Llama-3.1-8B-Instruct-4bit
7 workload results across 3 hardware configurations.
Fastest local config
232.2 decode tok/s
on RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB via llama.cpp. see full run
For a standard 4-bit quantization using MLX, here are measured results for short-chat generation.
On MLX 4-bit, Llama-3.1-8B-Instruct is fastest on the M3 Ultra (60-core GPU) at 130.2 tok/s, 346% faster than the M3 Pro (18-core GPU) (29.20 tok/s).
| Card | Decode tok/s | TTFT (136-token prompt) | Run |
|---|---|---|---|
| M3 Ultra (60-core GPU) | 130.2 (single run) | 340 ms | r_v2pbc0rq2l4 |
| M3 Pro (18-core GPU) | 29.20 (single run) | 669 ms | r_h0-use1ypnb |
Llama-3.1-8B-Instruct, 4-bit, MLX, chat-short, batch 1. Only cards with their standard memory are listed; CPU-only hosts are excluded.
Local runs (7 workload results)
Runs from contributors' own machines via MLX, llama.cpp, vLLM, exllamav2, or ollama. Signed on the submitter's hardware.
RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB
| Workload | Backend | Quant | decode tok/s | prefill tok/s | TTFT | Run |
|---|---|---|---|---|---|---|
| chat-short | llama.cpp | - | 229.4tok/s | no data | 35.6ms | r_qr4srge34da |
| chat-long | llama.cpp | - | 211.4tok/s | no data | 266ms | r_qr4srge34da |
| concurrent-decode | llama.cpp | - | 223.0tok/s | no data | no data | r_qr4srge34da |
| agent-trace | llama.cpp | - | 210.3tok/s | 36,720.3tok/s | 60.8ms | r_qr4srge34da |
| chat-short | llama.cpp | - | 232.2tok/s | no data | 33.8ms | r_kfrkg-vn376 |
M3 Pro (18-core GPU) + 36GB unified
| Workload | Backend | Quant | decode tok/s | prefill tok/s | TTFT | Run |
|---|---|---|---|---|---|---|
| chat-short | mlx@0.31.3 | - | 29.20tok/s | 203.3tok/s | 669ms | r_h0-use1ypnb |
M3 Ultra (60-core GPU) + 96GB unified
| Workload | Backend | Quant | decode tok/s | prefill tok/s | TTFT | Run |
|---|---|---|---|---|---|---|
| chat-short | mlx@0.31.3 | - | 130.2tok/s | 400.3tok/s | 340ms | r_v2pbc0rq2l4 |
Llama-3.1-8B-Instruct-4bit on hardware
Request a Llama-3.1-8B-Instruct-4bit run
No measured run yet for these pairs. Vote for the one you want; the most-requested runs go first. See all requests.
We never sell or share your email. Privacy