RigRight
An interactive advisor that, given the LLM you want to run and your budget, tells you exactly which GPU, motherboard, and bridge setup will hit your target tokens-per-second without wasted spend or surprise PCIe bottlenecks.
AI-curious power users and small-studio creators building local LLM rigs at home
- Workload-to-hardware matcher: pick your model size (7B, 13B, 70B) and use case (chat, code, long-context agents) and receive a recommended GPU stack with expected tokens/sec
- Bottleneck forecaster that warns when your consumer motherboard will drop to x8/x8 or x4/x4 with a second GPU installed, and predicts whether an NVLink bridge would actually pay off for your workload
- Build-vs-buy comparison showing when renting (RunPod, Vast) beats owning at your usage level
- Used-GPU price check surfacing current second-hand market rates so you don't overpay for hardware that loses value fast
The source analysis shows even technically literate builders are surprised by NVLink-vs-PCIe behavior, and that consumer motherboards from the $1000-1800 'EXTREME AI' tier quietly halve PCIe bandwidth when a second GPU is installed, meaning the wrong $300 bridge purchase is a common and recurring mistake.
Multiple 2025–2026 guide posts actively published on local LLM rig building (Sanj $1500 build, runaihome multi-GPU, compute-market 70B-405B, willitrunai) plus interactive calculator launches (aigpusim, bestgpuforai, abuz8ai) show a real, recurring audience; HN 'When Is NVLink Worth It?' engagement is modest (16 pts / 2 comments) so the builder-curious slice is niche, not mainstream.Build a $1500 AI Powerhouse: The 2025 Guide to Local LLM Hardware ↗Multi-GPU LLM Inference Guide — NVLink vs PCIe, Tensor Parallelism ↗
Goal-driven 'given a budget + target TPS, recommend GPU+motherboard+bridge' advisor is not yet built — existing players are either general GPU tok/s calculators (aigpusim, bestgpuforai, inventivehq, abuz8ai, kamilstanuch) or blog guides (runaihome, compute-market, willitrunai, SabrePC); none combine motherboard PCIe-lane splitting + NVLink bridge advisory + TPS budgeting in one interactive tool.AI GPU Simulator — LLM Performance, VRAM & Tokens/sec Calculator ↗LLM Tokens Per Second Calculator — Predict GPU Inference Speed ↗LLM Tokens Per Second Calculator - Inference Speed Estimator ↗
No evidence of paid LLM-rig advisor subscriptions; the visible monetization pattern in this space is content + affiliate (Sanj $1500 build post, runaihome guide, compute-market guide) and free interactive calculators (aigpusim, bestgpuforai, abuz8ai) — audience is price-sensitive hobbyists making one-off purchases, so premium SaaS willingness-to-pay is unproven.Build a $1500 AI Powerhouse: The 2025 Guide to Local LLM Hardware ↗AI GPU Simulator — LLM Performance, VRAM & Tokens/sec Calculator ↗
Local LLM trend is accelerating (Qwen 3 / DeepSeek-R1 closing the gap, per ikangai 2025 recap) and each new GPU generation resets the calculus — runaihome explicitly flags that the RTX 3090 Ti silently lost NVLink, the kind of trap that keeps the advisory need live; the structural PCIe-lane-splitting problem on consumer boards persists across generations.The Complete Guide to Running LLMs Locally: Hardware, Software, and Performance Essentials ↗Dual GPU for Local AI in 2026: NVLink vs PCIe Bandwidth and Real tok/s ↗
Buildable from public benchmark data already curated by willitrunai (exact tok/s across 2x/4x/8x configs, NVLink vs PCIe), public GPU/board specs, and free calculator precedents (aigpusim, abuz8ai) — main ongoing cost is keeping the model×GPU×bridge benchmark matrix current as new hardware and inference engines (vLLM, exllamav2) ship.Multi-GPU LLM Inference Guide — NVLink vs PCIe, Tensor Parallelism ↗LLM Inference Speed Calculator — Tokens/Sec on Your GPU ↗