8B LLM Starter Build
The smallest sensible desktop for running 7B and 8B class models well: a 16 GB graphics card, a fast 6 core CPU, 64 GB of system memory and a quick 2 TB drive.
See the parts listAI hardware directory
AI hardware, specced and priced in one place.
Every GPU, CPU, mini PC, server and part in this directory carries manufacturer specs, and every price we show names the maker or store that lists it. Find the card that fits your model, then find the store near you that sells parts.
1,122hardware products
421stores listed
4countries covered
171GPUs with VRAM data
Top GPUs by VRAM with US prices
Consumer cards from the directory with a US list price in the data, sorted by memory. VRAM decides which models you can run locally.
| GPU | VRAM | AI TOPS | TDP | US price |
|---|---|---|---|---|
| AMD Instinct MI350X | 288 GB | Not published | 1400 W | $39,999 (Estimate) |
| AMD Instinct MI355X | 288 GB | Not published | 1400 W | $42,999 (Estimate) |
| NVIDIA B300 SXM | 288 GB | Not published | 1400 W | $69,999 (Estimate) |
| NVIDIA GB300 NVL72 | 288 GB | Not published | 1400 W | $3,750,000 (Estimate) |
| AMD Instinct MI325X | 256 GB | Not published | 1000 W | $34,999 (Estimate) |
| AMD Instinct MI300X | 192 GB | Not published | 750 W | $29,999 (eBay) |
| NVIDIA B200 SXM | 192 GB | Not published | 1200 W | $54,999 (Estimate) |
| NVIDIA GB200 NVL72 | 192 GB | Not published | 1200 W | $3,000,000 (Estimate) |
| NVIDIA H200 NVL | 141 GB | Not published | 600 W | $39,999 (Estimate) |
| NVIDIA H200 SXM | 141 GB | Not published | 700 W | $42,999 (Estimate) |
Browse by category
Build recipes
Ready made parts lists assembled from the directory, each with its prices shown the honest way.
70B Single GPU Inference Build
A 24 GB card is the classic way to run a 70B model on one GPU at INT4, or a 32B model at FP16. This build pairs the RTX 4090 with a strong CPU and Gen5 class storage.
See the parts listDual GPU Training Build
Two RTX 5090 cards give 64 GB of combined VRAM for training runs and big inference jobs, with a 16 core CPU feeding them and a Gen5 drive for checkpoints.
See the parts listBudget AI Mini PC Setup
One small box, nothing else to buy: the Mac mini M4 with 16 GB of unified memory runs 7B and 8B models quietly on a desk and sips power.
See the parts listQuiet AI Workstation
A compact workstation with 36 GB of unified memory that runs 32B class models at INT4 without a fan roaring next to you: the Mac Studio M4 Max.
See the parts listRack AI Server
Data center class inference and training: 8 GPU systems and EPYC powered nodes. These are sold on vendor quote, so the build is priced per part rather than totaled.
See the parts list
Will it run your model? Ask the VRAM Fit Checker.
Pick a model and a precision, and the checker works out the memory it needs and lists every GPU, mini PC and workstation in the directory that can hold it.
Run the free checkerStores by state and region
421 PC hardware stores, from national chains to single city shops, listed by state across the US, India, Australia and the UK.
BuildSmith: the build planner app, in the works.
Tell it the model you want to run and your budget, and BuildSmith assembles a parts list from this directory, checks VRAM fit, PSU size and case room, and finds the stores in your state. Join the waitlist for the launch note.
Latest from the blog
November 14, 2026
The 16GB Line: GPU VRAM Tiers for Local LLMs, Priced and Counted
All 171 GPUs in the directory sorted into the seven VRAM tiers, with the cheapest way into each tier and the models that honestly fit there, from 8 GB quantization territory to the H200.
November 7, 2026Every GPU in Our Directory, Ranked by VRAM
All 171 GPUs in the directory lined up by memory, from 288 GB accelerators to 8 GB entry cards, and what each band is good for.
November 7, 2026The Cheapest Ways to 24 GB or More of VRAM, by List Price
Every GPU in the directory with 24 GB or more and a published US list price, ranked by price and by cost per gigabyte.
November 7, 2026Mini PCs Ranked by Published AI TOPS
The mini PCs in our directory that publish an NPU figure, ranked, with memory and list prices beside the headline number.
November 7, 2026Gen5 vs Gen4 SSDs for AI Work: What the Read Speeds Say
Peak read figures from the spec sheets in our directory, what they mean for model loading and datasets, and when Gen4 is the smarter buy.
November 7, 2026EPYC Turin vs Xeon 6 for AI Servers: The Spec Sheet Fight
Core counts, TDP and lineup shape for the two server CPU families in our directory, and what matters when the GPUs do the heavy lifting.