Discussion Reduced SATA performance on X570?

[tamz_msc]

https://www.reddit.com/r/Amd/comments/fwh7q0

This might be worthy of further investigation.

 

[SCdF]

So I just ran into this issue when I expanded past my 1TB nvme drive into a 4TB SSD, and was finding it was a runninng significant slower than it should for sequential reads, according to various benchmarks / marketing.

I'm not really sure what is going on, but this maybe be a cautionary tale to not believe block diagrams! I have a MSI x570 Gaming Edge Wifi (the mobo reviewed in the link at the start of this thread), and its block diagram (page 21) says that all 6 SATA ports go through the PCH.

However, in my testing ports 1 & 2 perform significantly worse than ports 3-6.

Reference CrystalDiskMark pulled from a benchmark in a review for a similar drive:


My Result from Port 1 (Port 2 looks similar):


This looks bad! Here is that same SSD though, in Port 5:



CDM seems to vary a lot, but ports 3-6 are all consistently 400-550 for Seq Q32T1 Read, whereas ports 1-2 always sit somewhere inside 3xx.
https://www.thessdreview.com/featur...ra-3d-ssd-review-1tb-twins-at-their-finest/3/
I'm not knowledgeable enough to know if this could be because Ports 1&2 go through a different flow than 3-6, but if you're seeing bad performance it's definitely worth ignoring what's written and just trying every port.

p.s., this was the last thing I tried, after reformatting multiple times with different block sizes, trimming, resetting and updating my BIOS to latest and trying other benchmarks and different versions of CDM.

p.p.s. I know the CDM version is old (and that CDM isn't great), I just wanted to match the benchmark site as quickly and easily as possible so I used the same major

 

[SamMaster]

So I just ran into this issue when I expanded past my 1TB nvme drive into a 4TB SSD, and was finding it was a runninng significant slower than it should for sequential reads, according to various benchmarks / marketing.

I'm not really sure what is going on, but this maybe be a cautionary tale to not believe block diagrams! I have a MSI x570 Gaming Edge Wifi (the mobo reviewed in the link at the start of this thread), and its block diagram (page 21) says that all 6 SATA ports go through the PCH.

However, in my testing ports 1 & 2 perform significantly worse than ports 3-6.

Reference CrystalDiskMark pulled from a benchmark in a review for a similar drive:
View attachment 44443

My Result from Port 1 (Port 2 looks similar):
View attachment 44444

This looks bad! Here is that same SSD though, in Port 5:

View attachment 44445

CDM seems to vary a lot, but ports 3-6 are all consistently 400-550 for Seq Q32T1 Read, whereas ports 1-2 always sit somewhere inside 3xx.
https://www.thessdreview.com/featur...ra-3d-ssd-review-1tb-twins-at-their-finest/3/
I'm not knowledgeable enough to know if this could be because Ports 1&2 go through a different flow than 3-6, but if you're seeing bad performance it's definitely worth ignoring what's written and just trying every port.

p.s., this was the last thing I tried, after reformatting multiple times with different block sizes, trimming, resetting and updating my BIOS to latest and trying other benchmarks and different versions of CDM.

p.p.s. I know the CDM version is old (and that CDM isn't great), I just wanted to match the benchmark site as quickly and easily as possible so I used the same major

Looking at the manual for your motherboard, SATA ports 1 and 2 use the ASMedia chip while the other four use the X570 chipset. That would explain the speed difference.

 

[SCdF]

Looking at the manual for your motherboard, SATA ports 1 and 2 use the ASMedia chip while the other four use the X570 chipset. That would explain the speed difference.

You're 100% correct, I completely missed that! I guess the real lesson here is a cautionary tale about reading comprehension...

 

[DrMrLordX]

ASMedia has one S too few in the name.

 

[00Logic]

66% of Windows I/O is RANDOM 4K
Less than 1% is the large sequential #s you guys are winging about here..!

Understanding SSD Advertised Performance and Its Purchase Implications – An SSD Primer - The SSD Review

Understanding SSDs Performance and Implications Today's SSD close up is going to teach us the most v

www.thessdreview.com

Games are trying to do bigger RANDOM reads(16-32K) but still all the random writes.

How Many Files Do Games Load? Does 4K Performance Matter to SSDs? | GamersNexus

stub SSD benchmarks generally include two fundamental file I/O tests: Sequential and 4K random R/W. At a very top-level, sequential tests consist of large, individual files transfers (think: media files), which is more indicative of media consumption and large file rendering / compilation. 4K...

www.gamersnexus.net

Further:
Around 80% of Windows I/O is at Que Depth 1.
~10%: QD2. (IIRC)
~1%: QD4.

Windows also does a ~75/25% mix of SIMULTANIOUS read/write I/O.
That drops the #s down to around 30% of the advertised #s..!

ie: You want a fast, responsive sys:
You should NOT give-AF what the large, sequential, high QD, 100% read numbers are!!!
Look to R4K, QD1, 75/25 Mix numbers..!

Also NB that AMD's drivers have better #s for said R4K etc I/0 than the MS drivers do..!
But they stopped bothering to make drivers because everyone keeps banging on #s they should be giving less than 1% of AF about!

Re evaluate your 'horrible #s' while looking at R4K, QD1, MIX #s. Preferably with AMD drivers! Do they still look so bad..?

Here are my #s for a 2.5" 500GB WD HDD with the most important rows and columns highlighted:

The R4K QD1#s are thx to 2x 32GB Readyboost caches on 1 and then 2 SATA SSDs.
(That's 16 000 000-32 000 000 R4K files. More 'all of them' than a cache...)
The high sequential #s are due to a small (fake #s) DRAM cache (read/write, coalescing?), running on the drive, necessary to properly unleash RB's potential... (legit R4K #s)​