Copilot on Windows 12 will require 16 GB of RAM


Major processor manufacturers, such as Intel, AMD, and Qualcomm, now sell processors with neural network processing units (NPUs) for use in PCs. And that’s good, since Microsoft’s new operating system, Windows 12, is also expected to require significant memory and NPU processing power to do AI in local mode. And this will have an impact on RAM.

“Windows 12’s AI processing will require 16 GB of memory, which will serve as a catalyst for increasing PC memory capacity and also accelerate renewal demand for consumer machines,” the IT firm said this week. TrendForce market research. “While the speed of DDR5 memory can reach 5600 Mbps, LPDDR5x memory is better suited to AI PCs that require bandwidths of at least 7500 Mbps and up to 8533 Mbps. We expect LPDDR5 memory to account for up to to 35% of total PC demand this year.”

Most laptops on the market today come with 8GB DDR4/5 memory as standard.

Copilot runs at the same time as Windows, so more power is required

After starting Windows 11, there is about 3-4 GB left, which is enough to run web browsers like Google Chrome and Microsoft Office. However, Microsoft is expected to double the minimum memory requirements for AI PCs and increase it to 16 GB in its new operating system, Windows 12 (whose name is very provisional), and which should be released this year.

The reasoning is: the Copilot feature will need this extra capacity because it runs concurrently with the operating system.

The main PC manufacturers we spoke with during CES 2024 agree with this. “AI PCs will be a growth driver this year, and the 16GB base memory increase is a no-brainer,” said one manufacturer.

AI will also have a strong impact on SSDs

AI PCs will also have a big impact on SSD performance. The need to increase data throughput to do generative AI will require the adoption of PCI Express 5.0-based SSDs, which exceed the current maximum of 8 GB per second.

PCI Express 5.0 SSDs are rare in laptops due to their high heat and high power consumption during high-performance operation. However, Taiwanese SSD controller manufacturers such as Siliconmotion and Pison have recently released low-power controller chips using sub-7nm microprocessor.

But current laptop processors still only use PCI Express 4.0 standards for SSDs. New processors will need to be brought to market to take advantage of this development. “With the launch of laptops capable of running generative AI, we expect to see PCI Express 5.0 SSDs in lightweight laptops starting in the second half of this year,” mentions a source from the storage world .

But there are still other prerequisites for the future Windows 12 to work properly.

40 TOPS for running an AI locally

TrendForce estimates that the minimum computing power required by Microsoft to run its AI is 40 TOPS (40 trillion operations per second). This is a simple count of the number of times an AI operation needs to be processed, rather than the type or shape of the data, like INT4/8/16 or FP16/32. Actual processing speed may vary depending on algorithm or software library, or data type.

Current laptop processors, such as the Intel Core Ultra and AMD Ryzen 8040, are known to exceed 40 TOPS via a combination of NPUs and GPUs. However, it is much more efficient to improve NPU performance using GPU because it consumes less power than NPU.

Intel’s “Luna Lake” laptop processor, which will be released in the second half of this year, will achieve a three-fold increase in NPU performance. “Lunar Lake will deliver up to three times higher NPU performance than the current Core Ultra (Meteor Lake),” said Intel Vice President Michelle Johnston Holters, who unveiled a prototype of the upcoming computer processor laptop during CES 2024.


Source: “ZDNet Korea”



Source link -97