Skip to content

Llama 3.3 70B on 4× RTX 3090: VRAM, speed and fit

Llama 3.3 70B at 8-bit runs on the 4× RTX 3090 with 56K of context at about 8.7 tokens a second.

Text generation · Chat · Ollama / LM Studio

Runs comfortably · on 4× RTX 3090 · 8-bit (Q8_0) · Ollama / LM Studio · Chat

GPU memory
90.3 GB
of 96.0 GB
Context
56K
auto
Speed
8.7 tok/s
7.0 when full
First reply
1.3 s
2K prompt

Where the memory goes

Model
69.8 GB
Context
17.5 GB
Overhead
3.0 GB
OS
0.0 GB
Free
5.7 GB
Memory
90.3 GB of 96.0 GB
Context
56K tokens · auto
Speed
8.7 tok/s · 7.0 when full
First reply
1.3 s for a 2K prompt
Running cost
$20/mo at home · $131/mo on a cloud 4× RTX A5000

Validation: Validated with conditions

What Llama 3.3 70B on 4× RTX 3090 was checked against
CheckTargetResultStatus
Fits in GPU memory96.0 GB usable90.3 GB neededPass
Headroom5% free5.7 GB free (6%)Pass
Context for chat16K or more56KPass
Writing speed15 tok/s or more8.7 tok/sCheck
First replyunder 3 s, 2K prompt1.3 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
Llama 3.3 70B
Parameters
70.5B
Precision
8-bit (Q8_0)
Context
56K, auto
Chats at once
1
Use
Chat
Runs in
Ollama / LM Studio
Hardware
4× RTX 3090
GPU can use
96.0 GB of 96.0 GB

Worth trying

  • 8-bit cache: Room for 104K context

Running cost

At home
$20/mo
920 W at the wall
Renting instead
$131/mo
4× RTX A5000 at $1.08/h

4 h a day, power at $0.18 per kWh, cloud at RunPod on-demand prices as of Nov 21, 2026.

How to run it

Ollama
ollama run llama3.3:70b
llama.cpp
llama-server -hf unsloth/Llama-3.3-70B-Instruct-GGUF
vLLM
vllm serve meta-llama/Llama-3.3-70B-Instruct
MLX (Mac)
mlx_lm.chat --model mlx-community/Llama-3.3-70B-Instruct-4bit

Llama 3.3 70B on similar machines

All machines: Llama 3.3 70B VRAM requirements. Everything that runs here: what can the 4× RTX 3090 run?

Other models that run on 4× RTX 3090

Questions

Can Llama 3.3 70B run on 4× RTX 3090?

Llama 3.3 70B at 8-bit runs on the 4× RTX 3090 with 56K of context at about 8.7 tokens a second.

Report VR-176S5NW, calculated Nov 22, 2026 from published specs, not benchmarked.