Nature

‘RAMmageddon’ hits labs: AI-driven memory shortage is impacting science

The soaring cost and limited supply of computer memory is slowing some projects — and spurring creative approaches.

Texture pop-in may be about to pop-out of existence thanks to even faster asset streaming ushered in by new Zstandard support in DirectStorage

According to Microsoft's own benchmarks, GDeflate is still the way to go for developers wanting to let the GPU handle asset loading, but Zstd is better for CPU-intensive projects. Zstd does also offer ...

The memory stock cycle of boom-bust-repeat is over, executives say

For decades, memory stocks were stuck in a trader's game, a boom-bust-repeat pattern, but executives now say that AI has structurally broken the old cycle, and prices are showing no signs of coming ...
IEEE

Efficient KV Cache Spillover Management on Memory-Constrained GPU for LLM Inference

Abstract: The rapid growth of model parameters presents a significant challenge when deploying large generative models on GPU. Existing LLM runtime memory management solutions tend to maximize batch ...
IEEE

Effective Memory Management and Sparse Aware Cache for Row-Wise Sparse CNNs

Abstract: Convolutional Neural Networks (CNNs) are used in several image processing tasks like image recognition and object localization. For edge applications such as drones and autonomous vehicles, ...