I’m experiencing a weird performance degradation of about 50% on slot 1 of my motherboard (connected to the CPU) when comparing with the secondary M.2 slot (connected to the B550 chipset.)
My SSD is a 1TB KingSpec (please don’t judge) M.2 NVMe PCI-E 3.0 X4 and my motherboard is the Asus TUF GAMING B550-PLUS (WI-FI).
When the SSD is connected in slot 1, I get about 1800MB/s Q8T1 transfer rate while on slot 2 I get around 3200MB/s transfer rate (CrystalDiskMark testes bellow).
My CPU is a Ryzen 5 3500X. As it supports PCI-E 4.0, and to exclude some type of speed detection issue, I’ve changed the M.2 slot 1 PCI-E setting from “auto” to “PCI-E 3.0” and the results are the same.
As the M.2 slot 2 shares bandwidth with SATA ports 5 and 6, and when this slot is populated those SATA ports are disabled, I would like to use the SSD on M.2 slot 1.
Could this be some kind of a motherboard / CPU / SSD / Software or a combo issue?
Hard to know for sure but my guess is that the CPU has some bent or missing pins, or there is debris in the socket preventing some pins from making full contact. Or you have uneven mounting pressure from the cooler. (Or you have a bad motherboard trace or a short, but those are less likely.) Can you share a screenshot of CrystalDiskInfo with the drive in the top slot?
The CPU was bought second hand some time ago and I believe all pins were OK.
Since the CPU is not LGA, there is little to no issues related to mounting pressure.
The Motherboard was bought new, and this is it first M.2 drive that I’ve installed on it. Both CrystalDiskInfo screenshot are in the first post, the image 1 relates to M.2 slot 1 and the second image is when the SSD is connected on slot 2.
I’ve found an excellent tool to check the NVMe status and the problem is that the SSD is currently running in PCI-E 3.0 x2 instead of x4:
It’s weird that it only effects sequential operations, and not random. Since the random writes are the same speed, I’d very confidently say you wouldn’t feel the difference doing anything that isn’t large file copies.
The 3500X only has 20 PCIe lanes of bandwidth, so onboard devices on the motherboard, and your GPU using them up, could be limiting the M.2_1 slot to x2. That’s the only reason I can think of.
Fair point, but in a truly extreme case, if the cooler is clamped down asymmetrically or if the backplate is slightly askew it could lead to some pins not being fully inserted.
I see CrystalDiskMark screenshots, but no CrystalDiskInfo screenshots…
…but that’s fine, this is what I wanted to see.
The issue is that some of the fingers on the M.2 connector aren’t making a complete electrical connection to the pins of the CPU. Since we know the SSD is good—since it runs at Gen3 x4 in the bottom slot—that leaves the following potential culprits:
Bad M.2 slot (or debris in the slot)
Bad trace(s) between M.2 slot and CPU socket (or a short)
Bad CPU socket (or debris in the socket)
Bad CPU (or some pins bent or missing or otherwise electrically non-conductive)
I think remounting the CPU is the next logical thing to try, and assuming that doesn’t solve the issue, ideally you will try a different CPU to determine if it’s the board or the CPU. Then you can RMA or otherwise replace whichever one is broken.
It makes perfect sense because the PCIe link only limits maximum throughput. A Gen3 x2 link has the same latency and other characteristics as a Gen3 x4 link. Given M peak IOPS of N block size, if M*N is less than the theoretical maximum throughput of the PCIe link then by definition it can’t be constrained by it. To give a counter example, with a 980 Pro or SK hynix P41 capable of 1M+ 4KiB IOPS, that performance metric will absolutely be constrained by a Gen3 x2 link.
I would agree, but based on what I see in the CrystalDiskMark screenshots, I don’t think OP was testing with the NVMe profile or the Peak Performance configuration, so we aren’t seeing the actual peak IOPS. It still might not be a problem per se in real-word usage, but we don’t currently have enough information to say either way.
All AM4 processors are limited to a maximum of 20 PCIe lanes plus four for the chipset, so this isn’t the issue. There are some AM4 processors with fewer than 20+4 lanes but the 3500X isn’t one of them.
It’s a Gen3 NVMe SSD, so whether the slot is Gen3 or Gen4 is irrelevant. The bugbear is that it’s only using two lanes instead of four when installed in the top slot.
Yes, I also believe that it could be a motherboard socket issue. Since this was my first experience with M.2, I thought that I could be missing something else.
In this weekend, I will try to remove the motherboard and inspect the socket to consider an RMA, although I’m not too fond of being some weeks with no PC…
Will post if I found something out. Thanks everyone for the input!
i have the exact same issue on a Gigabyte B550I AORUS PRO AX (rev. 1.1) with KC3000 2TB in PCIe 3.0 x4 (Chipset) slot and a Fury Renegade 2TB in the PCIe 4.0 x4 slot (CPU)
the chipset connected slot runs at x2 speed even when its reported as 4x in hwinfo
The manual suggests that the M.2 slot shares bandwidth with SATA 5 & 6, but it says the SATA ports will be disabled to provide 4x to the M.2. Maybe that was changed in a bios update, though. Is anything plugged into those SATA ports?
B550 only has 4 sata afaik, which is why the M.2 on the Asus board has a conflict with those SATA ports and the M.2 drive; there’s a secondary SATA controller for the last two ports.
2 PCIE lanes is enough bandwidth for 2 ports and a bit, so if ASUS updated the bios to use a 2x2 configuration rather than disabling the ports, that would be nice.
Then again, maybe it’s a limitation with the B550 chipset operation, and Asus and Gigabyte are just lying about the bandwidth to the slot. Wouldn’t be the first time a manual didn’t have the correct information about a product.
I can’t find any reason the Gigabyte board would operate at 2x on the second slot, aside from maybe sending PCIE over some of the USB ports or some such. That realtek 2.5gbe nic sure sets off some alarm bells though.
perhaps this is the issue, some patches might share bugfixes between windows 10 and 11
“The big picture: Windows 11 comes with plenty of improvements over Windows 10, but it also brings many new bugs that users will find inordinately frustrating. While Microsoft keeps pushing out updates to fix these, time and time again the updates also give rise to new issues. One such bug was seemingly introduced with update KB5023778 in March, whereby the SSD speeds in some machines are slowing down for no apparent reason.”
