Re: Does NVDA take up a lot of ram in win10?
@13
The point about pointers in this context is actually so relevant that I'm reasonably sure I could write synthetic benchmarks for Windows that had 32-bit programs which only do algorithmic things (i.e. no kernel calls) which are anywhere from 10% to 30% faster than the same source compiled as 64-bit in a day or two. This isn't my area of expertise; someone whose area of expertise it is can probably do so faster. To give you some idea, the V8 people went way out of their way to do pointer compression and got very impressive results, including CPU performance increases.
This is now going to get so technical that it's beyond most people on this site and certainly beyond the scope of Off Topic. But there's no real penalty for being in 32-bit mode save the missing registers--and for most C programs, that's not really a problem. You kind of have to be doing tons and tons of math in just the right way to feel that pain. But the thing is that a cache miss is upward of 100 cycles, on processors that can do something like 8 math operations per clock cycle. And most programs, the ones that don't specifically design for cache friendliness, are going to have tons and tons of pointers. It's like doubling your internet bandwidth. The prefetchers get significantly more bang for their buck. The caches get significantly more bang for their buck, too. For most programs, you're doing tons of jumps and storing lots of pointers. These days the gap has kind of closed, it would certainly be harder than it used to be, but that's only because modern caches are really really big, and it could certainly still be done with some ingenuity. What I can say is that if you have an array of 64-bit integers vs. an array of 32-bit integers plus an offset that you need to add to all of them, the array of 32-bit ints will almost certainly be faster, up to the point where the program is doing enough operations between elements to hide the prefetch of the next chunk, which is honestly way more than you'd think it is.
Something like Synthizer would rather you use 64-bit processors with it, but even there the difference is really slight, and this is only the case because I am putting some thought into these issues. At some point there'll be benchmarks for Synthizer and then I guess we could have actual numbers, but there's kind of no point.
-- Audiogames-reflector mailing list Audiogames-reflector@sabahattin-gucukoglu.com https://sabahattin-gucukoglu.com/cgi-bin/mailman/listinfo/audiogames-reflector
