Shrinking LLMs with Self-Compression
Discover how Self-Compression effectively reduces language model sizes, enhancing efficiency for on-device inference while maintaining predictive quality, ideal for resource-limited settings.
Vendor:
Imagination Technologies Group Ltd.
Category:
NPU
PowerVR Neural Network Accelerator
The Series3NX-F brings programmable extensibility to the Series3NX architecture.
Overview
The Series3NX-F brings programmable extensibility to the Series3NX architecture. It combines a Series3NX core with a neural network programmable unit (NNPU); a highly neural network optimised GPGPU, based on our industry-proven Rogue architecture. This provides developers with even greater flexibility in optimising their applications to run on the Series3NX architecture.
Key features
- Flexible bit-depth data type support
- As a fully flexible solution, the Series3NX supports neural network bit depths from 16 down to 4-bit, reducing bandwidth and increasing performance without compromising inference accuracy.
- Lossless weight compression
- Complementing its low-bit depth support, the Series3NX introduces a new lossless weight compression scheme that reduces network model sizes and bandwidth thus increasing overall performance.
- Advanced security enablement
- PowerVR Series3NX integrates with the industry-leading security architectures including a flexible infrastructure that enables integration into custom solutions, enabling rights holders to protect their content where required.
- Leading performance with low power consumption
- With the industry’s highest inference/mW, the Series3NX delivers class-leading neural network acceleration with the lowest power consumption.
Applications
- Security and surveillance
- Automotive (ADAS)
- Augmented and Virtual reality
Specifications
Identity
Part Number
PowerVR Series3NX
Vendor
Imagination Technologies Group Ltd.
Type
Silicon IP
Files
Note: some files may require an NDA depending on provider policy.
Provider
Learn more about NPU IP core
Embarrassingly Parallel Problems: Definitions, Challenges and Solutions
Benchmarking an NPU at Scale
With every SDK release, Quadric automatically recompiles and re-profiles the entire model zoo across a broad sweep of hardware configurations. The results land in DevStudio for any SoC architect to explore â no sales call, no handpicked numbers.
Your NPU Learned to Listen
Whisper runs end-to-end on Chimera. Encoder and decoder compiled as native GPNPU kernels, INT4 weights, FP16 attention, top-1 token match against the float32 reference. Scales to four cores with a flag, no recompile.
Can Your NPU Run DOOM? Chimera Can.
Is your NPU DOOMed? Quadric's Chimera GPNPU runs every AI model â and a complete DOOM engine. Find out why Quadric is different.
Heterogeneous NPU Data Movement Tax: Intel's Own Slides Tell the Story
At Quadric, we have long argued that heterogeneous NPU designs â those that stitch together multiple specialized fixed-function engines â carry an unavoidable hidden cost: data has to move. A lot. And data movement burns power, adds latency, and creates silicon-area overhead that scales with every new generation of AI models. Now, Intel has made that case for us.
Frequently asked questions about NPU IP cores
What is PowerVR Neural Network Accelerator?
PowerVR Neural Network Accelerator is a NPU IP core from Imagination Technologies Group Ltd. listed on Semi IP Hub.
How should engineers evaluate this NPU?
Engineers should review the overview, key features, supported foundries and nodes, maturity, deliverables, and provider information before shortlisting this NPU IP.
Can this semiconductor IP be compared with similar products?
Yes. Buyers can compare this product with similar semiconductor IP cores or IP families based on category, provider, process options, and structured technical specifications.