Positron Atlas - FPGA Inference Server
The Positron Atlas is an 8-card FPGA inference server delivering 4.5x better performance per watt than the NVIDIA DGX H200 at 2000W in a single 1U chassis.
Hardware
AMD Instinct MI325X - 256GB CDNA3 for Inference
AMD Instinct MI325X specs, benchmarks, and analysis. 256GB HBM3e at 6 TB/s, 2.6 PFLOPS FP8, CDNA3 architecture - the memory-capacity upgrade to the MI300X targeting large model inference.
Huawei Atlas 350 - China's FP4 Inference Accelerator
Huawei Atlas 350 specs, benchmarks, and analysis. Ascend 950PR chip, 112GB HiBL 1.0 HBM, 1.56 PFLOPS FP4, 600W - China's first domestically developed FP4-capable AI accelerator.
Microsoft Maia 200 - Azure's Inference Accelerator
Microsoft Maia 200 specs, benchmarks, and architecture analysis. TSMC 3nm, 216GB HBM3e, 10 PFLOPS FP4, 750W - Microsoft's first inference-only silicon deployed in Azure.
AMD Instinct MI455X
AMD's flagship CDNA 4 AI GPU with 432 GB HBM4, 40 PFLOPS FP4, and 2nm chiplet design targeting H2 2026.
Apple M5 Max
Apple's flagship SoC with 40-core GPU, per-core Neural Accelerators, 614 GB/s bandwidth, and 4x AI performance over M4 Max.
Meta MTIA 300
Meta's first mass-deployed RISC-V AI accelerator - 1.2 PFLOPS FP8, 216 GB HBM, powering Facebook and Instagram at scale.
NVIDIA Vera Rubin NVL144
NVIDIA's Rubin-based rack system with 144 R200 GPUs, 3.6 ExaFLOPS FP4, 20 TB HBM4 - arriving H2 2026.
AWS Trainium3 - Amazon's 3nm AI Accelerator
Complete specs, benchmarks, and analysis of AWS Trainium3 - Amazon's TSMC 3nm AI chip with 2.52 PFLOPS FP8, 144GB HBM3e, and NeuronLink-v4, powering Anthropic's Claude through Project Rainier.
Etched Sohu - Transformer-Only Inference ASIC
Full specs and critical analysis of the Etched Sohu - a transformer-specific ASIC claiming 500K+ tokens/sec on Llama 70B, built on TSMC 4nm with 144GB HBM3E. Bold claims, but no independent benchmarks yet.
Hailo-10H - Edge AI With On-Device LLMs
Complete specs, benchmarks, and analysis of the Hailo-10H - a 2.5W edge AI accelerator with 40 TOPS INT4, on-module LPDDR4, and the ability to run LLMs and VLMs on a Raspberry Pi at 10 tokens per second.
NVIDIA Rubin CPX - Inference GPU With GDDR7
Full specs, benchmarks, and analysis of the NVIDIA Rubin CPX - a purpose-built inference GPU with 128GB GDDR7, 30 PFLOPS NVFP4, and 3x faster attention versus Blackwell, targeting million-token context workloads.
