Click legend items to show or hide each series (filters both charts). Scroll to zoom, drag to pan.
💡 Reading the Chart
- X-axis (Price): Purchase price in EUR (Marktplaats.nl) or € equivalent for 500 hrs (cloud)
- Top chart (Bandwidth): Y-axis = memory bandwidth in GB/s - higher is better for LLM inference
- Bottom chart (VRAM): Y-axis = VRAM in GB - model size capacity
- Colors: 🟢 NVIDIA • 🔴 AMD • 🔵 Intel (manufacturer brand colors)
- Shapes: Circle, triangle, rect, cross, star — each series has its own shape
- Best Value: Look for points at the top-left (high bandwidth, low price)
- Filter: Click legend items at the top of the chart to show or hide each series
- Zoom & Pan: Scroll to zoom in/out, drag to pan. Double-click to reset zoom.
- Hover: Mouse over points to see detailed specs including VRAM
Best Budget Value
RTX 3060 12GB
12GB + INT8 @ €225
Best Bandwidth/Price
RTX 3080 10GB
760 GB/s @ €375
Best AMD Value
RX 7900 XTX
24GB + 960 GB/s @ €725
Best Intel Budget
Arc B580
12GB + 456 GB/s @ €255
Best Performance
RTX 5090
32GB + 1792 GB/s + FP4
🎯 Key Insights by Series
- 🟢 NVIDIA RTX 30/40/50: Excellent value, INT8/INT4/FP4 support, bandwidth 224-1792 GB/s
- 🟣 NVIDIA Cloud: T4 to B200 - price = € equivalent for 500 hrs rental
- 🖥️ NVIDIA DGX Spark: Compact AI system (~$4.7k / €2.8k–4.2k) — 128GB, 273 GB/s, 1 petaFLOP FP4
- 🔴 AMD ROCm: RDNA 2-4 + Instinct MI300X, 188-5300 GB/s, ROCm for inference
- 🔵 Intel ARC: Alchemist & Battlemage, XMX INT4, 188-616 GB/s, SYCL/llama.cpp
- Sweet Spots: RTX 3060 12GB (budget), Arc B580 (Intel), RX 7900 XTX (AMD), RTX 5090 (performance)