Jetson TX2 memory throughput

I’m running generic benchmarks before we develop our application and I’m seeing different numbers that are stated in the tech specs:

root@tegra-ubuntu:/tmp/mbw# ./mbw 512
Long uses 8 bytes. Allocating 2*67108864 elements = 1073741824 bytes of memory.
Using 262144 bytes as blocks for memcpy block copy test.
Getting down to business… Doing 10 runs per test.
0 Method: MEMCPY Elapsed: 0.14423 MiB: 512.00000 Copy: 3549.910 MiB/s
1 Method: MEMCPY Elapsed: 0.14417 MiB: 512.00000 Copy: 3551.314 MiB/s
2 Method: MEMCPY Elapsed: 0.14417 MiB: 512.00000 Copy: 3551.240 MiB/s
3 Method: MEMCPY Elapsed: 0.14416 MiB: 512.00000 Copy: 3551.585 MiB/s
4 Method: MEMCPY Elapsed: 0.14414 MiB: 512.00000 Copy: 3552.176 MiB/s
5 Method: MEMCPY Elapsed: 0.14409 MiB: 512.00000 Copy: 3553.310 MiB/s
6 Method: MEMCPY Elapsed: 0.14628 MiB: 512.00000 Copy: 3500.113 MiB/s
7 Method: MEMCPY Elapsed: 0.14430 MiB: 512.00000 Copy: 3548.090 MiB/s
8 Method: MEMCPY Elapsed: 0.14424 MiB: 512.00000 Copy: 3549.590 MiB/s
9 Method: MEMCPY Elapsed: 0.14417 MiB: 512.00000 Copy: 3551.388 MiB/s
AVG Method: MEMCPY Elapsed: 0.14440 MiB: 512.00000 Copy: 3545.805 MiB/s

This is the benchmark I’m running https://github.com/raas/mbw
What surprises me is that the numbers (3.5 GB/s * 2 = 7 GB/s) is far from what is advertised (8GB 128-bit LPDDR4 @ 1866Mhz | 59.7 GB/s). Is there a sound explanation for this discrepancy?

Thanks
Janis

Some reference for this topic:

https://devtalk.nvidia.com/default/topic/1030434/jetson-tx2/ram-speed/

Thanks, it provided the insight I needed.

For reference: after executing nvpmodel -m 0 and jetson_clocks.sh the numbers went almost to twice of the original ones. Moreover, with 6 running threads the sustainable mempcpy was 27.5 GiB/s when block size was 1MiB.