64-bit vs 32-bit

[Dukefrukem]

from Crytek people:

"Yes, 64-bit in general runs better than 32-bit. In fact I would recommend gamers run 64-bit only under very high configurations." i

Q&A: Crytek CEO specs out Crysis - PC News at GameSpot

Actual Tests:

[From the first test Take a look at the low resolution and settings figures though: on such a system, the tests were a cakewalk and this time the numbers were consistent and always had the 64 bit mode running slower than 32 bit.
...
[From the second test But look at the comparable settings to first system: DX9 Low at 800 x 600. Here we see that the 64 bit mode is still slower, albeit by a tiny fraction, than the 32 mode.... nearly. The High GPU test persistently gave the same sized gap with each run though, so there we have it: 64 bit faster than 32 bit, but only as a minimum value - the average and peak were essentially the same.

YouGamers - News - Crysis 32 vs 64 bit - the real story?

[Relly]

nice find Dukie

[Dopefish]

I expected this. There's nothing about a 64-bit system that should make it quicker.

Four benefits to running 64-bit over 32-bit:
Developers: Extra registers to use
Developers: Higher data ranges for integers.
Developers and Users: >4gb of memory
Users: It makes your dick bigger


Okay, so we have summed up the differences. How are any of these supposed to improve performance? They're not. 64-bit is 32-bit with a 64-bit extension. In a huge majority of tests out there, 64-bit hurts performance.

If anything, the extra overhead and hacked-together 64-bit operating system by Microsoft will definitely cause more problems.

I'll upgrade to 64-bit when I feel the need to exceed 3.2gb of memory.

[VincentVega]

One of the best uses of 64 bit (or simply having wider instructions) is that you can A) work with larger numbers and B) deal with more data in a given instruction. From this standpoint, it's a vastly superior architecture. more data in an instruction (such as bringing in 6 bytes instead of 4) means you can get done in one clock cycle what it used to take you in 2 or even more (if you count a branch miss, or a fetch instruction, blah blah blah).

simply put, wider instructions means you can add new instructions and extensions, that weren't possible before, and optimize things down to one hardware instruction that used to take more. It's like a temporary clock speed doubling or tripling for those few issues.

Since I'm a DBA, I want my RDBMS to be a 64 bit version. But I still won't opt for a 64 bit O/S because some other things still don't work the best with it.

[Dopefish]

I was going to mention something similar to that in my original post but decided to exclude it because most people here are not developers. Yes, there is indeed room for improvements, but only in very specific areas. This won't apply to video games or 99% of the software out there. This will apply to high-end corporate products and number/data crunching areas. Keep in mind that this is dealing with integers only. It has nothing to do with floating point operations in any way.

Also, to get the most out of 64-bit, even when developing for it, you need to make sure the compiler you're using will do a good job at optimizing the code. However, the explanation and details on this probably don't belong in this thread.

For the average user, you really want to be running 32-bit. End of story.

[VincentVega]

I've been running XP64 for 2 years, part of a free upgrade program.

I hate it.

This is why they invalidate your 32 bit license key when you do it.

[Dopefish]

I decided to mention some possible scenarios where a 64-bit application can be faster. If a developer takes advantage of the extra registers, there is room for improvement in performance. The use of more registers allow you to do more at once without having to waste cycles on reusing the same registers over and over. You can store data to be called later without having to store it in memory and call back upon it later. This all applies to the highly optimized compiler mentioned above. If the compiler you're using can take advantage of all of this, then it could perform better. One thing to keep in mind is that if you don't write in assembly yourself, or writing your own libraries, you are calling upon all the libraries that are part of Windows. Chances are these libraries are not optimized at all; They are more than likely just recompiles of the 32-bit versions.

Examples of 64-bit performance are out there. In all game benchmarks that are 64-bit, they are slower than their 32-bit counterpart. You will, however, notice that high-end business software will often perform faster, albeit minimally.

[VincentVega]

I think the world should compromise on 48bit

[Dopefish]

Funny you should mention that, since 64-bit processors' virtual address space is actually only a 48-bit implementation. There are no current plants to allow more at the moment, either, since the highest used right now is 44-bits. It will eventually happen, but nothing is on the table at the moment.

[VincentVega]

I wonder if one of the biggest possible advantages for 64bit might be in integrated graphics scenarios? wider is always better in those cases, case in point the ridiculous 320bit memory bus on video cards