Baseline with one large file
Small files measure the protocol overhead, not the link. Unbuffered I/O keeps the file cache out of the result so the number reflects disk and network together.
robocopy \\NAS\share C:\Temp\Disk testfile.iso /J /NP
Measure real throughput, read the negotiated dialect, and check the four settings that throttle a gigabit network.
Copy with **robocopy /MT** while watching Task Manager > Ethernet, then read the dialect with `Get-SmbConnection`. A gigabit link should carry about 110 MB/s; 40 MB/s points at a setting, not the cable.
Small files measure the protocol overhead, not the link. Unbuffered I/O keeps the file cache out of the result so the number reflects disk and network together.
robocopy \\NAS\share C:\Temp\Disk testfile.iso /J /NP
Many small files need multiple streams to fill the link. One large file cannot use them, which is why the two tests disagree.
robocopy \\NAS\share C:\Temp\Disk *.* /S /MT:16 /NP
SMB 3.1.1 encrypts and signs efficiently. A 2.x dialect means old firmware or a router inspecting SMB, and both cost throughput.
Get-SmbConnection | Select-Object ServerName,Dialect,NumOpens
Encryption costs processor time on both ends and on a weak CPU it halves the rate. Signing is cheaper and usually enough on a trusted LAN.
Get-SmbClientConfiguration | Select-Object RequireSecuritySignature,EnableSecuritySignature
Half duplex, a marginal cable or a 100 Mbit port produce exactly the 10 to 12 MB/s plateau people blame on SMB.
Get-NetAdapter | Select-Object Name,LinkSpeed,FullDuplex
An antivirus scanner reads every byte written and every byte read. On a file copy that is often the entire difference between 60 and 110 MB/s.
Add-MpPreference -ExclusionPath C:\Temp\Disk
The write cache filled up. The steady rate after a minute is the real one.
80 MHz on 5 GHz gives 40 to 60 MB/s in practice. Cable it before changing any setting.
The receiving disk or its write caching policy is the limit, not the network.
SMB1 is enabled on the NAS. Move the device to SMB2 or SMB3; SMB1 should not run at all.
Only when every device on the path supports them. One switch at 1500 MTU makes throughput worse, not better.
No. Receive side scaling on a single card also spreads the load across queues.
Remove the antivirus exclusion and restore signing or encryption in the same dialogs if you changed them for the test.
Free · no account needed · tracks your progress