I understood that it wouldn't exactly be double since I'm not using a proper controller and relying on the AMD CPU, but this was really unexpected. Not to say all of it got worse, SEQ1M Q1T1 on CrystalDiskMark for writes did improve by about 10% but that's it.
I'm running:
- Pro WS X570-ACE Motherboard
- AMD Ryzen 9 3950X
- 2x XPG S40G 4TB
- 4x 32GB 2666MHz Samsung DDR4 UDIMM
WITHOUT RAID, CrystalDiskMark 8.0.1 x64 5x 1GiB:
- SEQ1M Q8T1: 3700MB/s Read, 2800MB/s Write
- SEQ1M Q1T1: 3700MB/s Read, 2200MB/s Write
- RND4K Q32T1: 500MB/s Read, 500MB/s Write
- RND4K Q1T1: 60MB/s Read, 700MB/s Write
WITH RAID 0, CrystalDiskMark 8.0.1 x64 5x 1GiB:
- SEQ1M Q8T1: 3200MB/s Read, 3100MB/s Write
- SEQ1M Q1T1: 2400MB/s Read, 2200MB/s Write
- RND4K Q32T1: 500MB/s Read, 400MB/s Write
- RND4K Q1T1: 60MB/s Read, 300MB/s Write
So, slightly improved writes sequentially. Worse sequential reads, and significantly worse random 4K Writes.
RAID SETTINGS:
- RAID 0
- Stripe Size: 64KB
- Read Cache
- Write Back Cache
NVMe RAID is properly enabled and configured. Proper drivers were installed for the proper Operating System (Windows 10) and chipset (X570) during Windows installation. I've worked with these exact drives, CPU, memory, and motherboard for work but we use a RAID controller for that (even though the controller isn't even proper hardware RAID, it seems to have it's own chip(s) on there to do most the processing.) So it has to be something up with AMD-RAID. And yes, the drivers were also installed on Windows too (AMD Chipset software, AMD RAID software.)
I looked at a bunch of benchmarks between 64KB, 128KB, and 256KB stripe size and it really doesn't seem like it could be the culprit here. I searched AMD forums and the internet, information on this is unfortunately lackluster.
Edit -- Just in case it's helpful, I've disabled all caching and that resulted in 2200MB/s read instead of 2400MB/s SEQ1M Q1T1 and 2600MB/s write instead of 2400MB/s (for the same category.) And it reduced the RND4K Q1T1 write speed to 145MB/s (from 300MB/s.)
