Is there an improvement in throughput from installing a 32-bit operating system on a 64-bit machine?

Knuth recently objected to 64-bit systems, stating that for programs that fit in 4 gigabytes of memory, they "effectively throw out half the cache," because pointers are twice as large as on a 32-bit system.

My question is: can this problem be avoided by installing a 32-bit operating system on a 64-bit machine? And are there any bandwidth tests that demonstrate an advantage in this case?

+5
source share
4 answers

Answer: yes, it can to some extent, although the difference in performance is unlikely to be large.

Any test to verify this will have to make many pointers that will be difficult to separate from noise. Designing a test that will not be optimized is difficult. This article about failed java tests was published by someone in response to another question, but many of the principles described in it will apply to this.

+4
source

. , , , . , LK 16 : , 2 ^ 14/2 ^ 2 = 2 ^ 12 = 4096 32- , 2048 64- . 2K , 32- . , , , . , 64- , , 64- .

+6

, 64- . , 64- 4 ( , , , ). , , , . , 64- , 32- , , Intel Core Duo.

1 - 32-, , APPZ, . , 32- , /, 64- .

2 - 64-, , APPZ, . , 64- , /, 64- . , 4 , 64- . , ?

3 - 64- , / 0x00000000 - 0xFFFFFFFF ( !!!). ABI , /, /, 32 , 64- . , /ABI 64- . , 32-, APPZ 64-, 64- . , thunking; -P

:

, . , , . , " , , ". ...

+4

- ( x86) - 64- 32- .

64- :

  • 4
  • , , .

32- :

  • , .

:

  • . HD.
  • . ...

, , , . , ​​ , .

, , . 5% - , 32-.

, . "" 32- FireFox 64- KUbuntu (, )

+2

All Articles