6.25.5400 vs 6.30.5500 Lower RAM Copy result

[Deepcuts]

Hello,

X570 Aorus Xtreme F30 BIOS
AMD 3950X
G-Skill 3600 16-16-16-36

Using 6.25.5400, Cache and Memory Benchmark shows very different results for Copy comparing to the newer 6.30.5500.
On my setup, with 6.25.5400 I get around 60.500 MB/s Copy, while on the latest 6.30.5500 I get only 53.000 MB/s.
Read, Write and Latency values are the same.
That is quite the difference.
Repeated the test multiple times. Even cleared CMOS to be sure it is not something on my end.
Also, it looks to me like the copy takes a lot longer to complete compared to previous version.

Looks like a bug.

[Fiery]

1 hour ago, Deepcuts said:

Hello,

X570 Aorus Xtreme F30 BIOS
AMD 3950X
G-Skill 3600 16-16-16-36

Using 6.25.5400, Cache and Memory Benchmark shows very different results for Copy comparing to the newer 6.30.5500.
On my setup, with 6.25.5400 I get around 60.500 MB/s Copy, while on the latest 6.30.5500 I get only 53.000 MB/s.
Read, Write and Latency values are the same.
That is quite the difference.
Repeated the test multiple times. Even cleared CMOS to be sure it is not something on my end.
Also, it looks to me like the copy takes a lot longer to complete compared to previous version.

Looks like a bug.

No, it's actually normal.  We've had to increase the block size and cache invalidation scope in order to avoid the influence of the large L3 cache of Zen 2 and Zen 3 processors.  The longer execution time of the benchmark is caused by the more thorough cache invalidation, so that's normal too.  We do have plans to improve on that, but the combination of high amount of logical processors (32 in your case, incl. the SMT virtual CPU cores) and large L3 cache makes it very difficult to perform the cache invalidation properly without it taking loooong.

[Deepcuts]

Thank you. Understood.