2025-01-05

mug recall, Wells was a jerk, crowd surfing fail, boom tubes, fake AI frens, wrong audiences, advertising on news, Biden medal-ing, bomber manifestos, French wine decline, breast enhancement shots?


celebrity gossip


Edumacationalizing / Acedemia Nuts

Info Rental / ShowBiz / Advertising

TechSuck / Geek Bait

  • Modern Generic SVGA driver for Windows 3.1

  • Nascar and its teams are embracing 3D printing

  • How to use distributed shared memory in CUDA for inter-thread-block communication

    what the fine people at NVIDIA did with their Hopper Architecture is add a new address-space to the memory-model, which let’s us access a different’s thread-block memory without going to main memory. This is implemented via some inter-streaming-multiprocessor communication-net the details of which seem to still be obscured, but that also doesn’t matter too much for us. It is to note though that with the current architecture CUDA only supports this for thread-blocks that are in the same thread-block-cluster (clusters are limited to a total size of 16 blocks).

    you won’t be able to use this feature if you’re not running on a Compute-Capability 9.0 (or higher if you’re reading post-Blackwell release) CUDA-Card (i.e H100/H200), this also means that we need to compile our CUDA kernel to be specific to the architecture

Democrats / Biden Inc

Left Angst

Law Breaking / Police / Internal Security

World

Health / Medicine