RST/Raid VS AHCI Single NVMe drive (There's a significant difference!)

TL;DR - RST mode is significantly faster than AHCI mode, so if you don't require Linux on your machine, leave your bios set to RST mode. All test results are available in this imgur gallery https://imgur.com/a/0HR9HtQ

Hey guys, so I've been doing some doing a lot of benchmarking on my new Legion Y740's 1TB PM981 NVMe drive because I plan to replace it with a 1TB HP EX920 that I already have, so I can compare to see which drive is faster (I'll post these results later or tomorrow if anyone's interested).

Long story short, I wanted to dual boot linux as well, which requires you to be in AHCI mode to see the drive. Since I already had benchmarked PM981 a lot with the default settings/RST mode (should be the same or similar as RAID mode). I could not for the life of me find the difference the two modes made for non-raid setups, like my single NVMe configuration. Most of my searching just lead to people claiming there was virtually no performance difference, so I decided to run a few more tests and benchmark the same drive in AHCI mode to see if there's performance gain or loss.

I tested using CrystalDiskMark 7 Beta4 x64 using the "Real Word Performance [+Mix]" profile simply because queue depth 1 performance and the 70/30 read/write test is more relevant to real world use. I also tested with AS SSD, but got really strange seq. read numbers (always capped at 1k mb/s for some reason), so I just skipped including these results. As a bonus I also benched with UserBenchmark tool to see if it would verify my findings and testing, which it did. I did not run the TRIM command or any clean up tool like CCleaner at any point.

I tested on the first day with an entirely stock Y740 with no addition updates or changes. This is with RST mode enabled in the bios. (PS - Text results are a bit further below and might be easier to compare than these pictures)


Then again on the third day after some use. Still in RST mode.


Now after that test, I rebooted and tested again twice in AHCI mode, confirmed working as "Standard NVM Express Controller" under device manager instead of the Intel RSTe controller or whatever it usually is in RST mode.



Significant difference right? All three of the last three tests above were run consecutively in the same sitting, the only thing that changed in the last two tests was that the standard NVMe controller (AHCI mode) was used instead of the intel RSTe driver (RST mode).

The differences don't stop there. I saved the test results to text so I can see and compare everything including IOPS more easily.

Test 1 (Day 1, RST Mode)
[Read]
Sequential 1MiB (Q= 1, T= 1): 2339.229 MB/s [ 2230.9 IOPS] < 447.97 us>
Random 4KiB (Q= 1, T= 1): 47.077 MB/s [ 11493.4 IOPS] < 86.84 us>

[Write]
Sequential 1MiB (Q= 1, T= 1): 2396.870 MB/s [ 2285.8 IOPS] < 437.12 us>
Random 4KiB (Q= 1, T= 1): 138.856 MB/s [ 33900.4 IOPS] < 29.33 us>

[Mix] Read 70%/Write 30%
Sequential 1MiB (Q= 1, T= 1): 1967.188 MB/s [ 1876.1 IOPS] < 532.38 us>
Random 4KiB (Q= 1, T= 1): 58.282 MB/s [ 14229.0 IOPS] < 70.08 us>

Test 2 (Day 3, RST Mode)
[Read]
Sequential 1MiB (Q= 1, T= 1): 2394.951 MB/s [ 2284.0 IOPS] < 437.51 us>
Random 4KiB (Q= 1, T= 1): 46.770 MB/s [ 11418.5 IOPS] < 87.39 us>

[Write]
Sequential 1MiB (Q= 1, T= 1): 2397.587 MB/s [ 2286.5 IOPS] < 436.96 us>
Random 4KiB (Q= 1, T= 1): 136.784 MB/s [ 33394.5 IOPS] < 29.81 us>

[Mix] Read 70%/Write 30%
Sequential 1MiB (Q= 1, T= 1): 1811.030 MB/s [ 1727.1 IOPS] < 578.25 us>
Random 4KiB (Q= 1, T= 1): 59.678 MB/s [ 14569.8 IOPS] < 68.47 us>

Test 3 (Day 3, ACHI Mode)
[Read]
Sequential 1MiB (Q= 1, T= 1): 1637.455 MB/s [ 1561.6 IOPS] < 638.60 us>
Random 4KiB (Q= 1, T= 1): 48.819 MB/s [ 11918.7 IOPS] < 83.71 us>

[Write]
Sequential 1MiB (Q= 1, T= 1): 1980.046 MB/s [ 1888.3 IOPS] < 528.17 us>
Random 4KiB (Q= 1, T= 1): 118.345 MB/s [ 28892.8 IOPS] < 34.44 us>

[Mix] Read 70%/Write 30%
Sequential 1MiB (Q= 1, T= 1): 1821.950 MB/s [ 1737.5 IOPS] < 574.77 us>
Random 4KiB (Q= 1, T= 1): 56.761 MB/s [ 13857.7 IOPS] < 71.96 us>

Test 4 (Day 3, ACHI Mode)
[Read]
Sequential 1MiB (Q= 1, T= 1): 1618.866 MB/s [ 1543.9 IOPS] < 645.83 us>
Random 4KiB (Q= 1, T= 1): 45.470 MB/s [ 11101.1 IOPS] < 89.91 us>

[Write]
Sequential 1MiB (Q= 1, T= 1): 1936.693 MB/s [ 1847.0 IOPS] < 539.98 us>
Random 4KiB (Q= 1, T= 1): 118.793 MB/s [ 29002.2 IOPS] < 34.31 us>

[Mix] Read 70%/Write 30%
Sequential 1MiB (Q= 1, T= 1): 1549.789 MB/s [ 1478.0 IOPS] < 674.25 us>
Random 4KiB (Q= 1, T= 1): 54.214 MB/s [ 13235.8 IOPS] < 75.36 us>

Any new noticeable differences? Yes, even IOPS take a hit in AHCI mode.

I had already run the UserBenchmark test on the day I got my PC, so I figured since I had those results in RST mode, I could run it again in AHCI mode to see if it notices a difference. It definitely did. RST mode from day 1 scored 269%, AHCI mode in day 3 only scored 246%. Results here:
RST: https://www.userbenchmark.com/UserRun/20646518
AHCI: https://www.userbenchmark.com/UserRun/20680136

Personally I went into this very skeptical of RST, expecting there to be little to no difference if not within the margin of error. Was very surprised to see there be such a significant difference. I made sure to keep control variables like the testing environment as constant as possible to make sure results were unbiased as possible, but still expected variance to be within margin of error.

 

Bonus: I just switched back to RST after testing on ACHI, and tested again one last time to verify my findings. Here are my last RST mode results.

[Read]
Sequential 1MiB (Q= 1, T= 1): 2388.516 MB/s [ 2277.9 IOPS] < 438.62 us>
Random 4KiB (Q= 1, T= 1): 60.467 MB/s [ 14762.5 IOPS] < 67.58 us>

[Write]
Sequential 1MiB (Q= 1, T= 1): 2318.071 MB/s [ 2210.7 IOPS] < 451.85 us>
Random 4KiB (Q= 1, T= 1): 116.287 MB/s [ 28390.4 IOPS] < 35.06 us>

[Mix] Read 70%/Write 30%
Sequential 1MiB (Q= 1, T= 1): 1973.852 MB/s [ 1882.4 IOPS] < 530.63 us>
Random 4KiB (Q= 1, T= 1): 67.107 MB/s [ 16383.5 IOPS] < 60.87 us>

 

Have you tried ACHI with the Samsung NVMe driver? My guess is you would get similar results to RST.

 

Is there a driver from Samsung for the PM981? Last I tried to install a Samsung driver for this drive, it was incompatible.

 

Trying it now!

I had the same issue in RST Mode, I rebooted into AHCI mode then the driver installed without a hitch. I've rebooted now and am going to test it.

EDIT: Done testing. AHCI with the latest Samsung driver is faster but not as fast as RST Mode.



Test 1
[Read]
Sequential 1MiB (Q= 1, T= 1): 2010.076 MB/s [ 1917.0 IOPS] < 520.81 us>
Random 4KiB (Q= 1, T= 1): 51.335 MB/s [ 12533.0 IOPS] < 79.60 us>

[Write]
Sequential 1MiB (Q= 1, T= 1): 2346.887 MB/s [ 2238.2 IOPS] < 446.27 us>
Random 4KiB (Q= 1, T= 1): 132.042 MB/s [ 32236.8 IOPS] < 30.84 us>

[Mix] Read 70%/Write 30%
Sequential 1MiB (Q= 1, T= 1): 1811.208 MB/s [ 1727.3 IOPS] < 577.64 us>
Random 4KiB (Q= 1, T= 1): 57.770 MB/s [ 14104.0 IOPS] < 70.70 us>

Test 2
[Read]
Sequential 1MiB (Q= 1, T= 1): 1880.943 MB/s [ 1793.8 IOPS] < 556.42 us>
Random 4KiB (Q= 1, T= 1): 51.574 MB/s [ 12591.3 IOPS] < 79.25 us>

[Write]
Sequential 1MiB (Q= 1, T= 1): 2395.650 MB/s [ 2284.7 IOPS] < 437.17 us>
Random 4KiB (Q= 1, T= 1): 129.612 MB/s [ 31643.6 IOPS] < 31.40 us>

[Mix] Read 70%/Write 30%
Sequential 1MiB (Q= 1, T= 1): 1634.277 MB/s [ 1558.6 IOPS] < 639.89 us>
Random 4KiB (Q= 1, T= 1): 62.069 MB/s [ 15153.6 IOPS] < 65.80 us>

Edit 2: Booted back into RST Mode with the "Intel(R) Chipset SATA/PCIe RST Premium Controller" driver, with no intel software or anything installed/running.



[Read]
Sequential 1MiB (Q= 1, T= 1): 2319.805 MB/s [ 2212.3 IOPS] < 451.66 us>
Random 4KiB (Q= 1, T= 1): 49.054 MB/s [ 11976.1 IOPS] < 83.33 us>

[Write]
Sequential 1MiB (Q= 1, T= 1): 2393.699 MB/s [ 2282.8 IOPS] < 437.65 us>
Random 4KiB (Q= 1, T= 1): 138.938 MB/s [ 33920.4 IOPS] < 29.34 us>

[Mix] Read 70%/Write 30%
Sequential 1MiB (Q= 1, T= 1): 1988.224 MB/s [ 1896.1 IOPS] < 526.86 us>
Random 4KiB (Q= 1, T= 1): 57.839 MB/s [ 14120.8 IOPS] < 70.65 us>

Results still nicely lineup with older results.

 

I wonder if this could be specific to your computer model? I don't remember having this much difference going from AHCI to RAID mode on my Area-51m. I just reinstalled it a few days ago so don't feel like testing right now lol.

Maybe someone else can test on other models? @Papusan @Mr. Fox @S.K

Everyone I tagged has a different laptop brand lol

I didn't realize there's a new version with these options. I did a quick test on my M18x with 2x 840 Pro in RAID0:

 

raid on NVMe drives is pretty pointless, especially if you have a high end one that can easily do 3000 MB/s with just a single drive. all you're doing is upping the latency and higher risks for your files.

SATA on the other hand is understandable if someone wish to double their sequential.

 

The OP isn't running two NVMe drives in RAID 0, he is running it in raid mode vs AHCI...

 

Yeah I'm not running the kind of raid you're thinking about.. in fact raid mode has lower latency than AHCI in these tests. I have some suggested reading below if you care.

Here's a good explanation. (Source: https://www.reddit.com/r/hardware/c...ahci_mode_single_nvme_drive/f2u3gbn?context=3)

 

raid mode won't have lower latency than a single drive.
you are confusing and mixing AHCI/raid vs single drive/ raid arrays.

these will always hold true, it is and should be a rule applied when comparing similar things with very similar specs.

single stick of memory will have lower latency than duo dimms. 1 cpu has lower latency than 2 cpu, single core will have lower latency than duo cores, the same is applicable to single SSD vs two ssd in any raid array.

if you are talking about running a single SSD in AHCI mode vs raid mode, then theres be no difference. and quite honestly, in order to run in raid mode you'll need UEFI.

 

My dude, you have no idea what you're talking about. I'm not talking about RAID 0/1 or any of those multi drive arrays. The only confused person here is you. I'm talking about a motherboard setting that allows you to override the AHCI protocol your drive controller uses to use Intel's RST drivers. I don't know why you try to continue arguing when you don't have any idea what you're talking about. I'm not the only person here who's tried to clear your confusion. Not sure if you're just choosing not to read what's being said or just choosing to be ignorant on purpose. My laptop doesn't even supported raid with multiple drives. All the tests I've done here IS SINGLE DRIVE.

PS there is a difference. The whole point of all my testing was to see if there was a difference. Don't just arbitrarily say there isn't based on a guess. Especially when there's hard data and testing for you to see right there. And a very well informed explanation that I shared with you in my last response which I'm convinced your just choosing to ignore or just not read.

 

read the last part of my last reply before jumping on replying. i already stated if you talk about single drive for ahci vs raid mode, there are other factors involved which result in difference in performance that isn't coming from ahci vs raid.

uefi vs legacy, gpt vs mbr also affects performance, though difference is small. there is NO difference between AHCI mode and raid mode for a single drive, you'd crazy to think if such a large difference comes from different mode which might be coming from your hardware itself.

 

Read the last part of mine instead instead of entirely ignoring a very good explanation of the difference in AHCI and RST modes.

Also those things have nothing to do with my testing because I was strictly using UEFI/GPT.. why do you keep talking about random things that aren't relevant. It's like you're just saying the most advanced things you can think of then hoping you sound right lol.

 

i already did, i think you're not reading my reply properly though.

since you seem to be convinced that your findings are correct, i'll leave you with this. I am running on windows 8 and I don't use uefi so im on AHCI and mbr. not uefi and gdp, and performance is within 1-2% difference at best.

but i'll leave the rest to your own imagination.

 

Again I implore you to reread the explanation I provided from that Reddit comment I linked. If you don't understand what's being said there I think it's better we drop this discussion till you have a better understanding of the topic at hand because I don't think I'm explaining things for you to understand well enough because of most of the things you've talked about have been incredibly misinformed, different or not related to what was being tested or discussed here.

 

maybe you should read it again cause you're missing out on something, rather than quoting parts of it you don't like.

 

Just for fun I did this same experiment on my Aero 15x. Interestingly I had zero improvement on my SX8200 but better results on the stock Transend 512GB TS512GMTE510T SSD that comes in the laptop. I ran AS SSD and CDM 6.0.2 three times each to get the averages:



A little difficult to read, but you can see some decent read improvements on the TS512.