RISC-V And GPU Synergy In Practice: A Path Toward High-Performance SoCs


With the rapid growth of edge AI and high-performance computing demands, the division of roles within the processor industry is beginning to shift. Recent moves from the dominant CPU IP supplier has increased industry attention on the openness and ecosystem neutrality of the supply chain. Against this backdrop, the value of RISC-V CPU IP is becoming more evident. It offers chipmakers greater... » read more

Designing GPUs For Developers: A Conversation With Godot


Godot has rapidly established itself as one of the most important graphics engines in today’s ecosystem. Free, open source, and increasingly capable across 2D, 3D, mobile, desktop, and beyond, it represents a philosophy that resonates strongly with modern developers. In this conversation, Clay John—who leads Godot’s rendering team—shares how Godot thinks about performance, iteration,... » read more

Power, Not Area: Why Edge GPU Design Is Entering A New Era


For decades, semiconductor progress followed a familiar playbook: shrink the node, pack in more logic, raise the clock, and performance would follow. That model held remarkably well, and possibly much longer than it should have. As the industry moves below 2nm, GPU design is running into a hard physical reality. The limiting factor is no longer how much logic we can fit on a die. It’s how ... » read more

Modern Trends In Floating-Point


The requirement to support real numbers in computers has existed for as long as computers themselves, yet has always been a more complicated challenge than it at first appears. Why? Because computer-based representations can only represent a finite subset of the continuum of real numbers. Consequently, they can only ever be considered an approximation – thereby demanding a diligent understand... » read more

All Models Are Wrong, But Some Are Useful: Lessons From Everyday Life


Have you ever checked a weather forecast, packed an umbrella, and then spent the day under clear blue skies? Or trusted your navigation app to save time, only to end up stuck behind a tractor? These moments are frustrating—but they illustrate a fundamental truth: All models are wrong, but some are useful. This principle, coined by statistician George Box, applies everywhere—from predi... » read more

Building Custom Graphics Cards For Cloud Gaming


The global cloud gaming market will reach over $20B by 2030, with Asia Pacific representing 45% of the opportunity according to Grandview Research. However, incumbent GPU solutions were designed for data center compute, not the unique economics of cloud gaming, where profitability depends on maximizing concurrent users per GPU while maintaining a premium user experience. For companies devel... » read more

GPU Enables Surround View In Automotive Domain Controller


In recent years, the capabilities of Advanced Driver Assistance Systems (ADAS) have flourished. Nearly half of all car sales in the USA offer Level 2 capabilities (such as lane keeping and adaptive cruise control) or higher, and China is pushing the market further towards Level 3 (conditional automation with driver oversight) and beyond. The advanced functionality offered by these ADAS and a... » read more

GPU Driver Update Adds Support For Additional Vulkan And OpenCL Extensions


Here are some of the highlights of what has been updated in the latest Imagination GPU Linux and Android Driver Development Kits: Leveraging cooperative matrix in Vulkan To help accelerate graphics post-processing, neural shaders, physics simulations, and machine learning inference on the GPU, DDK 25.2 implements support for VK_KHR_cooperative_matrix. This extension provides Vulkan developers... » read more

How Fast A GPU Do You Need For Your User Interface?


Graphical user interfaces (GUIs) are common across all walks of life, from your smartphones to your TVs and even in your cars. Over the past decade, their complexity has evolved, moving from a simple background with basic icons into beautiful device differentiators, with 3D elements and micro-interactions that enhance the experience: shifting the perception when a phone tilts, or providing an... » read more

Shrinking LLMs With Self-Compression


Language models are becoming ever larger, making on-device inference slow and energy-intensive. A direct and surprisingly effective remedy is to prune complete channels whose contribution to the task is negligible. Our earlier work introduced a training-time procedure – Self-Compression [1, 4] – that lets back-propagation decide the bit-width of every channel, so unhelpful ones fade away. T... » read more

← Older posts