Memcpy latency and bandwidth | The CUDA Handbook

← All benchmarks · Chapter 6

Some of these measurements depend on the host as much as on the GPU: launch overhead is CPU dispatch cost, transfer rates are bounded by the host link and the host's memory system, and the CPU baselines run entirely on the host. Results for one GPU are reported separately per host architecture whenever it has been measured on more than one.

![Figure 6-4 (book): small host-to-device memcpy, book-era vs community hardware](/book/media/ch6/figure-6-4.svg) ![Figure 6-5 (book): small device-to-host memcpy, book-era vs community hardware](/book/media/ch6/figure-6-5.svg)

memcpy.dtoh.breakeven (us) -- concurrency/breakevenDtoHMemcpy.cu: time by size, 4K-64K step 4K

GPU	4096	8192	16384	20480	24576	32768	36864	45056	49152	53248	61440	65536	n
Ampere
GeForce RTX 3060	1.08	1.08	1.3	1.46	1.62	1.93	2.1	2.39	2.54	2.71	3.02	3.17	2

memcpy.dtoh.null-latency (us) -- concurrency/nullDtoHMemcpyAsync.cu: 0-byte device-to-host memcpy

GPU	us	n	range
Ampere
GeForce RTX 3060	1.06	2	1.06--1.06

memcpy.dtoh.null-latency-sync (us) -- concurrency/nullDtoHMemcpySync.cu: 0-byte device-to-host memcpy + sync

GPU	us	n	range
Ampere
GeForce RTX 3060	3.61	2	3.59--3.63

memcpy.dtoh.pinned-bandwidth (GB/s) -- memory/pinnedBandwidth.cu: 64MB pinned device-to-host, 100 iterations

GPU	GB/s	n	range
Ampere
GeForce RTX 3060	26.8	2	26.8--26.81

memcpy.htod.breakeven (us) -- concurrency/breakevenHtoDMemcpy.cu: time by size, 4K-64K step 4K

GPU	4096	8192	16384	20480	24576	32768	36864	45056	49152	53248	61440	65536	n
Ampere
GeForce RTX 3060	1.3	1.46	1.71	1.92	2.01	2.33	2.46	2.78	2.92	3.09	3.39	3.55	2

memcpy.htod.null-latency (us) -- concurrency/nullHtoDMemcpyAsync.cu: 0-byte host-to-device memcpy

GPU	us	n	range
Ampere
GeForce RTX 3060	1.51	2	1.49--1.53

memcpy.htod.pageable-staged (GB/s) -- concurrency/pageableMemcpyHtoD.cu: the book's staged pageable host-to-device memcpy (pinned double-buffer)

GPU	GB/s	n	range
Ampere
GeForce RTX 3060	14.67	2	14.45--14.89

memcpy.htod.pinned-bandwidth (GB/s) -- memory/pinnedBandwidth.cu: 64MB pinned host-to-device, 100 iterations

GPU	GB/s	n	range
Ampere
GeForce RTX 3060	26.8	2	26.8--26.81

Submissions

rtx-3060 · nwilt-2026-07-12-2.json -- Nicholas Wilt · driver 580.95.05 · CUDA 12.9
rtx-3060 · nwilt-2026-08-27.json -- Nicholas Wilt · driver 580.95.05 · CUDA 13.0