LLM

Local LLM Hardware & Token Speed Simulator

Hardware Profile:
Model & Quantization 8B Q4_K_M
Model Parameters 8 Billion
Quantization Precision
FP16 (16-bit)
Q8_0 (8-bit)
Q4_K_M (4.5b)
Q2_K (2.5b)
Active Context Window 4,096 tokens
Batch Size / Streams 1 request
Hardware System Specs
Dedicated GPU VRAM 24 GB
System RAM (CPU Offload) 64 GB
Memory Bandwidth 800 GB/s
Economics & Usage
Local Hardware Investment $1,600
Cloud AI API Subscription $20 / mo
Daily Prompt Requests 500 req/day
Deployment Feasibility
100% VRAM
Fully fits in high-speed GPU VRAM
Gen Throughput Limit
132.8 t/s
Memory bandwidth bound
3-Year TCO Break-Even
N/A
Cloud API cheaper at current rate
Memory Allocation Breakdown 6.0 GB / 24.0 GB
Base Weights: 5.0 GB
KV Cache: 1.0 GB
System RAM Offload: 0.0 GB
VRAM Overhead vs Context Length
3-Year Cumulative Cost ($)
Deployment Launch Parameters
FROM llama3:8b-instruct-q4_K_M # Hardware target: 24.0 GB VRAM | Context: 4096 tokens PARAMETER num_ctx 4096 PARAMETER num_gpu 99
Enjoy this tool? Build your own with Super