Hardware Tiers & Form Factors
How we categorize complete systems versus standalone upgrade components.
Tier 1: Starter Machines
Budget-conscious prebuilts with 12GB to 24GB of fast memory. Ideal for everyday chat, personal coding assistance, and image generation using 8B to 14B models.
Tier 2: Sweet Spot Workstations
The professional developer bracket featuring 24GB to 64GB memory. Built for multi-file coding agents, autonomous pipelines, and 32B parameter reasoning models (Qwen 2.5 32B, DeepSeek-R1 32B).
Tier 3: Maximum Power Systems
High-memory workstations with 96GB to 256GB+ unified memory or 32GB CUDA setups. Runs full 70B models with long context windows and 120B+ MoE checkpoints.
Standalone GPU Component Cards
Indicates a discrete graphics card sold on its own (such as an RTX 5080 or RTX 5090) rather than a complete computer. Clearly labeled to prevent mistaking a £700 upgrade card for a turnkey PC.
Memory Architecture & Compute Badges
The physical mechanism your computer uses to store model weights and run parallel tensor calculations.
Dedicated Graphics (dGPU)
The machine features a separate graphics card (NVIDIA RTX or AMD Radeon) with its own high-speed GDDR6/GDDR7 video memory. Delivers the fastest token generation speeds and native CUDA software support.
Unified System Architecture
Found in Apple Silicon (Mac Studio, Mac Mini) and AMD Ryzen AI Max systems. The CPU, GPU, and Neural Engine share a single high-bandwidth memory pool, allowing you to allocate up to 75%–80% of total system RAM to large AI models at low power draw.
Dual-GPU & Rack Configurations
Systems with 2 or more physical graphics cards (e.g. 2x RTX 3090 or dual RTX 5090s) connected over PCIe slots. Splitting model layers across two cards pools 48GB–64GB of fast GDDR VRAM.
Measured 70B Generation Speed
The benchmarked or estimated output rate when generating text with a 70-billion parameter model quantized to 4 bits (Q4_K_M). Human reading speed is ~4 to 6 tokens/sec; high-speed autonomous coding agents typically benefit from 15+ tokens/sec.
Model Fit & Value Badges
Calculated automatically from memory capacity, thermal limits, and index-wide price rankings.
48GB+ Fast Memory Milestone
Awarded only to machines with at least 48GB of accessible fast memory. A 70-billion parameter model compressed to 4-bit quantization requires ~40GB to load weights, plus 4GB–8GB for KV context cache. Machines with this badge run 70B models completely in fast memory without offloading to slow storage.
32GB Fast Memory Milestone
Awarded to 32GB–47GB hardware (e.g. RTX 5090 or Mac Mini M4 Pro 48GB). Confirms that 32B reasoning models (Qwen 2.5 32B, DeepSeek-R1 32B) fit cleanly at Q8 quantization or full precision with ample context buffer.
Cost-Per-Gigabyte Metric
Calculated by dividing the lowest verified retailer price by usable fast memory in gigabytes. Rigs that achieve the lowest number within their hardware tier or VRAM bracket earn competitive value honors.
Low Power & Thermal Profile
Awarded to hardware pulling 150 watts or less under full local inference. These machines run quietly in home offices and can run 24/7 background agent pipelines with minimal impact on electricity bills.
Editorial Recommendation
Designates a standout benchmark reference machine chosen by our editorial team for exceptional engineering balance, verified thermal stability, and unmatched value in its class.
Verified Deals & Freshness Badges
Automated proof of substantive price drops derived from append-only historical database tracking.
Age-Gated Deal Badges
Deal badges are mathematically restricted based on continuous logging history. A “30-Day Low” badge requires at least 30 days of logged prices before it can trigger; “90-Day Low” requires 90+ days. If an offer does not beat all previous recorded prices, no deal badge is displayed.
Verification Freshness
Every retailer URL is checked periodically by our sync scrapers to confirm stock availability and live pricing. Each badge is backed by a machine-readable ISO-8601 timestamp in the page HTML so search engines and users can verify data freshness.
Rig Finder Recommendation Badges
Labels assigned by our 3-Question Rig Finder when matching your workload, budget, and form factor.
Top Recommendation
The highest-scoring system that satisfies your exact workload target (e.g. 70B reasoning or daily coding) while remaining comfortably within your chosen budget bracket and form factor.
Strong Alternative
The runner-up recommendation that also satisfies your workload, providing a different memory technology (e.g. discrete CUDA vs. unified memory) or slightly different cost-to-performance ratio.
Closest Fallback Alternative
Shown when an exact filter combination has limited options (e.g. seeking an Upgradeable Tower under £1,500 for 70B models). Clarifies the nearest high-performing hardware that achieves your goal.
Not Ready to Buy a Full Rig?
Don't want to spend £3,500+ upfront? Rent the GPU power instead.
Buying a dedicated AI workstation is a serious investment—and if you aren't ready to spend that much right now, you don't have to. Renting cloud GPUs gives you instant access to high-VRAM hardware like RTX 4090s and H100s on demand. Spin up an instance for fine-tuning or running massive 70B+ models, pay strictly for the minutes you use, and avoid the upfront hardware cost, electricity bills, and maintenance.
Vast.ai Global GPU Marketplace
Lowest $/hrRunPod Community & Secure Cloud
Community & Secure CloudFindBestRig independently benchmarks cloud GPU providers to compare rental economics against physical hardware.