The M.2 NVMe SSD may be using an incorrect transfer mode.
The M.2 NVMe SSD may be using an incorrect transfer mode.
So curious, as I may be missing something, how is it that with his 16X GPU, and 3 4X drives that it could use all the lanes it wants. The OP did say " Samsung 970 ... (PCIe 3.0 x4) - Data: Sabrent ... (PCIe 3.0 x4) - Data: ADATA ... (PCIe 3.0 x4)" I could totally be missing something, but 16 + 4 + 4 + 4 is greater then the total PCIe lanes the CPU has, are you saying that some lanes are basically being used by more then one device concurrently? If so that is something new I learned!
It isn't 16+4+4+4, it is 16+4+4 with the last 4 going through the chipset's 4 lanes. 16 for gpu 4 for first m.2 4 for chipset (this 4 includes secondary m.2,ancillary pcie slots, usb, networking, audio, everythingelse) - These all share the bandwidth of the 4 lanes. And that could be the reason for the slow speeds. If you have lots of traffic on the NIC or USB or other slots, the m.2 will need to slow down and share bandwidth.
. It initially reached around 400MB/s and peaked at about 700MB/s before decreasing to 600MB/s. After renaming, I moved it back to the slower drive. This scenario could relate to caching during testing—starting near ~2GB/s but dropping sharply to roughly 800MB/s (suggesting a cache presence). That’s unusual since sequential reads should be limited to about 2GB/s regardless of caching. However, this particular drive typically handles sequential operations better than it did previously. I repeated the test twice. The files remain identical due to repeated conditions.
I performed the operation. I generated a 21GB file on the slower storage and transferred it to the faster one (D: to C. It initially reached around 400MB/s and peaked at about 700MB/s before decreasing to 600MB/s. After renaming, I moved it back to the slower drive. This scenario could relate to caching during testing—starting near ~2GB/s but dropping sharply to roughly 800MB/s (suggesting a cache presence). That’s unusual since sequential reads should be limited to about 2GB/s regardless of caching. However, this particular drive typically handles sequential operations better than it did previously. I repeated the test twice. The files remain identical due to repeated conditions.
Xxunidemonxx
11-16-2020, 08:47 AM #16Some specifications suggest a 10-lane connection to the chipset. The remaining PCIe ports are unused; USB only handles mouse and keyboard, the audio interface is external via USB with limited speed, there are no SATA drives, and networking appears to be stable. I plan to perform tests tomorrow by disconnecting everything, though I don’t expect much success.
farmerboy18
11-16-2020, 05:04 PM #17It's a four-lane connection to the chipset. The chipset supports ten lanes of devices, requiring four lanes for communication between the CPU and chipset.
Right, so I ran each drive individually in the M.2 or in the PCIe adapter and they run at ~850MB/s each. I also ran both at the same time, and then we get 900MB/s shared between the two. So basically, the motherboards come with 2x M2 slots, but the second one can never run at its intended speed as soon as you use the network or something? What the hell? Misleading advertisement?