Android's Virtualization Framework seems to be getting GPU support.
"Enable GPU feature. crosvm in virt APEX now supports GPU feature. This is to support VMs with display."
AVF was introduced in Android 13 and supported on select Pixels but is now supported by a variety of non-Pixel phones running Android 14. It's still only accessible to system apps, though with root, you can theoretically run your own VM.
Developer Danny Lin famously managed to get Windows 11 on ARM booting in a VM on his Pixel 6 two years ago (wow time sure flies), but at the time, there was no graphics acceleration.
"Enable GPU feature. crosvm in virt APEX now supports GPU feature. This is to support VMs with display."
AVF was introduced in Android 13 and supported on select Pixels but is now supported by a variety of non-Pixel phones running Android 14. It's still only accessible to system apps, though with root, you can theoretically run your own VM.
Developer Danny Lin famously managed to get Windows 11 on ARM booting in a VM on his Pixel 6 two years ago (wow time sure flies), but at the time, there was no graphics acceleration.
❤54👍18🎉7🔥4👏1😢1
Android 14 quietly introduced two changes to move the needle forward on deprecating older apps:
- Deprecated ABI warning: When you launch any 32-bit app, you'll see a dialog warning you that "this app isn't compatible with the latest version of Android. Check for an update or contact the app's developer." This is part of Google's multi-year efforts to deprecate support for 32-bit builds of Android. (Patrons/Twitter subscribers have access to a story from me that goes into more detail on these efforts).
- Updated deprecated target SDK warning: When you launch an app that targets SDK version 27 or lower (Android 8.1 Oreo or older), you'll see a warning that says "this app was built for an older version of Android. It might not work properly and doesn't include the latest security and privacy protections. Check for an update, or contact the app's developer."
The deprecated target SDK warning isn't actually new like the deprecated ABI warning, as it previously existed in Android 13 but the limit was set to SDK 23. In Android 14, the threshold for the warning to appear was bumped to SDK 28. Meanwhile, Android 14 also introduced a change that blocks the installation of apps that target SDK 22 or lower, ie. apps that used to trigger the deprecated target SDK warning in Android 13 are now blocked entirely in Android 14.
- Deprecated ABI warning: When you launch any 32-bit app, you'll see a dialog warning you that "this app isn't compatible with the latest version of Android. Check for an update or contact the app's developer." This is part of Google's multi-year efforts to deprecate support for 32-bit builds of Android. (Patrons/Twitter subscribers have access to a story from me that goes into more detail on these efforts).
- Updated deprecated target SDK warning: When you launch an app that targets SDK version 27 or lower (Android 8.1 Oreo or older), you'll see a warning that says "this app was built for an older version of Android. It might not work properly and doesn't include the latest security and privacy protections. Check for an update, or contact the app's developer."
The deprecated target SDK warning isn't actually new like the deprecated ABI warning, as it previously existed in Android 13 but the limit was set to SDK 23. In Android 14, the threshold for the warning to appear was bumped to SDK 28. Meanwhile, Android 14 also introduced a change that blocks the installation of apps that target SDK 22 or lower, ie. apps that used to trigger the deprecated target SDK warning in Android 13 are now blocked entirely in Android 14.
👍55😢27😭4🔥2😡1
Google is preparing to fully remove support for non-A/B updates from Android, leaving A/B updates as the only officially supported OTA update mechanism moving forward.
However, this won't stop certain device makers from continuing to use non-A/B as they can either use their own implementation or just revert Google's changes. Google would need to actually mandate the use of A/B through the CDD or GMS licensing, which they've tried to do but ended up reneging.
A/B updates = Seamless Updates, in case you weren't aware. Most OEMs have switched over to using A/B updates, with one notable exception: Samsung. If you want to learn more about how Android's update schemes work, check out this article I wrote (though note that Google ended up not going forward with this requirement).
However, this won't stop certain device makers from continuing to use non-A/B as they can either use their own implementation or just revert Google's changes. Google would need to actually mandate the use of A/B through the CDD or GMS licensing, which they've tried to do but ended up reneging.
A/B updates = Seamless Updates, in case you weren't aware. Most OEMs have switched over to using A/B updates, with one notable exception: Samsung. If you want to learn more about how Android's update schemes work, check out this article I wrote (though note that Google ended up not going forward with this requirement).
👍87🤬15❤9🤔3👎2🤡2👏1
One UI 6.1 will roll out to the Galaxy S23 series, S23 FE, Z Fold 5, Z Flip 5, and Tab S9 series at the end of March, bringing several "Galaxy AI" features Samsung unveiled with the Galaxy S24 series!
Here are the features the update will add:
* Chat Assist: Adjust message tone and translate messages in 13 different languages
* Live Translate: Voice and text translations for phone calls (Samsung Phone)
* Interpreter: A split-screen interface that shows text translations for live conversations
* Circle to Search with Google
* Note Assist: Create formats, generate summaries, and translate notes (Samsung Notes)
* Browsing Assist: Generate comprehensive summaries of news articles (Samsung Internet)
* Transcript Assist: Transcribe meeting recordings and generate summaries and translations (Samsung Voice Recorder)
* Generative Edit: Resize, reposition, or realign objects in photos (Samsung Gallery)
* Edit Suggestion: Get suggestions to edit photos (Samsung Gallery)
* Instant Slow-mo: Generate additional frames for slow-motion videos (apparently not available on S23 FE)
* AI wallpapers: Generate custom wallpapers using AI
Here are the features the update will add:
* Chat Assist: Adjust message tone and translate messages in 13 different languages
* Live Translate: Voice and text translations for phone calls (Samsung Phone)
* Interpreter: A split-screen interface that shows text translations for live conversations
* Circle to Search with Google
* Note Assist: Create formats, generate summaries, and translate notes (Samsung Notes)
* Browsing Assist: Generate comprehensive summaries of news articles (Samsung Internet)
* Transcript Assist: Transcribe meeting recordings and generate summaries and translations (Samsung Voice Recorder)
* Generative Edit: Resize, reposition, or realign objects in photos (Samsung Gallery)
* Edit Suggestion: Get suggestions to edit photos (Samsung Gallery)
* Instant Slow-mo: Generate additional frames for slow-motion videos (apparently not available on S23 FE)
* AI wallpapers: Generate custom wallpapers using AI
🔥47👍14👎9🤔1
You'll soon be able to launch Google Keep from a lock screen shortcut, making it even easier to quickly jot down notes on your Android tablet.
Here's a first look at the upcoming feature.
Here's a first look at the upcoming feature.
Android Authority
First look: Google Keep will let you jot notes from the lock screen on Android tablets
You'll soon be able to launch Google Keep right from the lock screen on your Android tablet, making it even easier to quickly jot down notes.
👍73🔥19☃6🤯5👏2👎1
Chrome OS is finally preparing to add support for peer-to-peer Wi-Fi. This will benefit not only Quick Share but also wireless display mirroring.
Read more here.
Read more here.
Android Authority
Quick Share on Chrome OS might finally be as fast as it is on phones
Chrome OS is finally adding support for peer-to-peer Wi-Fi, which will make sharing files via Quick Share faster and more convenient.
👍65🔥15💩3👏2
MediaTek has announced that its Dimensity 9300 and 8300 chipsets are now "optimized" for Google's Gemini Nano LLM!
Gemini Nano is Google's mobile-optimized LLM currently available on the Pixel 8 Pro and Samsung Galaxy S24 series. It powers on-device generative AI features like the Pixel Recorder's on-device summarization and Gboard's smart reply.
"This effort includes collaboration with Google using MediaTek’s NeuroPilot toolkit and porting Gemini Nano to MediaTek APU to improve performance."
Gemini Nano is Google's mobile-optimized LLM currently available on the Pixel 8 Pro and Samsung Galaxy S24 series. It powers on-device generative AI features like the Pixel Recorder's on-device summarization and Gboard's smart reply.
"This effort includes collaboration with Google using MediaTek’s NeuroPilot toolkit and porting Gemini Nano to MediaTek APU to improve performance."
👍89🔥26🤮21🤔3💩2🤡2👏1
Games like Pokemon Go and Dragonball Legends have reportedly been crashing for many users with Android phones powered by the Snapdragon 8 Gen 3, such as the Galaxy S24 Ultra and OnePlus 12.
According to a Niantic help article, this is due to "our underlying game engine Unity's support for these devices." Pokemon Go version 0.301.0 seems to have included a fix for this issue, per user reports, but other Unity-based games may separately need to receive updates for this issue to be resolved.
If you own a Snapdragon 8 Gen 3-powered device, let me know if you've encountered this issue (and what games have crashed for you)!
According to a Niantic help article, this is due to "our underlying game engine Unity's support for these devices." Pokemon Go version 0.301.0 seems to have included a fix for this issue, per user reports, but other Unity-based games may separately need to receive updates for this issue to be resolved.
If you own a Snapdragon 8 Gen 3-powered device, let me know if you've encountered this issue (and what games have crashed for you)!
😁45👍13😘7👎2🤔2🤮2😭2❤1👏1
Qualcomm wants to make it easier for developers to deploy AI features on phones powered by the Snapdragon 8 Gen 3 and laptops powered by the Snapdragon X Elite, so today they’re announcing the Qualcomm AI Hub.
Qualcomm AI Hub is a central location where developers can access resources and download and deploy AI workloads onto these Snapdragon platforms. It contains a library of over 75 pre-optimized AI models. These models are compressed and optimized to take advantage of available hardware accelerators (NPU, CPU, and GPU) within the Qualcomm AI Engine. This results in “4X faster inferencing times,” reduced memory use, and better power efficiency.
Developers can choose a model from the library (like Whisper, ControlNet, or Stable Diffusion), indicate which framework to use for deployment (TensorFlow Lite, PyTorch, etc.) then choose a target platform (specific chipset or device), and finally deploy it with a few lines of code. The AI model library automatically handles model translation from the source framework to popular runtimes.
Developers can sign up today to the Qualcomm AI Hub to run these models on cloud-hosted devices and get early access to new features and AI models. New models will be added in the future along with support for more platforms and OSes. Qualcomm says these models will also be available on both Hugging Face and GitHub.
Qualcomm AI Hub is a central location where developers can access resources and download and deploy AI workloads onto these Snapdragon platforms. It contains a library of over 75 pre-optimized AI models. These models are compressed and optimized to take advantage of available hardware accelerators (NPU, CPU, and GPU) within the Qualcomm AI Engine. This results in “4X faster inferencing times,” reduced memory use, and better power efficiency.
Developers can choose a model from the library (like Whisper, ControlNet, or Stable Diffusion), indicate which framework to use for deployment (TensorFlow Lite, PyTorch, etc.) then choose a target platform (specific chipset or device), and finally deploy it with a few lines of code. The AI model library automatically handles model translation from the source framework to popular runtimes.
Developers can sign up today to the Qualcomm AI Hub to run these models on cloud-hosted devices and get early access to new features and AI models. New models will be added in the future along with support for more platforms and OSes. Qualcomm says these models will also be available on both Hugging Face and GitHub.
👍62🔥6💋4👎3❤1
Qualcomm’s AI Research team is showing off some of the on-device AI capabilities of their latest chips.
First, Qualcomm is showing a demo of Large Language and Vision Assistant (LLaVa), a 7B parameter LMM that accepts multiple types of data inputs, running offline on an Android phone. In the demo, a person asks a chatbot what ingredients are shown in an image of some fruits and vegetables, then asks it to create a recipe with those ingredients and estimate the calorie count of the recipe.
Qualcomm is also showcasing a demo of Low Rank Adaptation (LoRA) on an Android phone. Running Stable Diffusion with a lightweight LoRA adapter enables changing the tone/personality of the model without needing multiple fine-tuned models. LoRA is not only useful for language vision models (LVMs) but also large language models (LLMs) as well to create things like more personalized assistants.
Finally, Qualcomm is also showcasing a 7B parameter LMM that can accept text and audio inputs running offline on a Windows PC. In the demo, a person uploads a sound byte and asks the chatbot to provide a detailed description of the audio, the environment it might’ve been recorded in, what the vibe is like, and what era the music is from. Another demo compares the performance of GIMP with a Stable Diffusion plugin on a Snapdragon X Elite PC versus a PC running an Intel Core Ultra 7 - 155H CPU.
First, Qualcomm is showing a demo of Large Language and Vision Assistant (LLaVa), a 7B parameter LMM that accepts multiple types of data inputs, running offline on an Android phone. In the demo, a person asks a chatbot what ingredients are shown in an image of some fruits and vegetables, then asks it to create a recipe with those ingredients and estimate the calorie count of the recipe.
Qualcomm is also showcasing a demo of Low Rank Adaptation (LoRA) on an Android phone. Running Stable Diffusion with a lightweight LoRA adapter enables changing the tone/personality of the model without needing multiple fine-tuned models. LoRA is not only useful for language vision models (LVMs) but also large language models (LLMs) as well to create things like more personalized assistants.
Finally, Qualcomm is also showcasing a 7B parameter LMM that can accept text and audio inputs running offline on a Windows PC. In the demo, a person uploads a sound byte and asks the chatbot to provide a detailed description of the audio, the environment it might’ve been recorded in, what the vibe is like, and what era the music is from. Another demo compares the performance of GIMP with a Stable Diffusion plugin on a Snapdragon X Elite PC versus a PC running an Intel Core Ultra 7 - 155H CPU.
🔥45👍13❤2🥰2