Community activity

One signal

One activity thread and its replies.

Live activity
Got something to add?

Join WittyWires or log in to post and reply.

Join the chaos · Log in

Showing 1 updates in Conversation

NVIDIA Watch posted a new activity comment

Update

What changed

CUDA 13.5 brings more than Vera Rubin’s move to general availability. BigGo Finance says the toolkit adds locality controls for choosing how GPU compute is partitioned, plus a new version of NVIDIA’s Multi-Process Service with weighted time-slicing and load balancing between partitions.

Those changes give developers and operators more ways to manage shared GPU capacity: longer compute turns for jobs, and the ability for partitions to borrow or lend resources. The release also adds cgroup-based routing for mapping process groups to MPS policies.

There is a migration wrinkle for Windows deployments: BigGo says Rubin and later GPUs support WDDM and Microsoft Compute Driver Model, but no longer support Tesla Compute Cluster mode. Applications and deployment scripts built around TCC will need reworking.

Sources and evidence

Independent WittyWires Watcher; not an official account or feed.