Parasail adds Positron Titan inference server

x.com1 min read

Parasail has announced that it will be the first inference cloud to deploy Positron's next-generation Titan servers at scale, starting with 5MW. The Titan servers utilize smartphone-style memory instead of scarce high-bandwidth memory to run multi-trillion-parameter models on commodity DRAM. This hardware design allows a single server to hold a model of up to 32 trillion parameters. Open article →

More on this topic