Designing specifically for inference
d-Matrix develops computing hardware for running trained AI models. Founded in 2019, it combines digital in-memory computing with chiplets, packaging and interconnects. The architecture separates fast integrated memory from larger capacity memory, addressing data movement as well as arithmetic. [1]
Corsair reaches early-access customers
The November 2024 Corsair disclosure described PCIe inference cards and the Aviator software stack. Corsair was sampling to early-access customers, while broad availability was then planned for the following year. Sampling establishes a real hardware stage without implying a completed volume rollout. [1]
Measured comparisons need their conditions
The release’s throughput, power and cost estimates were explicitly preliminary. They describe the company’s chosen configurations, rather than proving that Corsair replaces GPUs for training or wins every inference workload. Its relevance is an alternative memory-and-compute design for serving models. [1]