AI Hardware Library

Blackwell-generation datacenter GPU reported around 192GB HBM3e. Placeholder until detailed specs are verified.

An 8-GPU HGX H100 server with ~640GB of aggregate HBM3 — datacenter-scale training and high-throughput serving in one node.

Dell's flagship 8-GPU AI server, configurable with H100 or H200 — a rack-scale building block for AI clusters.

AMD's datacenter accelerator with a class-leading 192GB of HBM3 — exceptional memory capacity for very large models on a single card.

A next-gen Blackwell datacenter profile for frontier-scale training and serving — listed to anchor the top of the cloud tier.

An H100 with a much larger, faster memory system: 141GB HBM3e and ~4.8 TB/s, ideal for long-context and very large models.

A rentable Hopper card with 141GB of fast HBM3e — headroom for very large models and long context without multi-GPU splitting.

The defining datacenter accelerator for generative AI: 80GB HBM3, very high bandwidth, and transformer-optimized tensor cores.

A rentable H100 instance — top-tier inference and fine-tuning without capital expenditure, billed by the hour.

Blackwell-generation professional flagship reported at 96GB GDDR7 — a single-card path to very large local models.

A high-end workstation designed to hold multiple double-width pro GPUs — a deskside multi-GPU AI platform.

A dual-socket workstation built for multiple professional GPUs — suited to teams running heavier local AI on-prem.

Supermicro's deskside multi-GPU tower line — a flexible platform for on-prem local AI with several pro or datacenter cards.

Up to 192GB unified memory at ~800 GB/s — still one of the most memory-rich single-box options for running very large local models.

A top-tier Apple desktop profile: very large unified memory lets a single quiet machine hold flagship models — the configuration is representative until Apple ships this exact part.

A serious local AI workstation: four RTX 4090s pool to 96GB, enough to serve flagship 70B models and run many concurrent agents privately.

A professional tower that can house large pro GPUs (e.g. RTX 6000 Ada / A6000) for serious on-desk local AI.

The datacenter workhorse of the LLM boom: 80GB HBM2e with strong tensor throughput, now widely available used and in the cloud.

A widely available, often cheaper rentable 80GB card — a practical default for fine-tuning and serving when H100 supply is tight.

Up to 128GB of unified memory in a compact desktop — large enough to hold 70B-class models entirely on-device.

Placeholder entry for the Blackwell consumer flagship. Memory is widely reported as 32GB GDDR7; compute and power figures to verify.

A desktop AI appliance built on the GB10 Grace Blackwell superchip with ~128GB unified memory — a personal AI developer box.

A compact x86 mini PC whose large unified memory (up to ~128GB) lets the integrated GPU/NPU run sizeable local models.

A workstation tuned for local coding agents: ~48GB across two 24GB cards runs strong 32B coder models and serves a small engineering team privately.

A versatile 48GB datacenter card for inference and graphics — a popular, cost-effective cloud and on-prem serving option.

A cost-effective rentable 48GB card — a sweet spot for steady mid-scale inference without HBM-tier pricing.

More memory bandwidth and up to 64GB unified memory make this a surprisingly capable small-form-factor local-AI box.

A confidential evidence-search box for legal teams: ~48GB hosts a strong model and large-document retrieval for a DocMatch-style workflow entirely in the firm.

48GB ECC plus Ada-class compute at 300W: the workstation card for demanding local inference and light fine-tuning.

48GB ECC workstation card from AMD — large VRAM at workstation power, contingent on ROCm support for your stack.

48GB of ECC VRAM in a 300W workstation card — the classic choice for serious local model work without datacenter heat.

A back-office box for finance teams: extracts invoices, reconciles data and drives Odoo/ERP workflows on a private 14–32B model.

A tiny, low-power mini PC with shared memory: a frugal way to run a private 7–14B assistant for a small business with almost no noise or running cost.

The fastest consumer GPU for single-card local inference: 24GB VRAM with the highest consumer compute throughput.

A tiny, near-silent desktop that runs small-to-mid LLMs on unified memory — the cheapest credible Apple on-ramp to local AI.

24GB of VRAM at a consumer price — a strong value local-AI card if your stack supports ROCm/Vulkan well.

Still a local-AI favourite: 24GB of VRAM and strong bandwidth make it a value workhorse on the used market.

A budget two-GPU box: pooling two 12GB RTX 3060s gives 24GB total for bigger models or two assistants in parallel on a tight budget.

A quiet small-form-factor box for a single office: enough for a private assistant and light document RAG on a 7–14B model.

An on-site box for hospitality: runs a multilingual guest-messaging and front-desk automation agent without sending guest data to the cloud.

An affordable 16GB card that runs small-to-mid models via Intel's oneAPI/IPEX stack — best for tinkerers comfortable outside CUDA.

Intel's Battlemage budget card with 12GB VRAM — an inexpensive entry point for small local models on the Intel stack.

The budget entry point for local AI: 12GB of VRAM is enough for small quantized LLMs and assistants.