Hardware source register
Compare source-dated hardware specifications
Browse published memory, bandwidth, compute, power, and interconnect fields alongside the manufacturer or institutional source. The register preserves source terminology so readers can inspect the underlying product record rather than rely on a generated recommendation.
Source fields checked Jul 20, 2026 · re-check by Oct 18, 2026
Vendor-published fields
Specification register
Names and values retain the terminology used in the cited product source. Cross-vendor compute labels, power definitions, memory systems, and form factors may not be directly comparable. Verify the current specification and supported configuration at the original source before relying on any field.
| Product | Class / architecture | Published memory | Published bandwidth | Published compute label | Published power label | Interconnect | Original source |
|---|---|---|---|---|---|---|---|
| GeForce RTX 4060 Ti 16GB | Consumer discrete GPU Ada Lovelace |
16 GB · GDDR6 | 288 GB/s | 4,352 CUDA cores 353 AI TOPS |
165 W total graphics power | PCIe Gen 4 | NVIDIA RTX 40 Series specifications ↗ |
| GeForce RTX 4070 SUPER 12GB | Consumer discrete GPU Ada Lovelace |
12 GB · GDDR6X | 504 GB/s | 7,168 CUDA cores 568 AI TOPS |
220 W total graphics power | PCIe Gen 4 | NVIDIA RTX 4070 family specifications ↗ |
| GeForce RTX 4090 24GB | High-end consumer discrete GPU Ada Lovelace |
24 GB · GDDR6X | 1008 GB/s | 16,384 CUDA cores 1,321 AI TOPS |
450 W total graphics power | PCIe Gen 4 | NVIDIA RTX 40 Series specifications ↗ |
| Apple M4 Max | Consumer/pro unified-memory SoC Apple silicon |
128 GB · Unified memory (maximum configuration) | 546 GB/s | Up to 40-core integrated GPU Architecture-specific; not directly comparable to NVIDIA AI TOPS |
System-level notebook/workstation envelope | On-package shared CPU/GPU memory fabric | Apple M4 Pro and M4 Max announcement ↗ |
| NVIDIA H200 SXM | Data-center accelerator Hopper |
141 GB · HBM3e with ECC | 4800 GB/s | Tensor Core accelerator 3,958 FP8 Tensor TFLOPS with sparsity |
Up to 700 W configurable TDP | 900 GB/s NVLink; PCIe Gen 5 | NVIDIA H200 specifications ↗ |
Original references
Materials and manufacturing sources
These links identify the original institution or company record used during the provisional hardware research.
Material-system source
Silicon
Material-system source
Copper metals
Material-system source
Gallium germanium
Material-system source
Rare earths
Material-system source
Process inputs
Company project source
TSMC Arizona
Company project source
Intel Ohio One
Company project source
Micron Central New York
Company project source
SK hynix West Lafayette
Visual explainers are being rebuilt—not abandoned
The former AI-assisted lessons, model-fit conclusions, material narratives, fab lessons, and retooling guidance are not published as finished TriWei analysis. Their intended replacements and acceptance gates are listed in the Hardware Explainers workbench entry →