Skip to content

Qwen2.5 72B Instruct on 2× RTX 5090: VRAM, speed and fit

Qwen2.5 72B Instruct at 5-bit runs on the 2× RTX 5090 with 32K of context at about 23 tokens a second.

Text generation · Chat · Ollama / LM Studio

Runs comfortably · on 2× RTX 5090 · 5-bit (Q5_K_M) · Ollama / LM Studio · Chat

GPU memory
59.9 GB
of 64.0 GB
Context
32K
auto
Speed
23 tok/s
19 when full
First reply
0.8 s
2K prompt

Where the memory goes

Model
48.2 GB
Context
10.0 GB
Overhead
1.7 GB
OS
0.0 GB
Free
4.1 GB
Memory
59.9 GB of 64.0 GB
Context
32K tokens · auto
Speed
23 tok/s · 19 when full
First reply
0.8 s for a 2K prompt
Running cost
$17/mo at home · $117/mo on a cloud 4× RTX 2000 Ada

Validation: Validated

What Qwen2.5 72B Instruct on 2× RTX 5090 was checked against
CheckTargetResultStatus
Fits in GPU memory64.0 GB usable59.9 GB neededPass
Headroom5% free4.1 GB free (6%)Pass
Context for chat16K or more32KPass
Writing speed15 tok/s or more23 tok/sPass
First replyunder 3 s, 2K prompt0.8 sPass

Targets: a reply you can read along with, a first reply the use case tolerates, and 5% of memory left free. Calculated from published specs, not benchmarked.

The setup

Model
Qwen2.5 72B Instruct
Parameters
72.7B
Precision
5-bit (Q5_K_M)
Context
32K, auto
Chats at once
1
Use
Chat
Runs in
Ollama / LM Studio
Hardware
2× RTX 5090
GPU can use
64.0 GB of 64.0 GB

Running cost

At home
$17/mo
770 W at the wall
Renting instead
$117/mo
4× RTX 2000 Ada at $0.96/h

4 h a day, power at $0.18 per kWh, cloud at RunPod on-demand prices as of Nov 22, 2026.

How to run it

llama.cpp
llama-server -hf bartowski/Qwen2.5-72B-Instruct-GGUF
vLLM
vllm serve Qwen/Qwen2.5-72B-Instruct
MLX (Mac)
mlx_lm.chat --model mlx-community/Qwen2.5-72B-Instruct-4bit

Qwen2.5 72B Instruct on similar machines

All machines: Qwen2.5 72B Instruct VRAM requirements. Everything that runs here: what can the 2× RTX 5090 run?

Other models that run on 2× RTX 5090

Questions

Can Qwen2.5 72B Instruct run on 2× RTX 5090?

Qwen2.5 72B Instruct at 5-bit runs on the 2× RTX 5090 with 32K of context at about 23 tokens a second.

Report VR-0LKY387, calculated Nov 22, 2026 from published specs, not benchmarked.