our speed when that's not what 64-bit is a measure of (in fact, I'll bet 90% of those performance increases are due to core redesigns and the avalibility of the 8 new registers)....
OK, so I've listed a lot that we won't loose by going 64-bit, and even a little we will gain. But what will we loose?? Pretty much the only thing I can think of is money. No doubt, when this chip debuts, it's going to be pretty dang expensive. Probably somewhere near the Barton's debut price. But in time, that price will drop, and come down to more reasonable levels.
No performance hit from 64-bit technology for existing apps
+ Highly probable performance increase because of core redesign/clockspeed increase
+ Performance increase (though admittadly small) from recompile to 64-bit (even if >4GB memory isn't used)
-------------------------
Highly probable perfomance increase across the board which is greater than or equal to a new core alone, and all for the cost (probably) a new core.
* - actually, apps that run in real mode and virtual mode WILL require a recompile, however, real and virtual mode were made obsolete with the advent of the 386.
OK, your turn to spew forth!
JigPu