News: 2026-04-03
April 03, 2026 · Generated 09:31 AM PT
Executive Summary
- Legacy APUs Get Modern Support: AMD is standardizing the Linux experience for decade-old GCN 1.1 (Sea Islands) APUs by shifting them to the modern AMDGPU driver by default in Linux 7.1/7.2, bringing out-of-the-box Vulkan (RADV) and more stable OpenGL.
- Advanced CPU Power Tuning: The AMD P-State driver in Linux 7.1 is introducing granular power management controls, including CPPC Performance Priority and Raw EPP tuning, laying crucial software groundwork for upcoming Zen 6 processors.
- Qualcomm Heats Up Mobile Linux Market: Qualcomm’s MSM DRM driver updates in Linux 7.1 show rapid maturation of the Snapdragon X2 platform and Adreno GPUs, signaling stronger competition for AMD in the ARM-based Linux laptop segment.
🤖 ROCm Updates & Software
[2026-04-03] AMDGPU Driver Ready To Be The Default For Aging Kaveri / Kabini / Mullins APUs
Source: Phoronix (AMD Linux)
Key takeaway relevant to AMD:
- Transitioning legacy Sea Islands APUs from the experimental phase to the modern AMDGPU driver by default extends the usable lifecycle of these older chips, ensuring AMD developers and end-users have a standardized, highly-performant Linux stack with modern API support (Vulkan/RADV).
Summary:
- A new patch led by Valve’s Linux graphics team seeks to make the modern AMDGPU driver the default for decade-old AMD GCN 1.1 (Sea Islands/CIK) APUs, replacing the legacy Radeon Linux driver.
Details:
- Target Hardware: AMD GCN 1.1 Sea Islands (CIK) APUs, specifically Kaveri, Kabini, and Mullins architectures.
- Previous State: While Linux 6.19 made AMDGPU the default for GCN 1.1 dGPUs, the APUs were kept on the older Radeon driver. AMDGPU was available but remained an experimental, non-default option.
- Feature Parity Achieved: Developer Timur Kristóf (Valve) closed gaps in display code, adding Display Core (DC) support for analog connectors and DisplayPort bridge encoders to match the old driver’s capabilities.
- Benefits of AMDGPU:
- More stable OpenGL support through RadeonSI.
- Out-of-the-box Vulkan support via RADV.
- Improved general performance.
- Better display features through DC.
- Fallback Option: Users wishing to retain the legacy driver can append
amdgpu.cik_support=0 radeon.cik_support=1to their boot parameters. - Timeline: The patch was submitted targeting the Linux v7.1 merge window, though it may be delayed to v7.2 due to DRM-Next feature cut-offs.
[2026-04-03] AMD P-State Driver Introducing New Features With Linux 7.1
Source: Phoronix (AMD Linux)
Key takeaway relevant to AMD:
- AMD is significantly upgrading its power and thermal management capabilities for Ryzen and EPYC processors on Linux. These deep-level tuning options will allow developers and enterprise admins to squeeze maximum efficiency out of existing hardware while laying the foundation for next-generation Zen 6 chips.
Summary:
- The AMD P-State driver is receiving major feature updates in the Linux 7.1 kernel, including CPPC Performance Priority, Dynamic EPP, and Raw EPP controls for enhanced CPU frequency scaling.
Details:
- Maintainer Update: Mario Limonciello of AMD has officially taken over as the formal AMD P-State driver maintainer.
- AMD CPPC Performance Priority: Enables user-space to assign specific floor performance levels to individual CPU cores. Platform firmware uses these values for throttling decisions during power/thermal constraints. This exposes new
floor_freqandfloor_countsysfs attributes. - Dynamic Energy Performance Preference (EPP): Automatically adjusts the EPP profile based on whether the system is on AC or DC power (triggered by plug-in/plug-out events). This feature is toggleable via Linux kernel Kconfig or boot time options.
- Raw EPP: Bypasses the limitation of using only four static preset EPP options. It allows administrators to write any integer value from 0 to 255 directly into the sysfs file for highly granular workload tuning.
- Bug Fixes: Includes vital fixes for real-time
PREEMPT_RTkernel builds. - Hardware Implications: While functional on modern Ryzen and EPYC processors, these features are specifically noted as preparations for the upcoming AMD Zen 6 architectures.
🤼♂️ Market & Competitors
[2026-04-03] Snapdragon X2’s Adreno X2-85 GPU Sees Driver Improvements For Linux 7.1
Source: Phoronix (AMD Linux)
Key takeaway relevant to AMD:
- Qualcomm is rapidly expanding and optimizing its open-source Linux graphics stack to support its Snapdragon X2 platform. AMD must monitor this closely, as the introduction of robust Vulkan ray tracing and advanced power saving on ARM-based Linux directly threatens AMD’s dominance in the Linux/x86 laptop and handheld markets.
Summary:
- A major pull request for the MSM DRM driver in Linux 7.1 brings extensive support, feature enablement, and power optimizations for Qualcomm’s Adreno X2-85 and 840 GPUs.
Details:
- Snapdragon X2 (Adreno X2-85) Updates: Enables GPU preemption support, introduces SKU detection via a Speedbin table for different X2-85 variants, and applies fixes for bogus protect errors.
- Adreno 840 Updates: Gains preemption support, SKU detection, and Inter Frame Power Collapse (IFPC) as a dedicated power-saving feature.
- Vulkan Ray Tracing Prep: Introduces Application Qrisc Engine (AQE) support, which is a mandatory requirement for supporting the Vulkan ray pipeline on Qualcomm hardware.
- Additional Hardware Enablement:
- DPU support enabled for the Eliza SoC.
- Patches sent separately to enable the Adreno 810 GPU for the Milos SoC.
- Display & Code Fixes: Includes reworked alpha handling and RGB101010 support within the DSI code.
📈 GitHub Stats
| Category | Repository | Total Stars | 1-Day | 7-Day | 30-Day |
|---|---|---|---|---|---|
| AMD Ecosystem | AMD-AGI/GEAK-agent | 82 | +1 | +1 | +13 |
| AMD Ecosystem | AMD-AGI/Primus | 83 | 0 | +1 | +8 |
| AMD Ecosystem | AMD-AGI/TraceLens | 66 | 0 | +1 | +5 |
| AMD Ecosystem | ROCm/MAD | 33 | 0 | 0 | +2 |
| AMD Ecosystem | ROCm/ROCm | 6,312 | +8 | +24 | +92 |
| Compilers | openxla/xla | 4,130 | +1 | +14 | +100 |
| Compilers | tile-ai/tilelang | 5,461 | +8 | +28 | +149 |
| Compilers | triton-lang/triton | 18,838 | +11 | +59 | +288 |
| Google / JAX | AI-Hypercomputer/JetStream | 418 | 0 | 0 | +4 |
| Google / JAX | AI-Hypercomputer/maxtext | 2,200 | +6 | +13 | +43 |
| Google / JAX | jax-ml/jax | 35,292 | +6 | +55 | +295 |
| HuggingFace | huggingface/transformers | 158,747 | +57 | +251 | +1405 |
| Inference Serving | alibaba/rtp-llm | 1,080 | +1 | +4 | +23 |
| Inference Serving | efeslab/Atom | 336 | 0 | 0 | 0 |
| Inference Serving | llm-d/llm-d | 2,902 | +17 | +80 | +336 |
| Inference Serving | sgl-project/sglang | 25,385 | +49 | +291 | +1310 |
| Inference Serving | vllm-project/vllm | 75,151 | +148 | +629 | +3248 |
| Inference Serving | xdit-project/xDiT | 2,586 | +2 | +10 | +34 |
| NVIDIA | NVIDIA/Megatron-LM | 15,905 | +18 | +79 | +394 |
| NVIDIA | NVIDIA/TransformerEngine | 3,255 | +2 | +8 | +73 |
| NVIDIA | NVIDIA/apex | 8,939 | 0 | -1 | +11 |
| Optimization | deepseek-ai/DeepEP | 9,094 | +2 | +20 | +81 |
| Optimization | deepspeedai/DeepSpeed | 41,978 | +10 | +57 | +247 |
| Optimization | facebookresearch/xformers | 10,402 | +4 | +11 | +46 |
| PyTorch & Meta | meta-pytorch/monarch | 1,003 | 0 | +2 | +18 |
| PyTorch & Meta | meta-pytorch/torchcomms | 355 | 0 | +4 | +11 |
| PyTorch & Meta | meta-pytorch/torchforge | 665 | +1 | +5 | +37 |
| PyTorch & Meta | pytorch/FBGEMM | 1,549 | 0 | +1 | +12 |
| PyTorch & Meta | pytorch/ao | 2,757 | +3 | +11 | +44 |
| PyTorch & Meta | pytorch/audio | 2,857 | 0 | +5 | +23 |
| PyTorch & Meta | pytorch/pytorch | 98,780 | +28 | +156 | +851 |
| PyTorch & Meta | pytorch/torchtitan | 5,207 | +3 | +17 | +100 |
| PyTorch & Meta | pytorch/vision | 17,612 | +1 | +21 | +68 |
| RL & Post-Training | THUDM/slime | 5,107 | +26 | +103 | +552 |
| RL & Post-Training | radixark/miles | 1,042 | +8 | +21 | +100 |
| RL & Post-Training | volcengine/verl | 20,424 | +32 | +164 | +832 |