Auch das Smartphone-Speicher hat eine "gemeinsame Nutzung". Das neue Google-Handy hat einen "AI-spezifischen" Speicher.
Shortly before, Google presented the annually updated flagship Pixel 10 series at the "Made by Google" event. As a true trendsetter for Android smartphones, the Pixel 10 series has demonstrated its strengths with features such as real - time language translation, conversion of singing into songs, Magic Cue (Magic Tips), and Camera Coach (Camera - Coach).
While Apple's Apple Intelligence still has weaknesses, Google is currently unbeatable in the mobile AI industry. However, the successful "landing" of the Pixel 10 series in this market comes at a price. If consumers don't specifically buy the new models because of their AI capabilities, they have to pay for features they may not need.
The foreign tech magazine Android Authority recently analyzed the standard Pixel 10 model and found that Google has reserved 3.5 GB out of the 12 GB of RAM specifically for AI functions, which can only be called via the TPU (Tensor Processing Unit) of the Tensor G5. Apparently, Google has not only "downgraded" the GPU of the Tensor G5 but also adjusted the RAM to improve the AI functions at the device level.
In comparison, the Pixel 9, also with 12 GB of RAM, does not have a "AI - specific RAM" implemented. Obviously, the changes Google has made to the Pixel 10 probably come from the experience with the Pixel 9 series. Since the data processing of large AI models has to be done locally at the device level, a sufficiently large RAM is required to store the entire model and ensure a good user experience. If the flash memory is used, the duration of AI tasks such as chatting and image editing would increase to minutes.
Imagine if a user wants to use the "convert singing to song" function of the Pixel 10 and the flash memory is used for data transfer, it could take several tens of minutes for the device to generate a song. In today's fast - paced world, probably no user has that much patience. In addition, the on - device AI requires a fast cache and a higher RAM bandwidth, which inevitably occupies part of the RAM.
The problem is that the Pixel 10 not only has AI functions but also other apps occupy RAM. If a user opens too many apps, it can cause the system to freeze when running the on - device AI. Most users can't tell the difference between apps like ChatGPT, Doubao, and Yuanbao and the on - device AI, and they also don't know that the performance cost of the latter is comparable to that of demanding mobile games.
Therefore, Google has planned a 3.5 - GB "AI - specific RAM" on the Pixel 10 to solve the problem of delays that may occur when running the on - device AI and other apps simultaneously. The size of 3.5 GB was specifically calculated. The currently commonly used 7B model with 4 - bit quantization at the device level requires 3.5 to 4 GB of RAM.
Without planning an "AI - specific RAM", the problem of the conflict between the on - device AI and the multitasking requirements can also be solved, namely by "closing background apps". A few years ago, as the power consumption of SoCs increased significantly and the 5G modem chip became a major power consumer, Android manufacturers have increasingly closed background apps to improve battery life and use the physical RAM more efficiently.
It's not just the case with Android. Since the update to iOS 18.4, many users have found that the "closing of background apps" is very prominent. The most obvious example is that when opening WeChat, the startup page with the Earth is always displayed. At that time, many users suspected that Apple may have reserved a specific RAM for Apple Intelligence.
Although Google's approach is beneficial for the performance of Gemini AI on the Pixel 10, it poses a risk to the promise of seven - year system updates. In view of the Andy - Bill's Law, the computing power provided by the hardware is always consumed by new software applications. Therefore, old devices often have difficulty coping with new software versions.
The currently smoothly running Pixel 10 will probably reach its limits in four or five years. Whether the 8.5 - GB RAM for apps will be sufficient in seven years is a big question. This means that Pixel 10 users may find that their smartphones slow down earlier than expected. In short, Google has ensured with the "AI - specific RAM" that Pixel 10 users can use the full power of AI in the first two years, but the shortening of the actual lifespan of the smartphone as the price has not been made public.
The biggest problem with Google's "AI - specific RAM" concept is that it takes away the choice from users. Although the "AI - specific RAM" enables the Pixel 10 to fully utilize the on - device AI power and bring the fascination of AI support on smartphones closer to users, not all users need the Magic Cue and Camera Coach supported by the on - device AI. The app versions of ChatGPT and Doubao already meet the needs of many people.
In other words, the "AI - specific RAM" concept on the Pixel 10 means that consumers who are not interested in AI functions spend money unnecessarily. Although modern smartphones have more and more functions and there are functions that each type of consumer doesn't use, RAM is different from IP68 dust and water protection and NFC. It is a basic component of the smartphone and has a direct impact on performance.
For users who don't use the on - device AI, there is even suspicion of false advertising. The advertisement claims that the Pixel 10 has 12 GB of RAM. However, if you don't use the on - device AI, the actual RAM of the Pixel 10 is only 8.5 GB. As a netizen said: "Please list the shared areas separately and don't include them in the total area."
This article is from the WeChat account "3e Life" (ID: IT - 3eLife), author: 3e Jun. 36Kr published it with permission.