PC compatibility
Which GPUs fit
your actual build?
Enter your CPU, RAM, motherboard and PSU. We run a real power budget — not just “does the wattage look big enough” — and tell you which cards drop in cleanly, which are tight, and which need a server chassis.
Real power budget
CPU load + GPU + rig baseline, sized to 70% PSU load.
Connectors & class
Flags 12VHPWR and datacenter-only cards honestly.
RAM & PCIe
System RAM for loading, PCIe gen for bandwidth.
Any modern ×16 slot takes any card — gen affects bandwidth, ITX affects clearance.
—
Your GPU shortlist
—How we check
Precision, not a wattage glance.
Power budget
Draw = GPU + CPU peak + 90 W baseline (board, RAM, drives, fans). Recommended PSU sizes that to ~70% load for efficiency and transient spikes, then takes the higher of that and the card's vendor rating.
Class & connectors
Datacenter cards (A100, L40S, H100, H200) are passively cooled server parts — flagged server-only, not "compatible." Each card lists its power connector so you can check your PSU has it.
RAM & PCIe
System RAM should meet or beat the card's VRAM to load models comfortably. Any ×16 slot fits any card; PCIe gen only changes transfer bandwidth, so it's a note, never a blocker.
Fits the build — but is it the right buy?
Compatible is step one.
A card that fits your PSU still has to fit your models and your budget. The Workspace ranks the ones that do, then calls buy-vs-rent.
Evidence & method
How this calculation works
Checks a model's estimated memory footprint against your PC's GPU and system memory to see whether it runs, and how.
Data sources
- NVIDIA, AMD & Intel GPU documentation
- Hugging Face model cards
- Official model papers
- Bitpute Methodology
Assumptions
- Weights = parameters × bits-per-weight ÷ 8
- Runtime overhead = 0.75 GB + 5% of weights (CUDA context, allocator, buffers)
- Recommended VRAM = total × 1.10 (10% headroom)
- Usable VRAM = card capacity × 0.95
- KV cache counted only when the model architecture is known; when applied, batch size 1, FP16 KV, context as entered
- Single GPU, inference workload
Limitations
- Actual VRAM varies by framework (PyTorch, llama.cpp, vLLM, TGI), driver/CUDA version and OS
- Quantization implementations differ; real bits-per-weight can vary from the nominal value
- When architecture is unknown the KV cache is omitted, so long-context use will exceed the estimate
- Estimates are for planning and comparison, not a guarantee
Data status: Hardware specs — vendor documentation. Model metadata — community model cards, not independently verified.
Related
Why does this estimate differ from other calculators?
- Different footprint formulas and rounding
- Different KV-cache assumptions (batch, context, precision)
- Framework and inference-engine differences
- Reserved and fragmented VRAM
- Driver and CUDA overhead
- Precision and quantization choices