GPU Computing Hardware Planning

Llama3 Inference Speed on GPUs (tokens/second)

Performance Comparison of Mainstream GPUs on Llama3 8B / 70B

GPU	Memory(VRAM)	8B Q4_K_M	8B F16	70B Q4_K_M	70B F16
RTX 4090	24GB	127.74	54.34	Out of Memory	Out of Memory
RTX A6000	48GB	102.22	40.25	14.58	Out of Memory
L40S	48GB	113.60	43.42	15.31	Out of Memory
RTX 6000 Ada	48GB	130.99	51.97	18.36	Out of Memory
A100	80GB	138.31	54.56	22.11	Out of Memory
H100	80GB	144.49	67.79	25.01	Out of Memory
M2 Ultra	192GB	76.28	36.25	12.13	4.71

---

VRAM Requirements for Llama3 Models

Model	Q4_K_M (Quantized)	F16 (Original)
Llama3 8B	4.58 GB	14.96 GB
Llama3 70B	39.59 GB	131.42 GB

Source

GitHub - XiongjieDai/GPU-Benchmarks-on-LLM-Inference: Multiple NVIDIA GPUs or Apple Silicon for Large Language Model Inference?GitHub

---

Hardware Configuration Recommendations

MaiAgent recommends two combinations suitable for different groups:

1. Two H100(80GB): Higher budget, prioritizing quality and performance

2. L40S(48GB) and RTX 6000 Ada(48GB): Standard budget, focusing on cost-effectiveness

For more detailed information, please contact MaiAgent's professional consultants at [e-mail]