Six weeks after launch New architecture v9 Arm also introduced the first processor cores built on it. Then it will add four new generation graphics chips. Either way, the performance and efficiency increase. We’ll see them in the first products next year.
The most powerful processor in the new Arm family will be the Cortex-X2. for him Predecessor x1 It plays the role of the top core in Snapdragon 888 and Exynos 2100. In the new, Arm believes in the rapid adoption of both the best cell phones and laptops. X2 promises better branching prediction, has 30% larger buffer and one cycle shorter pipeline.
According to Arm, its correct power should be increased Compared to X1 by up to 16%The performance will be in machine learning Double. Armv9 also brings vector modules to aid in artificial intelligence and scientific computations, in case the X2 SVE2 is in 128-bit width. SVE means more silicon, but Arm saved it by removing 32-bit instruction support from the kernel, so it’s a whole 64-bit processor.
New cell phones with Cortex-X2 cores offer up to 30% higher than existing ones (Source: Arm)
Cortex-X2 will be used most often in cell phones In a single-core variant, But the cutting-edge chips could also offer four cores along with the new Cortexy-A710, which will be discussed in the next chapter. They will then be able to target laptops as well Eight core processors It consists of only one set of X2 cores. Single-threaded power for the X2 according to Arm Up to 40% higher than Intel i5-1135G7.
How does IPC compare to other barks? As a basis, let’s take the two-year-old model A77, which we still find in many cell phones today. Cortex-A78 performance increased by 7%. Cortex-X1 is 20% more powerful than the A78. Then Cortex-X2 mentioned 16% before X1.
The kernel cores change into a three-digit instruction. The new Cortex-A710 is the successor to last year’s A78, and it will take on the role of serious, powerful and efficient workers. Here, too, there have been many changes to the level of fine architecture. Arm also shortened the pipeline and one cycle to a total of ten, reduced the number of instructions per cycle from 6 to 5, improved branching predictor, and increased L1 TLB in half.
Thanks to the tweaks, the Cortex-A710 should be too 10% more powerful than the A78Or just as powerful 30% more economical. The AI performance has doubled here, too. Unlike the X2, the A710 still supports 32-bit instructions, but according to Arm, all cores from 2023 will only work with AArch64. Thus, potential A711 or A720 cores will already be combed out.
Cortex-A710 (ZDROJ: ARM)
However, performance increases for all reported arm cores must be taken with some caution. The new DSU-110 platform, in which the company connects individual sets of different processor cores, can now hold up to 16MB of shared L3 cache, while its predecessors had a maximum of 8MB. Thus, larger cache can play an important role, its availability and capacity are not guaranteed in final products. It’s optional, but Arm relies on it in its marketing, full-size.
The Cortex-A710 was marked large in the arm hierarchy, and the new Cortex-A510 would be LITTLE. Four years later We’ve seen that since the launch of the A55 cores New energy saving units, Which can be found on a number of inexpensive cell phones and other electronic devices even without support for the most powerful A7x cores.
Due to the long hiatus, Arm has brought back most of the changes from the entire new family here. It still processes instructions in order, but re-processes two instructions instead of three for each action. The L1 Data and Help cache now contains at least 32 KB (instead of 16 KB) and the optional L2 cache can contain up to 512 KB (instead of 256 KB). The core is only 64-bit and because it is also built on Armv9, it offers SVE2 module, especially with 64 or 128-bit vectors.
Cortex-A510 (ZDROJ: ARM)
However, one change may have a somewhat negative effect on performance, but it is a positive effect on chip size and efficiency. Chip manufacturers can use what’s called an integrated microarchitecture, in which a single dual-core complex is used instead of two separate cores, where both cores share a L2 TLB and L2 cache and a NEON / SVE2 module.
One way or another, Arm believes the modifications will fundamentally increase performance. It should be Cortex-A510 Up to 35% more powerful than the A55 Its integer operations and up to three times faster in machine learning. The performance has already accommodated the large five-year-old Cortex-A73 core processors, the IPC is 10% higher in the A510, and the consumption is up to 35% lower.
Mali-G710, G610, G510 a G310
Arm not only develops processor cores, but also graphics cores. GPUs are used in Mali by Samsung, Huawei, Mediatek, and others, making them the most widely used in the world. To the family It will be added Three new graphics named Mali-G710 (the G78 successor in 2020), Mali-G610 (successor to the G68 in 2020), Mali-G510 (successor to the G52 in 2018), and Mali-G310 (the G31 successor in 2018).
They will all work on the third version of Valhall’s architecture and of course will vary in performance. It will be on top Mali-G710Compared to its predecessor Increases gaming performance with the same consumption by up to 20%Or, with the same energy, it will reduce consumption by 20%.
Mali-G710 (Source: Arm)
The architecture is again scalable, so Mali-G710 can contain from 7 to 16 compute units. The predecessors had up to 24, but they were smaller and less complex. Arm also promises to redesign the Task Manager to be more suitable for modern APIs like Vulkan.
However, Mali-G710 supports Vulkan 1.2, OpenGL ES 3.2, and OpenCL 2.0 Still without DirectXSo in the world of Windows we will not meet him. In addition, the Arm increased the overall performance in particular, but did not add new technologies, as it is now expanding into computers and controllers, especially real-time ray tracing, variable rate shading, etc.