A clock cycle is the interval between consecutive pulses of a chip’s internal clock. GPU memory latency and instruction costs are counted in cycles, making them comparable across clock speeds. section 1.4
Frequency and latency
At 1.5 GHz, a GPU completes 1.5 billion cycles per second, so one cycle lasts under a nanosecond. Hopper-generation HBM access is measured at 479–656 cycles. section 1.4 section 1.4
Table 1.2 Measured memory access latency by hierarchy level, in GPU clock cycles, on Hopper-generation GPUs
| Memory level | Latency (cycles) |
Registers |
~1 |
Shared memory / L1 cache |
29 to 41 |
L2 cache |
258 to 744, since Hopper’s L2 is split into two partitions and the figure depends on which one a request hits and whether it is a hit or a miss |
Global memory (HBM) |
479 to 656 |
Compare cycle-counted access latencies across GPU memory levels. section 1.4

