Dual GPU AI Workstation Motherboard

Dual GPU AI Workstation Motherboard | Generated by AI

Home 2026.07

Question: You have an i9-13900KF + RTX 4070 + DDR5 64GB build and want to add a second RTX 3060. Which motherboard should you choose?

Answer:

Your main requirement changes the motherboard choice:

You are not just building a gaming PC anymore. You are building a small AI workstation.

The important thing is PCIe slot layout, not CPU compatibility.

Your i9-13900KF has:

16 PCIe 5.0 lanes from CPU
- usually used by first GPU: PCIe x16
Z790 chipset provides extra PCIe lanes
- second GPU usually comes from chipset: PCIe x4

For AI workloads (PyTorch, CUDA), a second RTX 3060 running at PCIe 4.0 x4 is usually OK. The GPU is mostly limited by VRAM and compute, not PCIe bandwidth.

A typical layout:

CPU
 |
 +-- RTX 4070
 |      PCIe 5.0 x16
 |
Z790 chipset
 |
 +-- RTX 3060
        PCIe 4.0 x4

The second card is not for SLI/NVLink. It is for:

extra VRAM
running another model
separate CUDA process
distributed training experiments

My recommendation: ATX Z790 board

Do NOT buy mATX.

Your case is “Mini” but if you want dual GPU, you need physical space and airflow.

Good choices:

1. MSI MAG Z790 Tomahawk WiFi DDR5 (my first choice)

MSI MAG Z790 TOMAHAWK WIFI DDR5 ATX Motherboard

$209.99

Why:

i9-13900KF power delivery is handled easily
4 DIMM DDR5 slots
two usable PCIe GPU slots
good VRM cooling
many M.2 slots
good BIOS

Expansion:

PCIe 5.0 x16  -> RTX 4070

PCIe 4.0 x4   -> RTX 3060

PCIe 4.0 x4 M.2
PCIe 4.0 x4 M.2
...

MSI lists the board as Z790/LGA1700 supporting 12th/13th/14th gen Intel CPUs with strong VRM design. (MSI-US Official Store)

2. ASUS TUF Gaming Z790-PLUS WIFI

ASUS TUF GAMING Z790-PLUS WIFI

$268.40

Also very good.

Expansion:

PCIe 5.0 x16

PCIe 4.0 x16 slot
   (electrically x4)

PCIe 4.0 x4

ASUS specifies:

1× PCIe 5.0 x16
1× PCIe 4.0 x16 (max x4)
1× PCIe 4.0 x4

which is exactly the configuration needed for two GPUs. (ASUS Global)

Your PSU problem

Your current:

Corsair CX650F RGB 750W

is borderline.

Power:

i9-13900KF
~250W peak

RTX 4070
~200W

RTX 3060
~170W

Motherboard + disks + fans
~80W

Peak:

250 + 200 + 170 + 80
≈700W

750W works, but not much headroom.

For AI workloads where GPU runs 100% for hours:

I would upgrade to:

850W Gold

Examples:

Corsair RM850x
Seasonic Focus GX-850
MSI MPG A850G

Your case is also a concern

Your RTX 4070 Gaming X Trio is huge.

Then adding RTX 3060:

slot 1:
RTX 4070
████████

slot 2:
RTX 3060
████

Problems:

second GPU may block airflow
temperatures increase
motherboard spacing matters

A better case:

ATX mid tower
7 PCIe slots
front mesh airflow

Examples:

Lian Li Lancool series
Fractal Design Meshify
Corsair 4000D Airflow

One more AI-specific point

For your use case:

RTX 4070 + RTX 3060:

RTX 4070
12GB VRAM

RTX 3060
12GB VRAM

Total:
24GB VRAM

But:

you cannot combine them like one 24GB GPU automatically.

Example:

Qwen model:

12GB model

works.

But:

20GB model

does not simply use:

4070 12GB + 3060 12GB

unless using model parallelism.

Tools:

PyTorch Distributed
DeepSpeed ZeRO
Accelerate

can split models, but bandwidth between GPUs is slow compared with NVLink.

My final build recommendation:

CPU:
i9-13900KF

Motherboard:
MSI MAG Z790 Tomahawk WiFi DDR5

GPU:
RTX 4070 12GB
+
RTX 3060 12GB

RAM:
64GB DDR5 6000

PSU:
850W Gold

Case:
ATX airflow case

This is actually a very reasonable “poor man’s AI workstation”. It gives you:

20 CPU cores / 32 threads
64GB RAM
24GB total NVIDIA VRAM
CUDA compatibility
ability to experiment with multi-GPU training

References:

ASUS TUF GAMING Z790-PLUS WIFI specifications (ASUS Global)
MSI MAG Z790 TOMAHAWK WIFI specifications (MSI-US Official Store)

Back Donate