After updating to the new 368 bios, my NUC 5i5RYH was very slow. The machine is basically only used for web browsing and some other light tasks as a secondary PC, so I'm not asking much of it. Just running the Windows file browser and web browsers was very slow.
-Responsiveness was very sluggish. It felt like an old atom PC with a slow spinning drive.
-Program launching was slow and would often stall. I updated some drivers right after the bios update and some the installers would just sit there for minutes with no progress. Then they would eventually finish.
-Even maximizing a window was slow and would take a few seconds.
-Browsing directories was slow.
The difference was very noticeable.
On benchmarking with Crystal Disk Mark, random 4k read/write SSD performance with high Que Depth was more than cut in half.
This is with the following SSD: SAMSUNG 850 EVO M.2 2280
Sequential Read (Q= 32,T= 1) : 550.542 MB/s
Sequential Write (Q= 32,T= 1) : 528.400 MB/s
Random Read 4KiB (Q= 32,T= 1) : 99.827 MB/s [ 24371.8 IOPS]
Random Write 4KiB (Q= 32,T= 1) : 86.986 MB/s [ 21236.8 IOPS]
Sequential Read (T= 1) : 480.189 MB/s
Sequential Write (T= 1) : 463.596 MB/s
Random Read 4KiB (Q= 1,T= 1) : 25.436 MB/s [ 6210.0 IOPS]
Random Write 4KiB (Q= 1,T= 1) : 64.319 MB/s [ 15702.9 IOPS]
Sequential Read (Q= 32,T= 1) : 549.253 MB/s
Sequential Write (Q= 32,T= 1) : 529.140 MB/s
Random Read 4KiB (Q= 32,T= 1) : 221.303 MB/s [ 54029.1 IOPS]
Random Write 4KiB (Q= 32,T= 1) : 183.992 MB/s [ 44919.9 IOPS]
Sequential Read (T= 1) : 484.048 MB/s
Sequential Write (T= 1) : 460.311 MB/s
Random Read 4KiB (Q= 1,T= 1) : 31.335 MB/s [ 7650.1 IOPS]
Random Write 4KiB (Q= 1,T= 1) : 77.179 MB/s [ 18842.5 IOPS]
I then went back to the 367 bios and all the above issues were resolved. The system is back to its usually fast responsiveness.
This seems similar to this 5i3RYH issue with the 368 bios:
Overall, something seems off about this update. Hopefully the Spectre/Meltdown microcode fix can be improved upon so that it does not destroy I/O performance on the NUC in Windows.
Benchmarks (other than storage ones) won't fully reproduce this. The best way to reproduce this would be to just gauge the responsiveness of the system before and after the 368 update.