MAG Tomahawk X870 M2_1 and M2_2 not achieving advertised speed

leipp124c02a1

New member
Joined
May 11, 2025
Messages
22
tl;dr;
Devices in both M2_1 and M2_2 ports show connecting at 5.0x4.
Each device only reaches 4.7GB/s which is not even gen 4.0x4 max speed of 7GB/s.
I'd expect closer to 14GB/s.

details:
BIOS: latest A44
linux kernel: tried various 6.12.n and also the latest 6.14.6
2x Crucial T705 gen5 NVMe in M2_1 and M2_2 ports.
lspci -vv shows both connected at gen 5.0x4, 32GT/s link speeds.
Testing with dd if=/dev/nvme0n1 of=/dev/null bs=4096 count=10M shows a mere 4.7GB/s on either device.
Testing with hdparm -tT [--direct] /dev/nvme0n1 shows similar slow speeds.
I checked the temps before, during, and after each test run and they never exceeded the 40C range (I think 44C was the highest temp I saw).
They should see close to 14GB/s.
Meanwhile on my older board my older gen 4.0x4 NVMe sees 7GB/s as expected, connected at 4.0x4 16GT/s link speed.
I've tried disabling M2_2, it had no effect on M2_1.
Setting either or both to gen4 speeds they show correctly as 16GT/s link speed and still show only 4.7GB/s performance.
I tried disabling the PCI ASPM, no change.
Not sure what else to try.
 
I installed a spare Samsung 980 Pro in the M2_4 port.
It connects correctly at gen 4 x4 16GT/s.
However, it only sees 577 MB/s (after several runs of even slower results).
It's not thermally throttled--temp never got over 60C. Meanwhile in another system, the same device at 65C+ sees 7GB/s consistently.
This system is entirely idle--there is no configured network, no other I/O occurring.

What is wrong with the PCIe/NVMe I/O performance on this motherboard???
Is there some setting somewhere that will let it hit max potential?

I bought this board specifically since it advertises the ability to run both M2_1 and M2_2 at full gen 5.0 x4--yet it can't even manage to see gen 4.0 x4 max numbers. Very disappointing so far. What other knobs are there to turn?
 
Gets more strange.
After sitting idle for over an hour I sat down and just ran the test again on the Samsung 980 pro in M2_4 port and got 1.9GB/s several times in row. Max temp was 53C. Better than 577MB/s, but still a very far cry from the 7GB/s it should see.
The T705s in M2_1 and M2_2 still saw the same 4.7GB/s. Temps in the 40C range.
 
So just a random thought as I ran into a similar issue after doing a drive clone to my Samsung 9100 Pro. I was getting 4x4 speeds after the clone instead of 5x4 like it should have been. In my case after doing some thinking and research, I performed a 4k Alignment on the drive and it fixed the unusual speed issues. Might be worth a try.
 
Both drives are set to native 4k-mode. I also benchmarked them before I switched them, and while still in 512 block mode saw the same speeds.
There is no file system on either device--I'm just testing using direct block access with dd.
 
Okay, so just to be thorough, I reran several tests, each 3 times.
Crucial T705 in M2_1 connected at 5.0x4 32GT/s, temps peaked at 35C, native 4k block mode:
dd if=/dev/nvme0n1 of=/dev/null bs=512 count=80M = 2.9GB/s
dd if=/dev/nvme0n1 of=/dev/null bs=4096 count=10M = 4.7GB/s
dd if=/dev/nvme0n1 of=/dev/null bs=1M count=40K = 4.5GB/s

This should see close to 14GB/s

Samsung 980 Pro in M2_4 at 4.0x4 16GT/s, temps peaked at 60C, 512 byte block mode:
dd if=/dev/nvme0n1 of=/dev/null bs=512 count=80M = 2.4GB/s
dd if=/dev/nvme0n1 of=/dev/null bs=4096 count=10M = 1.9GB/s
dd if=/dev/nvme0n1 of=/dev/null bs=1M count=40K = 1.5-1.9GB/s

This should see close to 7GB/s

For control, the same Samsung 980 Pro in an older ASRock machine at 4.0x4 16GT/s, 512 byte block mode, on an active system, temps peaked at 70C:
dd if=/dev/nvme0n1 of=/dev/null bs=512 count=80M = 1.1GB/s <== might have been in a power save mode still
dd if=/dev/nvme0n1 of=/dev/null bs=4096 count=10M = 6.3GB/s <== much closer to 7GB/s, as expected

This gets close to 7GB/s, as expected.

The MAG Tomahawk X870 isn't even delivering close to gen 4x4 speeds, let alone anything approaching gen 5x4 speeds...
 
Last edited:
Well, mine was too. After rebuilding it though it worked fine. Just saying it might be worth a try. 👍
There's nothing to rebuild.
These are brand new, empty drives, no filesystem, the write count on them is still 0.
Still running integrated GPU for video, no video card installed yet--so PCIE_1 is empty--minimized variables.
It's booted to a linux install media off a USB thumbdrive.
This is the same way I've sanity-tested every new build for the past decades using spinning hard disks (IDE/SCSI/SATA), SATA ssds, and now NVME ssds. I've always been able to confirm peak device sequential read speeds this way before beginning the rest of any install or migration adding in more variables.

Something is definitely off on this motherboard.
 
Clear CMOS Button and tried again or reinstall Bios again see if that works, if I'm correct the specs are advertise

  • 1 x M.2, 22-60/80, PCIe Gen4x2
  • 1 x M.2, 22-60/80, PCIe Gen4x4
  • 1 x M.2, 22-60/80, PCIe Gen5x4
  • 1 x M.2, 22-80/110, PCIe Gen5x4
  • 1 x PCIe 3.0 x16 (x1 Bandwidth)
  • 1 x PCIe 4.0 x16
  • 1 x PCIe 4.0 x16 (x4 Bandwidth)
  • 1 x PCIe 5.0 x16
  • 4 x SATA 6GB/s
But does not state weather this broad shares lanes and that's why your not getting the full speeds? possible defected broad?
 
I have tried clearing the CMOS and resetting all BIOS to factory. Did not change anything.
It's not an issue with lane sharing. There's a block diagram on pg 73 in the manual that shows all the lane sharing--as well as numerous images online of the X870 PCIe lane allocations between the CPU and the X870 chipset.
Both Gen5x4 ports have 4 dedicated lanes and should each see gen 5.0x4 speeds (14MB/s).
The M2_4 port should be able to max out the 4 lanes to the chipset and see gen 4.0x4 speeds (7MB/s) on an otherwise idle system.
It's really simple. With every other motherboard I've tested in the past I could just plug in new devices, run a baseline test and max out the device to confirm specs.
With this MSI board this has not been the case--it simply will not achieve acceptable performance--not even gen 4.0x speeds.

MSI needs to fix this.
 
Checked the ASROCK X870E Taichi (BIOS 3.25. CPU: R5-9600X) and ASUS ProArt X870E-Creator WiFi (BIOS 1401), the M.2 on M2_1 slot runs with 4.7 to 5.1 GB/s in linux with command "dd if=/dev/nvme0n1 of=/dev/null bs=4096 count=10M" which is close to the speed on MAG X870 TOMAHAWK WIFI.
 
Checked the ASROCK X870E Taichi (BIOS 3.25. CPU: R5-9600X) and ASUS ProArt X870E-Creator WiFi (BIOS 1401), the M.2 on M2_1 slot runs with 4.7 to 5.1 GB/s in linux with command "dd if=/dev/nvme0n1 of=/dev/null bs=4096 count=10M" which is close to the speed on MAG X870 TOMAHAWK WIFI.
so... way too slow on those boards also.
Maybe this is somehow a bug in Agesa?
My AM4 X570 boards have had no issue maxing out gen4x4 nvme devices. Looks like AM5 X870(e) boards are struggling, not even achieving gen4x4 speeds let alone gen5x4.
Currently, this X870 board would represent a downgrade, not an upgrade.
 
Last edited:
Tech support said: "based on both internal findings and confirmed cases, we have identified a hardware-level inconsistency affecting the performance of certain M.2 slots on a small batch of boards."
They did not elaborate on what build date or serial numbers were affected, nor which M.2 slots were affected. In my case at least 3 of the 4 M.2 slots are affected--all of the x4 slots.
 
Checked the ASROCK X870E Taichi (BIOS 3.25. CPU: R5-9600X) and ASUS ProArt X870E-Creator WiFi (BIOS 1401), the M.2 on M2_1 slot runs with 4.7 to 5.1 GB/s in linux with command "dd if=/dev/nvme0n1 of=/dev/null bs=4096 count=10M" which is close to the speed on MAG X870 TOMAHAWK WIFI.
Even with lower speed in Linux, the speed in CrystalDiskMark on Windows 11, Crucial T705 in M2_1 reaches 11000 MB/s read speed.
 
Back
Top