Show HN: Llmspeed – feel how fast a local LLM runs before you buy the hardware

17 October 2026 2 min read هدهد تك | Hudhud Tech

Llmspeed is a tool designed to help users gauge the performance of local language models (LLMs) on various hardware configurations before making a purchase. By selecting specific hardware and model combinations, users can experience the estimated speed of the model in real-time.

For example, a configuration using an NVIDIA RTX 4090 with a Llama 3.1 model can achieve speeds between 80 to 139 tokens per second, with a time-to-first-token (TTFT) of approximately 0.3 seconds. These measurements are based on community benchmarks and specifications.

Hardware Options

Users can browse through a variety of hardware options available on Llmspeed, including:

NVIDIA RTX 3090 24GB
AMD Ryzen AI Max+ 395 128GB (Strix Halo)
NVIDIA RTX 5090 32GB
NVIDIA RTX 4090 24GB
Mac Studio M5 Ultra 512GB
MacBook Pro M5 Max 128GB
NVIDIA RTX 5060 Ti 16GB
NVIDIA DGX Spark 128GB
NVIDIA RTX PRO 6000 96GB
Mac Studio M3 Ultra 512GB
AMD RX 7900 XTX 24GB
NVIDIA RTX 3060 12GB
Mac Studio M4 Max 128GB
NVIDIA RTX 5070 Ti 16GB
AMD RX 9070 XT 16GB
Intel Arc Pro B70 32GB
Mac Mini M4 Pro 64GB
AMD Radeon AI PRO R9700 32GB
NVIDIA RTX 5080 16GB
MacBook Pro M5 Pro 48GB
AMD RX 7900 XT 20GB
NVIDIA RTX 4070 Ti Super 16GB
MacBook M4 Pro 24GB
NVIDIA RTX 4070 12GB
MacBook M3 Max 48GB
CPU-only (Ryzen 9 9950X, 96GB DDR5)
MacBook M2 Pro 16GB
H100 80GB (cloud)
A100 80GB (cloud)
H200 141GB (cloud)

Model Options

Users can also explore various models available on Llmspeed, including:

Qwen 3.6/3.8 27B
GLM-5.3 Flash
GPT-OSS 20B
Gemma 4 26B A4B
Qwen 3.5/3.6 35B A3B
Ornith 1.5 35B A3B
DeepSeek V4 Flash
GPT-OSS 120B
Gemma 4 31B
Llama 3.1 8B
Qwen 3.5 9B
DeepSeek R1 Distill 14B
Ornith 1.5 9B
Muse Glimmer 30B
Qwen3 30B A3B

Estimates for performance are based on short prompts on individual devices, with measured numbers sourced from community benchmarks and projections derived from similar setups. Llmspeed is open source and available on GitHub.