Question SN850X Write Speeds Faster then Read Speeds?

[GSource]

Just recently got a 4TB SN850X and decided to test it with CrystalDiskMark and found something a little bit peculiar. My sequential write speeds are consistently faster than my sequential read speeds. I've never had a drive of any kind behave like this. Is this normal or a sign of something not being quite right? Do these numbers seem normal?

For context, this drive is in my third M2 slot, so it's chipset driven. The slot is PCIE 4x4.

My latest shows the following:
SEQ1M Q8T1:
- Read = 6228 MB/s
- Write = 6338 MB/s

SEQ1M Q1T1:
- Read = 4599 MB/s
- Write = 5350 MB/s

RND4K Q32T1:
- Read = 1006 MB/s
- Write = 916 MB/s

RND4K Q1T1:
- Read = 80 MB/s
- Write = 239 MB/s

 

[Darkbreeze]

No, it's not normal and that drives specs don't indicate that should be the way it is. I'd suggest making sure you have the latest firmware installed for your drive, and the latest version of CrystalDiskMark. And also maybe, try a couple of other disk benchmark utilities to see if you get the same results.

I'd also make sure you have the MOST recent motherboard chipset drivers installed for your motherboard model AND that you have the latest motherboard BIOS version installed.

 

[GSource]

No, it's not normal and that drives specs don't indicate that should be the way it is. I'd suggest making sure you have the latest firmware installed for your drive, and the latest version of CrystalDiskMark. And also maybe, try a couple of other disk benchmark utilities to see if you get the same results.

I'd also make sure you have the MOST recent motherboard chipset drivers installed for your motherboard model AND that you have the latest motherboard BIOS version installed.

Everything is up to date in my system! If it's helpful, here are my specs:
MB: MSI X670E Tomahawk
CPU: 9800X3D
GPU: 7900 XTX

I checked other tests as well, but similar results.

 

[Darkbreeze]

Might check in the BIOS and make sure that your NVME settings are set to Auto rather than say, Gen 3, Gen 4, etc. Shouldn't really matter if it's set to the correct Gen, but I've seen specific settings cause some weird things that setting to Auto has corrected.

 

[Darkbreeze]

I'd suggest you remove and reinstall the drive to see if that changes anything. One of the contacts may not be making great contact.

Or, try it in a different M.2 slot to see if anything changes. Otherwise, this is not typical, I've never seen it before and I can't find any indication that anybody else has for this drive either.

All else fails, I'd contact WD about a replacement.

 

[evermorex76]

My SN850X shows significantly higher write speeds in the 4K tests in AS-SSD and intermittently higher 4K-64Thrd tests in AS-SSD (which would match the results you get although I can't change the queue depth), but sequential reads are higher. Your sequential rates aren't hugely far apart but still shouldn't look like that. I assume you've run this multiple times and made sure there are no background apps or anything that might be affecting it.

These are my DiskMark numbers with an SN850X as my C drive on an X570 in the primary slot. Some of these tests, the ones with queue depth 1, are I believe inherently less optimal for reads, particularly random, since the controller and OS can't have requests already being assigned and have to wait for the previous one to be processed. That just works out better for writes.

7313/5631
5177/5270
722/534
80/212

Also note that the third M.2 slot is running on the secondary X670 chip (X670E is composed of two linked X670 chips), not the primary. So traffic to that slot requires passing over two PCIe x4 hops (CPU to primary chipset, primary to secondary) which could easily affect performance. It's not clear whether the second M.2 slot is running on the primary chipset or the secondary. They should support 1 and 2 M.2 slots, respectively, but MSI chose to make slot 4 shared with PCIe slot 4, so I suspect both of them are using the secondary chipset, but you could test in the second M.2 slot to see if performance is better.

 

[Darkbreeze]

Also note that the third M.2 slot is running on the secondary X670 chip (X670E is composed of two linked X670 chips), not the primary. So traffic to that slot requires passing over two PCIe x4 hops (CPU to primary chipset, primary to secondary) which could easily affect performance.

This shouldn't matter because both the write and read operations are running on the same bus from the same slot. And besides which, no matter if it was running on SATA, IDE, or any gen of NVME, writes should always be lower than reads because there is always more involved in the process. From old mechanical drives to the latest M.2 drives, reads should always be faster. Any drive that isn't, there is some kind of problem somewhere.

 

[evermorex76]

This shouldn't matter because both the write and read operations are running on the same bus from the same slot. And besides which, no matter if it was running on SATA, IDE, or any gen of NVME, writes should always be lower than reads because there is always more involved in the process. From old mechanical drives to the latest M.2 drives, reads should always be faster. Any drive that isn't, there is some kind of problem somewhere.

Anything that adds an additional hop and additional processing has the potential to affect performance in odd ways. The X670E (and newer E chipsets) are cobbled together. And as I showed, my own SN850X has similar results, although not in sequential performance, AND mine is significantly faster in sequential performance. All drives I've tested myself have shown significantly faster speeds in writes when running random tests and my understanding is that is normal. OP's sequential test is so close it COULD be just a variance in a particular run combined with those two hops limiting performance so it simply can't go over the 6.4GBps mark, even though the drive should be able to match my 7.3GBps in reads. (My drive has been in use for some time of course so performance may not match a brand new one.)

The only way to know for certain would be installing it in the primary M.2 slot (and testing in the second slot would be helpful as well).

 

[Darkbreeze]

OP's sequential test is so close it COULD be just a variance in a particular run

Actually, it can't, because he's done several different runs using multiple different benchmark utilities, and got the same results.

I checked other tests as well, but similar results.

Besides which, the drive is supposed to have a 600 MB/s faster sequential read speed than it does write speed, add to that the 110 MB/s difference in faster write speed that it actually has and that becomes a 710 MB/s discrepancy which I would not call "not hugely far apart" or "just a variance". It seems rather significant to me when you look at it like that. I agree that it should be tested in the other slots to see if the problem follows the drive or if it's perhaps a problem with that M.2 slot, but it seems the OP got the answers he was looking for so perhaps this is a moot discussion?