MasInt
is a new computer designed to run emerging supermassive superintelligences
MasInt is founded to build supercomputers that accelerate emerging superintelligent AIs. MasInt founders are former NVIDIA employees who worked on CUDA and Cutlass. We believe that it is possible for AIs to learn faster than scaling laws by taking advantage of emerging memory technologies. MasInt phase 1 will build a prototype of a computer combining GPUs with high capacity and high bandwidth LPDDR and 3D-NAND memories to show that it is possible to significantly increase accuracy for complex engineering applications including operating system driver development, relational data analytics, and machine learning systems.
Posts
-
LLM Decompression: Reverse-Engineering Models Into Datasets
-
ScalarLM Benchmarking MI300X BF16 GEMM
-
ScalarLM Benchmarking MI300X Memcpy Peer
-
ScalarLM Benchmarking MI300X Memcpy
-
Tokenformer: A Scalable Transformer Architecture
-
Introducing ScalarLM v0.5: Unifying LLM Inference and Training for RL Agents
-
Building ROCm Containers for ScalarLM: A Comprehensive Guide
-
Large Language Models for Automatic Code Repair
-
Welcome to my blog
subscribe via RSS