Can GeForce RTX 4070 run gpt-oss-20b?

⚠️ Runs slowly (CPU offload needed)

Needs CPU offload (16.6 GB total). Estimated 10 tok/s

Memory Needed
16.6 GB
Q4_K_M, 8K context
Est. Speed
~10 tok/s
Generation speed
GPU VRAM
12 GB
GDDR6X

💡 Recommended

For best quality that fits: Q4_K_M quantization

All Quantizations

Q8_0
Won't run
Memory
26.2 GB
Speed
N/A
Bits/weight
8.5
Quality
Highest
Q6_K
Runs slowly
Memory
21.1 GB
Speed
7 tok/s
Bits/weight
6.6
Quality
High
Q5_K_M
Runs slowly
Memory
19.0 GB
Speed
8 tok/s
Bits/weight
5.7
Quality
High
Q4_K_M
RECOMMENDED
Runs slowly
Memory
16.6 GB
Speed
10 tok/s
Bits/weight
4.85
Quality
Good
Q3_K_M
Runs slowly
Memory
14.7 GB
Speed
12 tok/s
Bits/weight
3.9
Quality
Lower
Q2_K
Runs slowly
Memory
11.7 GB
Speed
16 tok/s
Bits/weight
3
Quality
Lower

Run with Ollama

ollama run gpt-oss:20b

This will download and run the Q4_K_M quantization

View all GPUs for gpt-oss-20bView all models for GeForce RTX 4070