DeepSeek's DSpark just made Nvidia's most important new bet harder to close

DeepSeek just released DSpark, an inference module that makes its AI models 60% to 85% faster without new hardware. Nvidia is ...

DeepSeek open sources DSpark, a new framework to speed up LLM inference by up to 85%

DSpark can make decoding faster, but acceptance quality still determines how much speed the system actually realizes.
VentureBeat

IndexCache, a new sparse attention optimizer, delivers 1.82x faster inference on long-context AI models

Processing 200,000 tokens through a large language model is expensive and slow: the longer the context, the faster the costs spiral. Researchers at Tsinghua University and Z.ai have built a technique ...
Semiconductor Engineering

AI Inference Needs A Mix-And-Match Memory Strategy

Interactive LLMs (chat, copilots, agents) with strict latency targets Long‑context reasoning (codebases, research, video) with massive KV (key value) cache footprints Ranking and recommendation models ...
Developer Tech

NVIDIA: DFlash block diffusion accelerates autoregressive LLMs

Deploying DFlash block diffusion on NVIDIA hardware accelerates autoregressive LLMs during latency-sensitive inference.
Morningstar

SambaNova and Intel Announce Blueprint for Heterogeneous Inference: GPUs for Prefill, SambaNova RDUs for Decode, and Intel® Xeon® 6 CPUs for Agentic Tools

Coding agents are exposing the limits of GPU-only infrastructure, making each phase of the pipeline mission-critical: efficient prefill, high-throughput decoding, and high-performance agent task ...
Semiconductor Engineering

Transformers At The Edge: Efficient LLM Deployment

Since the groundbreaking 2017 publication of “Attention Is All You Need,” the transformer architecture has fundamentally reshaped artificial intelligence research and development. This innovation laid ...