B BrunoSanApplied AI for decision-makers

LOCAL AI::FINDER V1

Local AI.
Matched to your machine.

Choose a model, upgrade an existing PC, or buy a complete local-AI system. BrunoSan separates verified specs, observed prices, derived fit and benchmark evidence.

Local AI → LLMs →

01 · DECISION

What are you trying to do?

Start with the decision — not the product catalogue.

02 · FIT ENGINE

Your constraints

Only values that affect the recommendation. Change them and the ranking updates immediately.

WHAT YOU GET

Find the local AI setup that actually fits you.

Choose the model you want to run and tell BrunoSan the constraints that matter. We check memory, RAM, PSU, OS, software stack, price and local-runtime evidence — then show the hardware that truly fits.

FIT INPUT

EUR
GB
W

ROI

ROI scenario

These are your assumptions, not a BrunoSan forecast.

EUR
%
%
EUR/kWh
h/day
Live preview updates automatically. Use the button to commit and share this configuration.

03 · OUTPUT

Best fits

Ranking uses fit, constraints and your selected priority. It does not use affiliate commission.

Fit Index is a deterministic requirements score. Without matching benchmark evidence it is capped at 95 and rounded to 5-point steps.

DEFAULT SCENARIO SNAPSHOT :: HW::intel-arc-pro-b60-24gb

Arc Pro B60 24GB

Intel · new · DERIVED_FIT

95FIT INDEX / 100
VERIFIED SPECOBSERVED PRICEDERIVED FITCURRENT OBSERVATION · 2026-09-13
24GB
VRAM / FAST MEMORY
456GB/s
BANDWIDTH
200W
BOARD POWER
€729PRICE OBSERVED
CRAWLER-VISIBLE DEFAULT EXAMPLE

Default inputs: upgrade existing PC · €900 budget · Qwen3-Coder-30B-A3B · 32 GB system RAM · 650 W PSU · open software stack. This is a deterministic fit example, not a benchmark.

Benchmark truth: No model/runtime-specific tokens/s evidence is published yet, therefore no tokens/s number is claimed.

Context truth: Fast-memory fit does not claim maximum context or KV-cache headroom.

Verified spec source → · Observed price source →

Bandwidth is used only as a performance proxy when no model-specific benchmark is present; cross-vendor TOPS are not ranked against each other.

04 · HARDWARE RADAR

Hardware radar

A compact V1 catalogue. Price is an observation, never a permanent product property.

GPU DETAIL LIBRARY →
ProductMemoryBandwidthPowerStackObserved priceTruth
Arc Pro B60 24GBIntel · GPU · HW::intel-arc-pro-b60-24gb24 GB456 GB/s200 WIntel XPU · oneAPI · SYCL€7292026-09-13OBSERVED PRICEspec · price
GeForce RTX 3090 24GB (used)NVIDIA · GPU · HW::nvidia-rtx-3090-24gb-used24 GB936 GB/s350 WCUDA · PyTorch · vLLM€8502026-09-13MARKET ESTIMATEspec · price
Arc Pro B65 32GBIntel · GPU · HW::intel-arc-pro-b65-32gb32 GB608 GB/s200 WIntel XPU · oneAPI · SYCL€1 2342026-09-13OBSERVED PRICEspec · price
Arc Pro B70 32GBIntel · GPU · HW::intel-arc-pro-b70-32gb32 GB608 GB/s230 WIntel XPU · oneAPI · SYCL€1 5242026-09-13OBSERVED PRICEspec · price
Radeon AI PRO R9700 32GBAMD · GPU · HW::amd-radeon-ai-pro-r9700-32gb32 GB640 GB/s300 WROCm · HIP · PyTorch€1 7632026-09-13OBSERVED PRICEspec · price
Radeon RX 7900 XTX 24GBAMD · GPU · HW::amd-radeon-rx-7900-xtx-24gb24 GB960 GB/s355 WROCm · HIP · PyTorch€1 2542026-09-13OBSERVED PRICEspec · price
GeForce RTX 4090 24GB (used market)NVIDIA · GPU · HW::nvidia-rtx-4090-24gb-used24 GB1008 GB/s450 WCUDA · PyTorch · vLLM€2 4002026-09-13MARKET ESTIMATEspec · price
EVO-X2 128GBGMKtec · Mini PC · HW::gmktec-evo-x2-128gb128 GB256 GB/s120 WROCm · Vulkan · llama.cpp€3 2302026-09-13OBSERVED PRICEspec · price
MS-S1 MAX 128GBMinisforum · Mini PC · HW::minisforum-ms-s1-max-128gb128 GB256 GB/s120 WROCm · Vulkan · llama.cpp€3 9992026-09-13OBSERVED PRICEspec · price
DGX Spark 128GBNVIDIA · Mini PC · HW::nvidia-dgx-spark-128gb128 GB273 GB/s140 WCUDA · PyTorch · vLLM€4 8002026-09-13OBSERVED PRICEspec · price

05 · METHOD

Truth before ranking.

BrunoSan keeps product, variant, offer and benchmark as separate objects. A fit can be derived from memory and software facts; tokens/s require model/runtime-specific benchmark evidence.

VERIFIED_SPEC

VERIFIED_SPEC — manufacturer or licensed hardware source.

OBSERVED_PRICE

OBSERVED_PRICE — time-stamped market observation.

MARKET_ESTIMATE

MARKET_ESTIMATE — market range, not a live offer.

DERIVED_FIT

DERIVED_FIT — deterministic fit from stated requirements.

BENCHMARKED

BENCHMARKED — model/runtime-specific measurement.

Benchmark rule

No matching benchmark means no tokens/s claim. BrunoSan shows missing evidence instead of manufacturing precision from bandwidth or cross-vendor TOPS.

Affiliate policy

Offers may contain affiliate links when merchant feeds are connected. Commission is applied after ranking and can never influence fit score, order or recommendation.

06 · NEXT STEP

From one GPU to a private AI stack.

Need this for a team, company knowledge or a governed on-prem deployment? BrunoSan can turn the hardware decision into an operating architecture.

Plan a Local-AI setup →