- Fri Jul 22, 2005 4:10 pm
#48571
Well I am only giving my opinion, that may be wrong as I am not aware of all the 64-bit ported products and how they compare with their equivalent 32-bit versions.
I would like to say that there is some kind of marketing strategy that pretends to impose the equation: 64-bit = faster, but I don't see that point.
64-bit hardware has been created to jump over the 32-bit limit for memory addressing, that is the famous 4 Gb. For me, 64-bit allows to manage huge dataset impossible to handle before. Do we always need this? Depends on the application.
Now Maxwell can handle 4-5 millions in 2 Gb. With a 64-bit machine and memory we can rise to any incredible amount of triangles. However, speed performance will not scale in the same way.
In our short experience with 64-bit platforms, we haven't got a significative increase in performance, but this may change as long as we keep working on them.
64-bit, 128-bit, whatever doesn't mean 'faster'. It just means that will address more memory, but we also have to look at the floating point units, paralelization schemes, compiler options, cache size, bus speed, if the code is well adapted to 64-bit structures, etc, etc. Take away from your minds the idea of 'more bits = faster'
I would like to say that there is some kind of marketing strategy that pretends to impose the equation: 64-bit = faster, but I don't see that point.
64-bit hardware has been created to jump over the 32-bit limit for memory addressing, that is the famous 4 Gb. For me, 64-bit allows to manage huge dataset impossible to handle before. Do we always need this? Depends on the application.
Now Maxwell can handle 4-5 millions in 2 Gb. With a 64-bit machine and memory we can rise to any incredible amount of triangles. However, speed performance will not scale in the same way.
In our short experience with 64-bit platforms, we haven't got a significative increase in performance, but this may change as long as we keep working on them.
64-bit, 128-bit, whatever doesn't mean 'faster'. It just means that will address more memory, but we also have to look at the floating point units, paralelization schemes, compiler options, cache size, bus speed, if the code is well adapted to 64-bit structures, etc, etc. Take away from your minds the idea of 'more bits = faster'
Next Limit Technologies


- By Jochen Haug