Cambricon MLU590 - China's Inference Accelerator
Full specs and analysis of the Cambricon MLU590 - 192GB HBM2e, ~2,400 GB/s bandwidth, TSMC 7nm, and what it means for AI inference outside the NVIDIA ecosystem.
Hardware
Cerebras WSE-3 - The Wafer-Scale AI Engine
The Cerebras Wafer-Scale Engine 3 is the largest chip ever built - an entire TSMC 5nm wafer with 900,000 AI cores, 44GB of on-chip SRAM, and 21 PB/s of memory bandwidth powering the CS-3 AI supercomputer.
Google TPU v6e Trillium
Google Cloud TPU v6e Trillium specs, benchmarks, and pricing. 32GB HBM per chip, ~1,600 GB/s bandwidth, optimized for Transformer training and inference at cloud scale.
Google TPU v7 Ironwood
Google TPU v7 Ironwood specs, architecture, and performance estimates. Google's next-gen inference-optimized TPU with massive memory per chip, announced at Cloud Next 2025.
Groq LPU - Deterministic Inference at Scale
Groq's Language Processing Unit (LPU) is a purpose-built inference ASIC that trades HBM for 230MB of on-chip SRAM, delivering deterministic latency and record-breaking tokens-per-second for LLM serving.
Huawei Ascend 910B
Huawei Ascend 910B specs, benchmarks, and real-world performance. 64GB HBM2e, ~1,200 GB/s bandwidth, ~600 TFLOPS FP16 - the chip that trained DeepSeek.
Huawei Ascend 910C
Huawei Ascend 910C specs, benchmarks, and performance analysis. 96GB HBM2e, ~1,800 GB/s bandwidth, ~800 TFLOPS FP16 - China's flagship AI chip under US sanctions.
Intel Gaudi 3 - Challenging NVIDIA on Price
Intel Gaudi 3 is a TSMC 5nm AI accelerator with 128GB HBM2e and 1,835 TFLOPS FP8 performance, positioned as a cost-effective alternative to NVIDIA H100 for training and inference workloads.
NVIDIA A100 - The GPU That Built Modern AI
Complete specs, benchmarks, and analysis of the NVIDIA A100 80GB SXM - the Ampere-architecture GPU that remains the most widely deployed AI accelerator in the world.
NVIDIA B200 - Blackwell Flagship GPU
Complete specs, benchmarks, and analysis of the NVIDIA B200 - the Blackwell-architecture flagship GPU with 192GB HBM3e, 8 TB/s bandwidth, and up to 9,000 TFLOPS FP8.
NVIDIA GB200 NVL72 - Rack-Scale Blackwell
Complete specs, benchmarks, and analysis of the NVIDIA GB200 NVL72 - the 72-GPU rack-scale Blackwell system delivering 1,440 PFLOPS FP4 for trillion-parameter AI training and inference.
NVIDIA GB300 NVL72 - Blackwell Ultra Rack
Complete specs, benchmarks, and analysis of the NVIDIA GB300 NVL72 - the Blackwell Ultra rack-scale system with 288GB HBM3e per GPU, 1.5x more FP4 compute, and 2x attention performance over GB200.
