This week, Xumi introduced the flagship chip set at home for smartphones and tablets, Xring O1. This Xiaomi 15s Pro and Xiaomi Pad are featured on 7 ultra. Some people suspected how much Xiaomi actually did here – finally, O1 uses standard CPU and GPU core from the bun of parts. The team of Gickeron made a deep dive and discovered that it was actually a heavy -on -custom chip.
Xring O1 is developed on TSMC N3E Node, which is similar to 9400 Meditik’s dimensions. And it also uses some of the same parts-such as CPUs for Cartax-925 and GPU for Amlas-G 925. However, the design is very different, very different.
For those who start, the board has two cortex-X925 prime core, not just one. And Xumi left the X4 instead of choosing two variations of Cartax-A725 for the big and middle core. And he added two A520 core as a safety net.
They are an interesting element of the cortex-A725 core O1 design. There are two clusters. In the dye shot shown in the video you can see that the two performance A725 core are physically different from four performance A725 core – they are slightly larger and are limited to very low clock speed (1.9GHz vs 3.4GHz).
This allows the chip set to be replaced between the core while using the low power possible to meet the performance requirements. Low power A725 cores are so good that two A520 core may not be necessary.
The final result is quite impressive – Xiomi’s Xring O1 CPU defeats dimension 9400 in terms of performance and power efficiency. This is not at all at the Snap Dragon 8 Elite level, but it is not far away.
Zering O1 is quite small at 109mmmmiron – it’s about the same size as Apple A18 Pro. However, both Xumi and Apple chips use the external modems, while dimensions are built -in in 9400 and Snapdragon 8 Elite. This describes a part of the contradiction (and creates a problem after which we will come later).
The Xing team formed on-chip memory in an interesting way-most of these days have several megabytes of system level cache (SLC), which is the last stop before bringing data from Ram.
O1 has no SLC. Instead, Xiaomi formed CPU with a lot of cache – L3 16 MB jointly jointly between all 10 cores, 2 MBL 2 each for X 925 core and 1 MBL 2 and 1MB L2 for each 725 core. GPU has 4MB cache and NPU has 10 MB – look at it, it’s huge!
The NPU is not a arm design, but instead is a custom Xiaomi creation. 6 -core NPUCPUs raise almost AS on the chip, like the NPUCPU. The ISP is also made by the fourth -generation design Xiaomi. The previous generations have been highlighted as separate chips on the Mother Board of the Xiaomi flagship, but it is more efficient to build.
Now we come to the GPU-the 16-core Amlas-G 925 from Bazu. The Zereng team did a wonderful job to form the CPU, but it may have gone wrong with the GPU. For one, it is quite large-12 core G925 in the 9400.
Also, the Gakrwan team has speculated that the SLC error has damaged GPU’s performance – it is very fast, but it uses more power than GPU in peak performance.
The more efficient CPU performs well with the fact that the GPU rarely bowed in full inclination in real -life gaming tests. Nevertheless, it is an area that should be improved in future designs.
Another problem with performance – the outer modem (in this case uses more power), which hurts the standby endurance of the Xiaomi 15S Pro. That is why Apple worked so hard on the integrated 5G modem (which started with the iPhone 16E).
Xiaomi is not yet present, but he took the first step – he introduced his first 4G modem with Xring T1, a chip set for the corrected Xumi Watch S4. It is much more difficult than 5G4G (it took Apple to be true), but this will be another major milestone towards Vertical integration for Xiaomi.
With this, we leave you to watch the video, which goes into great detail how Xring O1 is made and how it performs:
