AI Compute Card Encyclopedia
AI Compute Card Encyclopedia
A comprehensive catalog of 160+ AI compute cards covering training, inference, and edge computing across all scenarios. Spanning 14 vendors including NVIDIA, AMD, Intel, Google, AWS, Huawei, Cambricon, Cerebras, and more.
Quick Start
| Feature | Description | Link |
|---|---|---|
| Full Comparison Table | Side-by-side comparison of 160+ compute cards with filtering | View Table |
| Future Roadmap | 2025–2028 product launch timeline | View Roadmap |
| TCO Calculator | Total cost of ownership estimation (power/PUE/NPV) | Use Calculator |
| Industry News | Latest releases, acquisitions, IPOs, and in-depth analysis | Browse Blog |
Browse by Type
| Type | Description | Representative Products |
|---|---|---|
| Datacenter Training GPU | Primary LLM training hardware | H100 / H200 / B200 / MI300X |
| Datacenter Inference GPU | High-throughput online inference | L4 / L40S / RTX 6000 Ada |
| Training ASIC | Dedicated training accelerators | TPU v6e / Trainium 3 / Ascend 950 |
| Inference ASIC | Dedicated inference accelerators | TPU 8i / Inferentia 2 / Ascend 310 |
| Wafer-Scale Training | Single-wafer solutions for massive models | Cerebras WSE-3 / WSE-4 |
| Workstation GPU | Professional graphics + AI inference | RTX 6000 Ada / RTX Pro 6000 |
| Consumer GPU | Personal AI development & inference | RTX 5090 / RTX 5080 |
| Edge AI | On-device inference acceleration | Jetson Orin / Edge TPU |
| Innovative Architectures | Non-traditional architecture exploration | Groq LPU / Graphcore IPU |
Browse by Vendor
| Vendor | Flagship Product Line | Representative Models |
|---|---|---|
| NVIDIA | General-purpose GPU + CUDA ecosystem | H100 / B200 / Rubin R200 |
| AMD | Instinct series + ROCm | MI300X / MI400 / MI455X |
| Intel (Habana) | Gaudi series + OpenVINO | Gaudi 3 / |
| TPU tensor processors | TPU v6e / TPU 8t / TPU 8i | |
| AWS | Trainium / Inferentia | Trainium 3 |
| Cerebras | Wafer-Scale Engine (WSE) | WSE-3 / WSE-4 🔮 |
| Huawei | Ascend series (domestic China) | Ascend 910C / 920 / 950DT |
| Cambricon | Siyuan MLU series | MLU370 / MLU690 |
| Meta | MTIA inference accelerator | MTIA v3 |
| Microsoft | Maia inference accelerator | Maia 200 |
| Baidu | Kunlun XPU | Kunlun R200 |
| Moore Threads | MTT series (domestic GPU) | MTT S5000 |
| MetaX | Xiyun series (domestic GPU) | C600 |
| Iluvatar CoreX | Tian-gai series (domestic GPU) | BI-V150 |
| Enflame | T-series (domestic DTU) | T21 |
Status: ✅ Released | 🔄 Upcoming | 🔮 Forward-looking | ❌ Cancelled
Latest Releases
- 2026-07-20 — WAIC 2026: Huawei Atlas 950 SuperPoD live demo, 1024 cards / 1 EFLOPS FP8, wins SAIL Award ✅
- 2026-07-11 — AMD Advancing AI 2026: MI400 three SKUs confirmed, Helios rack 225-245kW, OpenAI+Meta lock 12GW ✅
- 2026-07-11 — Moore Threads MTT S5000 launched, domestic GPU targeting H100, ¥55,000 ✅
- 2026-07-10 — NVIDIA Vera Rubin starts shipping, first VR200 NVL72 delivered, Samsung HBM4 in production ✅
- 2026-07-10 — Ascend 950PR monthly capacity jumps 10×, ByteDance locks 350K units at $5.6B 🔄
- 2026-04 — Google TPU 8t (training) + 8i (inference) split architecture officially announced ✅
- 2025-11 — Huawei Ascend 920 launched: 1,800 TFLOPS FP16, 4 Tbps inter-chip interconnect ✅
Why MirrorFrog?
- Comprehensive — 160+ compute cards, 14 vendors, complete specifications
- Continuously Updated — Keeping pace with vendor release cadence, 2025–2026 launches updated promptly
- Neutral Comparison — No vendor bias, purely data-driven
- Open Source — Open source on GitHub, PRs welcome
- TCO Tool — Built-in Total Cost of Ownership calculator with power/PUE/NPV estimation
Contribute
Found an error or want to add a new card? Open an Issue or submit a PR!
Last updated: 2026-07-22 | ← Home | Comparison Table → | Roadmap → | TCO Calculator → | Blog →