បន្ទះឈីបនិងបច្ចេកវិទ្យាស៊ីជម្រៅ · ផ្លូវការ
NVIDIA Explains Five Co-Design Methods to Get More Value From Memory in AI Factories
NVIDIA Technical blog ·
មើលត្រួសៗ
In data centers that run AI services, how usefully companies use high-bandwidth memory (HBM, fast memory attached next to the GPU) has become important, because it is fast but expensive and scarce. NVIDIA says it is solving this problem by designing chips, software, CPUs, and storage together.
ចំណុចសំខាន់ៗ
- NVFP4 is a format that reduces numbers to 4 bits, which greatly shrinks the memory a model takes up.
- When expert models are spread across several GPUs, each GPU has more memory left over for cache.
- The Dynamo software splits conversation memory (the KV cache) across GPUs, CPUs, and storage, so multiple servers can share it.
- The large memory of the Vera CPU temporarily holds conversation memory for tasks waiting on tool calls.
ហេតុអ្វីវាសំខាន់
In an era when AI works in multiple steps, how much work can be done with the same memory determines the cost and speed of AI services.
ប្រភពដើម
Five Ways NVIDIA Co-Design Maximizes the Value of Every Byte in the AI Factory
NVIDIA Technical blog · Nov 6, 01:00 KST
Read on NVIDIA Technical blog ↗AI summarized the original article. Check the original for accuracy. Copyright belongs to NVIDIA Technical blog.
បន្ថែមពីវិស័យនេះ
- Daniel Ek Backs Europa Fusion, a Stealthy UK Fusion Startup Led by an AI Researcher
Spotify founder Daniel Ek is backing Europa Fusion, a stealth startup in the UK. Its founder, Nantas Nardelli, is an AI researcher who left Pacific Fusion to pursue fusion power with a different approach. The startup plans to focus more on simulations and data than on physical hardware.
TechCrunch · Nov 21, 04:53 KST - Nvidia Board Ousted Jensen Huang Mainly Over Lying, Sources Say
Nvidia's board reportedly pushed out CEO Jensen Huang mainly over his repeated lying, not just the affair allegations. Sources say he lied about his relationships and what he told the board over about a year.
WIRED · Nov 21, 04:52 KST - AWS CloudFormation StackSets Adds Custom Execution Roles and Nested Stack Support
AWS CloudFormation StackSets now lets service-managed stack sets use custom execution roles. This lets you limit what StackSets can do in each account. Nested stacks, macros, and transforms also work with these stack sets.
AWS What's New · Nov 21, 02:45 KST - AWS Transform Adds UI Component Modernization for ASP.NET Websites
AWS Transform, Amazon's tool for upgrading old software, can now move the visual parts of legacy ASP.NET websites to new frameworks. When it finds paid third-party UI parts, it can swap them for free Microsoft or open-source ones, or keep the vendor's library with guidance on how to license it.
AWS What's New · Nov 21, 01:30 KST