Used AI GPU Evaluator & Local Model Fit Analyzer

Roofline Engine v2.4
VRAM Allocation Fit
Comfortable Fit
6.5 GB / 24.0 GB Used
Roofline Operational Limit
Memory Bandwidth Bound
936 GB/s @ 3.56 FLOP/B
Est. Generation Speed
121.2 tok/s
Batch Size 1 Peak Prompt/Gen
Second-Hand Value ($/VRAM)
$28.33 / GB
$19.10 per FP16 TFLOPS

VRAM Allocation Footprint Map

Required: 6.5 GB / Total: 24.0 GB
Weights: 4.50 GB
KV Cache: 1.00 GB
Activations & Overhead: 1.00 GB
Headroom: 17.50 GB

Interactive Roofline Saturation Model (D3.js)

Log-Log Scale (FLOP/Byte vs TFLOPS)

Second-Hand Hardware Inspection Checklist

GDDR6X VRAM Junction Thermals
RTX 3090/3080 series prone to thermal pad oil bleed. Require FurMark/Memtest VRAM temp check under 96°C.
PCIe Lane & Host Bandwidth
Ensure x16 slot bandwidth for fast model loading and multi-GPU tensor parallel offloading.
PSU Transient Power Spikes
Ampere GPUs exhibit 100ms power spikes up to 1.5x TDP. Quality 750W+ ATX 3.0 PSU recommended.
Crypto Mining Thermal Stress
Inspect PCB for discoloration near memory controllers and verify stable PCIe Gen4 negotiating speed.

Used Market Comparison Matrix

GPU Model Price VRAM Bandwidth $/GB VRAM Fit Status Est. Tok/s
Enjoy this tool? Build your own with Super