Local AI hardware,
benchmarked honestly.
The independent hardware reference for builders who run AI on their own machines, anywhere in the world.
Today's leaderboard is curated from public benchmark logs, community submissions, and cited vendor disclosures. Each record shows its verification status so builders can judge confidence at a glance.
benchmark records
devices tracked
workloads tracked
hands-on tutorials
Start here
Three steps in the order I actually take them when someone asks me what to buy: figure out the GPU, check the model fits, then assemble the rig around it.
Why builders can trust it
If you cannot fact-check us, we have failed. Every claim has a source, every aggregation has a tier, every gap has a date.
What we claim, precisely
MyAIHardware is strongest when it behaves like an open reference desk for local AI builders, not a black-box review site. The premium move is transparency.
4
clear verification tiers
75
devices tracked in the corpus
public
API, feed, and changelog links
Hardware Reference
Detailed specs, benchmarks, and AI usefulness for every device
NVIDIA GeForce RTX 5090 32GB
NVIDIA GeForce RTX 4090 24GB
NVIDIA GeForce RTX 5080 16GB
NVIDIA GeForce RTX 5070 Ti 16GB
NVIDIA GeForce RTX 4070 Ti 12GB
NVIDIA GeForce RTX 4060 Ti 16GB
NVIDIA GeForce RTX 3090 24GB
NVIDIA GeForce RTX 3060 12GB
NVIDIA H100 SXM5 80GB
NVIDIA H200 141GB
NVIDIA B200 192GB
AMD Instinct MI300X 192GB
NVIDIA L40S 48GB
NVIDIA A100 SXM4 80GB
Apple M4 Max (40c GPU, 128GB)
Apple M3 Ultra (80c GPU, 512GB)
AMD Radeon RX 7900 XTX 24GB
Intel Arc B580 12GB
NVIDIA RTX 6000 Ada 48GB
MyAI Bench, Live
Top hardware running Llama 3 8B Q4 right now
AMD Instinct MI300X 192GB
NVIDIA GeForce RTX 5090 32GB
NVIDIA GeForce RTX 4090 24GB
NVIDIA B200 192GB
AWS Trainium2
Your AI Hardware Playbook
Five guides that answer every builder's first question
Don't guess your build. Calculate it.
Two free tools that pay for themselves on your first GPU purchase. Configure a complete AI workstation or check if a model will fit your VRAM before you buy.
Local AI Tutorials
Step-by-step walkthroughs from install to inference
GPU Price Tracker
Trending price drops from our latest reference dataset
Built in public
Real benchmarks. No fluff.
Every benchmark should be legible: where it came from, how much to trust it, and how to export it into your own workflow.
Community Spotlight
Illustrative build concepts today. Real community submissions open in Q3 2026.
Preview mode: These cards are realistic example configurations to help builders reason about budgets, thermals, and throughput before the live submission system launches.
The $1,200 Used-3090 Whisper Box
by Illustrative build concept
Single RTX 3090 (used eBay), Ryzen 5 7600, 64GB DDR5. Designed as a realistic reference configuration for Whisper Large v3 and Llama 3 8B builders with a roughly $1,200 target budget.
Hot Right Now
- NVIDIA B200 announced+2,340%
- Copilot+ PC sales+890%
- AMD MI325X leaks+670%
- Apple M4 Neural Engine+520%
- Groq LPU pricing+410%
Never Miss a Benchmark
Weekly AI hardware news + deals
Performance Leaderboards
See how the latest AI hardware stacks up
Latest News
Explore by Category
Deep dives into every type of AI accelerator
Shop Trending on Amazon
Popular AI hardware with current pricing
Pick the model. Compare the hardware.Trust the evidence.
The new product spine answers the three questions buyers actually have: what runs well, what breaks first, and which device is the smarter buy for this exact workload.
Rankable devices
54
GPUs plus Apple silicon paths
Verified rows
1
Explicit verification state on the public corpus
Confidence rows
9
Benchmark rows with confidence intervals
Source links
156
Traceable citations behind the scores
Model-to-hardware picks
Evidence-led recommendations
Llama 3 70B Q4
48GB+ native, 24GB with partial offload compromises.
8× NVIDIA H100 SXM5 80GB
412 tok/s
Verified local-lab ingestion
Your RTX 5080 is in the evidence layer now
Fresh ingest from NVIDIA GeForce RTX 5080 16GB using ollama-local-api, 2048 context, and consecutive warm-model runs.
kumru-2b
2.4B / Q4_K_M / n=5
brooqs-mistral-turkish-v2-latest
7.2B / Q4_0 / n=5
Public trust layer in practice:
- Source links and verification status live on every benchmark-backed hardware page.
- Run counts, confidence intervals, and freshness are surfaced instead of buried.
- Affiliate intent is separated from benchmark evidence and labeled clearly.
Head-to-head pages
Comparison pages worth publishing
RTX 4090 vs RTX 5090
Flagship-to-flagship deltas with shared benchmark rows instead of spec-sheet guesswork.
Open compare pageRTX 5080 vs RTX 3090
New-gen efficiency and software freshness versus used-market brute force.
Open compare pageApple M4 Max vs RTX 5080
Unified-memory laptop class against a desktop GPU on real local-AI workloads.
Open compare pageCommunity Voices
What our readers say about MyAIHardware
“The RTX 5090 review saved me $3,000. Their benchmarks caught thermal throttling that no other reviewer mentioned. Absolute gold.”
Alex K.
ML Engineer
“Finally, a site that understands AI hardware. The NPU database helped us choose the right laptops for our entire team.”
Priya M.
AI Startup Founder
“Their 'What Can It Run?' tool is brilliant. I use it before every hardware purchase. The price tracker alone has saved me thousands.”
Jordan T.
Data Scientist
Built in the open
Benchmark data is curated from public sources with attribution. We're building toward an in-house test lab (Q3 2026). Want to contribute benchmarks? Submit a run.
The Numbers Don't Lie
Our growing database of tested AI hardware
GPUs Benchmarked
NPUs Tested
Laptops Reviewed
Edge Devices
ASICs Analyzed
Benchmarks Run
Stay Ahead of the AI Curve
Get weekly AI hardware news, benchmark updates, and deals in your inbox. Founding-subscriber list, be one of the first.