My question is: Can this problem be avoided by installing a 32-bit operating system on a 64-bit computer? Are there any bandwidth-intensive benchmarks that prove the advantage in this situation?
Any benchmark for testing it must perform a lot of pointer resolution, which is difficult to separate from the noise. It is very difficult to design a benchmark that will not be optimized. Someone is replying to another This article about flawed java benchmarks was published in a question, but many of the principles described in it are applicable to this.
Knuth recently objected to 64-bit systems, saying that it is suitable for 4 GB memory programs, “they effectively discarded half of the cache”, because the pointer is twice that of a 32-bit system.
My question is: by installing 32-bit on a 64-bit computer Can the operating system avoid this problem? Are there any bandwidth-intensive benchmarks that prove the advantage in this situation?
The answer is: Yes, it can be to a certain extent, although the performance difference is unlikely to be large.
Any benchmark to test it must perform a lot of pointer resolution, which is difficult to separate from the noise. It is very difficult to design a benchmark that will not be optimized. Someone posted This article about flawed java benchmarks in response to another question, But many of the principles described therein apply here.