If you are running a homelab with multiple Proxmox hosts, a NAS, and Docker containers, your 1GbE network is probably the bottleneck. VM migrations crawl, NFS storage feels sluggish, and pulling container images takes longer than it should. A 10GbE upgrade changes that — but only if you pick the right hardware and configure it correctly.

This guide covers the full stack: choosing a MikroTik CRS switch and SFP+ NICs, cabling decisions, VLAN segmentation, Proxmox host integration, Linux performance tuning, and real-world throughput validation.

Why Upgrade to 10GbE in Your Homelab

The jump from 1GbE to 10GbE is a 10x throughput increase. Here is where it matters most:

  • NFS / NAS storage — A single 1GbE link caps at ~110 MB/s, which is slower than most SATA SSDs. 10GbE unlocks sequential reads at 800-1100 MB/s.
  • VM live migrations — Moving a 64 GB VM over 1GbE takes ~9 minutes. Over 10GbE, under a minute.
  • Container registry pulls — Downloading multi-gigabyte images at 110 MB/s is painful. 10GbE drops that to seconds.
  • Cluster storage replication — Ceph or DRBD between Proxmox nodes benefits massively from 10GbE bandwidth.

If your homelab runs a single server with everything local, 1GbE is fine. Once you add a second host, shared storage, or backup infrastructure, 10GbE becomes the unlock.

Choosing Your 10GbE Hardware — Switch, NICs, and Cabling

MikroTik CRS Switches

The MikroTik CRS3xx series is the go-to for cost-effective 10GbE switching. Three models cover most homelab scenarios:

Model SFP+ Ports 1GbE Ports Form Factor Notes
CRS309-1G-8S+ 8 1 Fanless desktop Best for small homelabs, silent
CRS317-1G-16S+ 16 1 1U rackmount 16 ports, more airflow noise
CRS326-24S+2Q+RM 24 SFP+ + 2x40GbE 0 1U rackmount Overkill for most; future-proof

The CRS309 is the sweet spot: fanless, 8 SFP+ ports, and RouterOS 7 with full hardware offloading. At about 8W idle, it runs cool and silent with DAC cables.

Network Interface Cards

Your Proxmox hosts need SFP+ NICs. Reliable options:

  • Intel X710-DA2 — 2-port 10GbE SFP+, Linux-native driver (i40e), low power, well-supported
  • Mellanox ConnectX-3 Pro — Older but cheap on eBay, driver (mlx4_core), solid performance
  • Mellanox ConnectX-4 Lx — 2-port 25GbE (downward compatible to 10GbE), driver (mlx5_core)
  • Integrated on server motherboards — Many Supermicro and Dell boards include x710 or x520 on-board

Check existing NICs on your Proxmox host:

lspci | grep -i ethernet

Verify current link speed:

ethtool eth0 | grep Speed

Cabling: DAC vs Fiber vs Transceivers

Choosing the right cable type saves money and avoids signal issues:

  • Direct Attach Copper (DAC) — Best for runs under 5m (same rack or adjacent racks). Passive, no transceivers needed, lowest power, cheapest. Use SFP+ passive DAC cables.
  • Fiber + SFP+ transceivers — For runs over 5m or between rooms. Multimode fiber (OM3/OM4) with 10GBASE-SR transceivers supports up to 300m.
  • 10GBASE-T SFP+ modules — Only if you need RJ45 copper 10GbE. Higher power draw (~3-4W per port), runs hot, avoid if possible.

For most homelabs, DAC cables between the switch and server NICs are the right choice. Example: a 2m passive SFP+ DAC is about $8-15 on FS.com vs $30+ for a pair of transceivers plus fiber.

MikroTik CRS Switch Configuration for 10GbE

Start by resetting the CRS to a clean switching configuration through RouterOS:

/system reset-configuration run-after-reset="
/interface bridge add name=bridge1 vlan-filtering=yes protocol-mode=rstp
/interface bridge port add bridge=bridge1 interface=sfp-sfpplus1 hw-offload=yes
/interface bridge port add bridge=bridge1 interface=sfp-sfpplus2 hw-offload=yes
/interface bridge port add bridge=bridge1 interface=sfp-sfpplus3 hw-offload=yes
/interface bridge port add bridge=bridge1 interface=sfp-sfpplus4 hw-offload=yes
/interface bridge port add bridge=bridge1 interface=sfp-sfpplus5 hw-offload=yes
/interface bridge port add bridge=bridge1 interface=sfp-sfpplus6 hw-offload=yes
/interface bridge port add bridge=bridge1 interface=sfp-sfpplus7 hw-offload=yes
/interface bridge port add bridge=bridge1 interface=sfp-sfpplus8 hw-offload=yes
/interface bridge port add bridge=bridge1 interface=ether1
/interface vlan add name=vlan10 interface=bridge1 vlan-id=10
/interface vlan add name=vlan20 interface=bridge1 vlan-id=20
/ip address add address=10.0.30.1/24 interface=vlan10
/ip address add address=10.0.20.1/24 interface=vlan20
"

This configures VLAN filtering on the CRS bridge, assigns all SFP+ ports to the bridge with hardware offloading, and creates VLAN interfaces for management access.

For VLAN segmentation, define which ports carry which VLANs:

/interface bridge vlan
add bridge=bridge1 vlan-ids=10 tagged=bridge1,sfp-sfpplus1,sfp-sfpplus2 untagged=
add bridge=bridge1 vlan-ids=20 tagged=bridge1,sfp-sfpplus1,sfp-sfpplus2 untagged=sfp-sfpplus3,sfp-sfpplus4

Verify that all ports are hardware-offloaded:

/interface bridge port print where hw-offload=no

If any port shows hw-offload=no, check that the CPU interface (/interface bridge port without a physical interface) is not accidentally forwarding traffic. On CRS3xx switches, the switch chip handles all forwarding when properly configured.

Proxmox Host 10GbE NIC Setup

After physically installing the SFP+ NIC, ensure the kernel driver is loaded:

lsmod | grep -E "i40e|mlx4_core|mlx5_core"

For Intel X710-based NICs:

modprobe i40e
echo "i40e" >> /etc/modules

Configure a static IP on the 10GbE interface in Proxmox’s /etc/network/interfaces. Example with a bridge for VM traffic:

auto enp4s0f0
iface enp4s0f0 inet manual
    mtu 9000

auto vmbr10
iface vmbr10 inet static
    address 10.0.30.10/24
    mtu 9000
    bridge-ports enp4s0f0
    bridge-stp off
    bridge-fd 0

For dual-NIC bonding with LACP (802.3ad), the MikroTik switch side needs a bond interface:

/interface bonding add name=bond1 mode=802.3ad slaves=sfp-sfpplus1,sfp-sfpplus2
/interface bridge port add bridge=bridge1 interface=bond1 hw-offload=yes

And on the Proxmox host:

auto enp4s0f0
iface enp4s0f0 inet manual
    mtu 9000

auto enp4s0f1
iface enp4s0f1 inet manual
    mtu 9000

auto bond0
iface bond0 inet manual
    mtu 9000
    bond-slaves enp4s0f0 enp4s0f1
    bond-mode 802.3ad
    bond-miimon 100
    bond-lacp-rate 1

auto vmbr10
iface vmbr10 inet static
    address 10.0.30.10/24
    mtu 9000
    bridge-ports bond0
    bridge-stp off
    bridge-fd 0

Linux 10GbE Performance Tuning for Maximum Throughput

Getting full line-rate 10GbE requires tuning beyond the defaults. The Linux network stack defaults to conservative values optimized for 1GbE.

Ring Buffer Size

Default ring buffers are often 256 or 512 descriptors, which can starve 10GbE throughput under load.

ethtool -g eth0
ethtool -G eth0 rx 4096 tx 4096

Make this persistent via a systemd service or by adding the command to /etc/rc.local.

IRQ Affinity and RSS Queues

10GbE NICs expose multiple MSI-X interrupt vectors, one per RSS queue. Distribute them across CPU cores:

cat /proc/interrupts | grep eth0

List the MSI-X IRQs:

ls /sys/class/net/eth0/device/msi_irqs/

On systems with 4+ cores, irqbalance handles this automatically. Verify:

systemctl status irqbalance

For manual pinning, set each IRQ’s smp_affinity to a different core mask:

echo 1 > /proc/irq/125/smp_affinity   # core 0
echo 2 > /proc/irq/126/smp_affinity   # core 1
echo 4 > /proc/irq/127/smp_affinity   # core 2
echo 8 > /proc/irq/128/smp_affinity   # core 3

TCP Stack Tuning

Append these to /etc/sysctl.d/90-network-tune.conf:

net.core.rmem_max = 134217728
net.core.wmem_max = 134217728
net.ipv4.tcp_rmem = 4096 87380 33554432
net.ipv4.tcp_wmem = 4096 65536 33554432
net.core.default_qdisc = fq
net.ipv4.tcp_congestion_control = bbr
net.core.netdev_budget = 600
net.core.netdev_budget_usecs = 4000

Apply immediately:

sysctl -p /etc/sysctl.d/90-network-tune.conf

BBR (Bottleneck Bandwidth and Round-trip) congestion control is critical for long flows. It handles bufferbloat better than CUBIC on high-speed links.

Jumbo Frames

Enable 9000-byte MTU on every hop: switch ports, host interfaces, and storage targets.

On the switch, set MTU on each SFP+ port:

/interface ethernet set sfp-sfpplus1 mtu=9000
/interface ethernet set sfp-sfpplus2 mtu=9000

On the Proxmox host:

ip link set dev enp4s0f0 mtu 9000

Make it permanent in /etc/network/interfaces by adding mtu 9000 to each interface stanza.

Testing Throughput with iperf3

Validate the setup end-to-end between two 10GbE-connected hosts:

iperf3 -c 10.0.30.20 -P 4 -t 30

Expected output on a properly tuned link:

[ ID] Interval           Transfer     Bitrate
[  5]   0.00-30.00  sec  34.0 GBytes  9.77 Gbits/sec                  receiver

If you see under 9 Gbps, check CRC errors and IRQ distribution.

Storage Performance with 10GbE

Mounting NFS shares over the 10GbE VLAN makes a dramatic difference. Optimize mount options:

10.0.30.5:/volume1/nfs /mnt/nfs nfs4 rw,hard,intr,rsize=1048576,wsize=1048576,noatime,vers=4.2 0 0

For Proxmox, add a dedicated NFS storage pointing to your NAS over the 10GbE VLAN:

pvesm add nfs fast-storage --server 10.0.30.5 --export /volume1/nfs --path /mnt/pve/fast-storage --options vers=4.2,rsize=1048576,wsize=1048576

Backup operations that used to take 45 minutes over 1GbE finish in 5-7 minutes over 10GbE. VM storage migrations drop from 9 minutes to under a minute.

Power and Thermal Considerations

10GbE can add measurable heat to your rack. Some practical notes:

  • DAC cables are passive and generate negligible heat. Plug-and-play.
  • SFP+ optical transceivers consume 0.5-1W each and produce noticeable heat. A switch with 8 populated SFP+ transceivers can run 10-15°C hotter than with DACs.
  • 10GBASE-T SFP+ modules consume 2.5-4W each and need active cooling. Avoid if possible.
  • CRS309 with DACs runs fanless at ~45-50°C case temperature in ambient 25°C. With fiber transceivers, expect 55-60°C.
  • NIC power draw — A dual-port Intel X710 draws about 8-10W idle. Mellanox ConnectX-3 draws about 9-12W.

Troubleshooting Common 10GbE Issues

If the SFP+ link does not establish, check transceiver compatibility. MikroTik devices work with generic SFP+ modules when set sfp-sfpplus1 advertise=10G-full is set, but DAC cables (MikroTik or FS.com compatible) are more reliable.

CRC and Error Counters

Run on the Proxmox host:

ethtool -S eth0 | grep -E "crc|error|miss"

Any non-zero CRC counter indicates signal integrity issues. Try reseating the cable, cleaning fiber end-faces with a cleaning pen, or swapping the DAC.

Port Flapping

A link that drops and reconnects repeatedly is often a power-saving setting on the NIC:

ethtool --set-priv-flags eth0 disable-fw-lro on

Or disable energy-efficient Ethernet:

ethtool --set-eee eth0 eee off

This usually means a cable or transceiver mismatch. Manually advertise 10GbE on the MikroTik port:

/interface ethernet set sfp-sfpplus1 advertise=10G-full

Then check:

/interface ethernet monitor sfp-sfpplus1 once

Conclusion

Deploying 10GbE in a homelab is straightforward when you follow the right sequence: pick a MikroTik CRS switch and DAC cables, configure hardware-offloaded bridging with VLANs, install and tune the NIC on your Proxmox host, and validate with iperf3.

Quick Checklist

  • Choose switch form factor: CRS309 (desktop, silent) or CRS317 (1U, more ports)
  • Buy passive DAC cables for each server-to-switch link
  • Reset CRS switch, configure bridge with vlan-filtering and hw-offload
  • Install SFP+ NIC in Proxmox host, load driver
  • Configure /etc/network/interfaces with bridge and optional bond
  • Tune ring buffers, IRQ affinity, TCP sysctls, and jumbo frames
  • Validate with iperf3 — expect 9.4+ Gbps
  • Mount NFS shares with optimal rsize/wsize over the 10GbE VLAN

The performance difference is transformative. Once you have done a VM migration in 30 seconds or a backup in 5 minutes, you will not want to go back to 1GbE.