Snapdragon X2 Elite users reported an 87% decrease in AI performance after the driver update

Many Snapdragon X2 Elite chip users reported a serious performance drop after installing Adreno driver version 32.0.181.0. The reports focus on two main tasks including video editing and running AI models locally on the computer.

One Zenbook A16 user reported that rendering time for a 7-minute 1080p video in Adobe Premiere Pro increased from 1 minute 40 seconds to about 10 minutes. The slowdown is equivalent to 6 times compared to before updating the driver. This user recommends that the community not upgrade to the new version.

Another case on the Surface Laptop 8th Edition recorded a decrease in text creation performance of the large language model from 69% to 87%. When testing Qwen3 models, the inference speed drops sharply. Qwen3-0.6B dropped from 98.1 to 12.8 tokens per second. Qwen3-4B reduced from 29.5 to 9.0 tokens per second. Qwen3-Coder-30B-A3B-Instruct model reduced from 26.8 to 5.9 tokens per second.

The user confirms that the system, software and operating system have not changed. Performance returned to normal when reverting to old driver version 32.0.172.2. The drop mainly occurred on the OpenCL backend of llama.cpp. CPU and NPU performance remained virtually unchanged. The issue has been reported to Qualcomm's support channel pending a patch.

Snapdragon X2 Elite users reported an 87% decrease in AI performance after the driver update Many Snapdragon X2 Elite chip users reported a serious performance drop after installing Adreno driver version 32.0.181.0. The reports focus on two main tasks including video editing and running AI models locally on the computer. One Zenbook A16 user reported that rendering time for a 7-minute 1080p video in Adobe Premiere Pro increased from 1 minute 40 seconds to about 10 minutes. The slowdown is equivalent to 6 times compared to before updating the driver. This user recommends that the community not upgrade to the new version. Another case on the Surface Laptop 8th Edition recorded a decrease in text creation performance of the large language model from 69% to 87%. When testing Qwen3 models, the inference speed drops sharply. Qwen3-0.6B dropped from 98.1 to 12.8 tokens per second. Qwen3-4B reduced from 29.5 to 9.0 tokens per second. Qwen3-Coder-30B-A3B-Instruct model reduced from 26.8 to 5.9 tokens per second. The user confirms that the system, software and operating system have not changed. Performance returned to normal when reverting to old driver version 32.0.172.2. The drop mainly occurred on the OpenCL backend of llama.cpp. CPU and NPU performance remained virtually unchanged. The issue has been reported to Qualcomm's support channel pending a patch.

Screen height is not available for this content. Please rotate your device screen vertically.

Snapdragon X2 Elite users reported an 87% decrease in AI performance after the driver update

Many Snapdragon X2 Elite chip users reported a serious performance drop after installing Adreno driver version 32.0.181.0. The reports focus on two main tasks including video editing and running AI models locally on the computer.

One Zenbook A16 user reported that rendering time for a 7-minute 1080p video in Adobe Premiere Pro increased from 1 minute 40 seconds to about 10 minutes. The slowdown is equivalent to 6 times compared to before updating the driver. This user recommends that the community not upgrade to the new version.

Another case on the Surface Laptop 8th Edition recorded a decrease in text creation performance of the large language model from 69% to 87%. When testing Qwen3 models, the inference speed drops sharply. Qwen3-0.6B dropped from 98.1 to 12.8 tokens per second. Qwen3-4B reduced from 29.5 to 9.0 tokens per second. Qwen3-Coder-30B-A3B-Instruct model reduced from 26.8 to 5.9 tokens per second.

The user confirms that the system, software and operating system have not changed. Performance returned to normal when reverting to old driver version 32.0.172.2. The drop mainly occurred on the OpenCL backend of llama.cpp. CPU and NPU performance remained virtually unchanged. The issue has been reported to Qualcomm's support channel pending a patch.