Leaderboard

NVIDIA

RTX 4070 SUPER

12 Go

VRAM

—

Best generation

—

Efficiency (ref.)

—

Elec. cost / M tokens (ref.) · 0,25 €/kWh

What this card can run

Gemma 3 270M Q4_K_M

0.25 Go

Not measured

Qwen2.5 0.5B Q4_K_M

0.4 Go

Not measured

Qwen2.5 Coder 0.5B Q4_K_M

0.4 Go

Not measured

Qwen3 0.6B Q4_K_M

0.48 Go

Not measured

Qwen3.5 0.8B Q4_K_M

0.58 Go

Not measured

LFM2.5 1.2B Q4_K_M

0.73 Go

Not measured

Llama 3.2 1B Q4_K_M

0.81 Go

Not measured

Qwen2.5 1.5B Q4_K_M

0.99 Go

Not measured

Qwen2.5 Coder 1.5B Q4_K_M

0.99 Go

Not measured

R1 Distill 1.5B Qwen Q4_K_M

1.12 Go

Not measured

Qwen3 1.7B Q4_K_M

1.28 Go

Not measured

Qwen3.5 2B Q4_K_M

1.4 Go

Not measured

Gemma 2 2B Q4_K_M

1.71 Go

Not measured

Qwen2.5 3B Q4_K_M

1.93 Go

Not measured

Qwen2.5 Coder 3B Q4_K_M

1.93 Go

Not measured

Llama 3.2 3B Instruct Q4_K_M

2.02 Go

Not measured

Ministral 3 3B Q4_K_M

2.15 Go

Not measured

Granite 4.1 3B Q4_K_M

2.17 Go

Not measured

Phi-4 mini 3.8B Q4_K_M

2.49 Go

Not measured

Qwen3 2507 4B Q4_K_M

2.5 Go

Not measured

Qwen3 4B Q4_K_M

2.5 Go

Not measured

Gemma 3n E2B Q4_K_M

2.79 Go

Not measured

Qwen3.5 4B Q4_K_M

3.01 Go

Not measured

Gemma 4 E2B Q4_K_M

3.46 Go

Not measured

Gemma 3n E4B Q4_K_M

4.24 Go

Not measured

Mistral 7B 7B Q4_K_M

4.37 Go

Not measured

Qwen2.5 7B Instruct Q4_K_M

4.68 Go

Not measured

Qwen2.5 Coder 7B Q4_K_M

4.68 Go

Not measured

R1 Distill 7B Qwen Q4_K_M

4.68 Go

Not measured

Meta Llama 3.1 8B Instruct Q4_K_M

4.92 Go

Not measured

Qwen3 8B Q4_K_M

5.03 Go

Not measured

R1 0528 (Qwen3) 8B Q4_K_M

5.03 Go

Not measured

Ministral 3 8B Q4_K_M

5.2 Go

Not measured

Gemma 4 E4B Q4_K_M

5.41 Go

Not measured

Gemma 2 9B Q4_K_M

5.76 Go

Not measured

Qwen3.5 9B Q4_K_M

6.17 Go

Not measured

Gemma 4 12B Q4_K_M

7.66 Go

Not measured

Ministral 3 14B Q4_K_M

8.24 Go

Not measured

Qwen2.5 14B Instruct Q4_K_M

8.99 Go

Not measured

DeepSeek R1 Distill Qwen 14B Q4_K_M

8.99 Go

Not measured

Qwen2.5 Coder 14B Q4_K_M

8.99 Go

Not measured

Qwen3 14B Q4_K_M

9 Go

Not measured

Phi-4 Q4_K_M

9.05 Go

Not measured

Coder V2 Lite 16B-A2.4B Q4_K_M

10.36 Go

Doesn't fit in VRAM

Smooth ≥ 60 tok/s (faster than you read) · Comfortable 30-60 · Slow < 30 · "Doesn't fit" = model size + 2 GB headroom > VRAM (CPU offloading is excluded from the protocol: the numbers would be meaningless).

Measurements by model

No runs for this card yet. Your move: run the bench.

VRAM used = peak system GPU memory during the run (includes the OS, roughly 0.5–1 GB more than the model alone).