Skip to main content

Single-GPU Local Models

“Runs on one GPU” can mean weights load with CPU offload, one short request completes, or a useful context runs at interactive speed. These are different claims.

Memory Budget

Raw weight memory starts with:

weight GiB ≈ parameters × bits per weight ÷ 8 ÷ 2^30

A nominal 7B model at 4 bits is about 3.3 GiB of raw weights; a nominal 24B model is about 11.2 GiB. Real artifacts add quantization metadata, embeddings or unquantized tensors, allocator/workspace overhead, and runtime buffers.

KV-cache memory is architecture-dependent but grows approximately with layers, retained KV heads/head dimension, bytes per cache element, batch, and sequence length:

KV bytes ≈ 2 × layers × KV width × sequence × bytes × batch

Always use runtime measurements or an architecture-aware calculator for the actual model. Reserve output tokens and concurrent requests before declaring a fit.

For MoE models, active parameters estimate compute per token; the total experts still contribute to weight storage unless a specialized offload scheme says otherwise.

Starting Tiers, Not Guarantees

Available VRAMReasonable first trialLikely compromise
8 GB3–4B at 4/5-bit; selected 7B at short contextcapacity, context, or CPU offload
12 GB7–8B quantized; cautious 12–14B experimentscontext and speed headroom
16 GB12–14B quantized; supported gpt-oss-20b MXFP4 pathtemplate/KV/runtime constraints
24 GB20–24B quantized; selected 30B MoE artifactslong-context and concurrency pressure

CPU offload expands placement, not VRAM. It may reduce throughput enough to make an autonomous loop impractical.

Candidate Facts

  • Qwen3-4B: 4B, native 32K, thinking/non-thinking control, broad runtime support; the card warns against greedy decoding in thinking mode because of degradation/repetition.
  • Phi-4-mini-instruct: 3.8B and advertised 128K context for constrained use; its card explicitly notes limited factual capacity and stronger Python coverage.
  • gpt-oss-20b: 21B total/3.6B active; OpenAI states the MXFP4 configuration runs within 16 GB of memory and requires Harmony formatting.
  • Mistral Small 3.1 24B: its card says a quantized build can fit an RTX 4090 or 32 GB Mac, while BF16 serving needs roughly 55 GB. Precision is part of the claim.

These facts create candidates, not a quality ranking.

Worked Fit Procedure

  1. Define task, concurrency, context, output reserve, and latency target.
  2. Pick an exact artifact and compute raw weight size.
  3. verify template and runtime support before download if possible.
  4. Start at native/short context and concurrency 1.
  5. Measure peak VRAM/RAM, offload, first-token latency, decode speed, and accepted result.
  6. Increase context independently; stop before OOM or unacceptable latency.
  7. Keep 10–20% operational headroom rather than tuning to one lucky peak.

Fit Versus Usefulness

A smaller model that reliably emits valid tools may outperform a larger offloaded model in an agent loop. A larger context that lowers speed and attention quality may be worse than retrieval into a shorter window. Decide with the local evaluation protocol, not parameter count.

Bias and Safety

These tiers are shaped by consumer NVIDIA-style VRAM and common quantizations. Apple unified memory, AMD, integrated GPUs, NPUs, power limits, and server accelerators behave differently. Model-card memory claims may use a favorable runtime and context. Bind local servers to loopback, review licenses and telemetry, and do not equate local execution with automatic privacy or safety.