Good morning!! Today I’ve discovered something which I can’t quite explain. I’ve performed tests for pageable and pinned memory in a 1080 Ti and a 2080 Ti, for ~8MB of data to be more precise. The reason of choosing this amount of data is because it is the size of data I use for many of my PhD tests. When I perform the bandwidth tests on both cards, I get the following results for this 8MB chunk of data:
1080 Ti:
Paged Memory Bandwidth: 7.5 GB/s
Pinned Memory Bandwidth: 12.725 GB/s
2080 Ti:
Paged Memory Bandwidth: 12.746 GB/s
Pinned Memory Bandwidth: 13.182 GB/s
The results are similar for both DtoH and HtoD. I got the best numbers from several tests. I know that pinned memory is not necessarily faster than paged memory but still surprises me. Was there any improvement from Pascal to Turing that I am not aware of? I’ve tried googling for it but without luck. The machines have a different CPU/RAM as well, the 1080 Ti is paired with an intel i7-3770 with DDR3 memory while the 2080 Ti is paired with an i9-9900K with DDR4. I do not know if that is significant when performing this kind of tests.
Thanks in advance to anyone who sheds some light on this matter :)