And in the Cortex-78CPU just released, aren't you satisfied with its performance improvements?Even the 5nm process can only improve the performance by 20%And it's still pulled by frequency. Yes, cortex-78 is positioned in this way. The real high-performance CPU is the new cortex-x1, which is also the core of the first cortex-x series.
Different from cortex-a78 in terms of performance, power consumption and area, cortex-x1 uses a new architecture, which is designed by arm to achieve significant performance growth. Its performance is 30% higher than that of previous generation CPU, 22% higher than that of cortex-a78, and machine learning performance is 100% higher.
Cortex-X1, in addition to the surge in performance, also offers more flexible customization features, and you know Cortex-A there isn't much that can be changed from public licensing to caching, andCortex-X1 allows customers to customize, with more different features, allowing customers to participate in the early stages of R & D.
Of course, Cortex-X1 authorization fee is certainly not the same, more expensive than Cortex-A series is not running. Now, as Qualcomm and Samsung have abandoned ARM core of their own research in recent years, other manufacturers besides Apple have changed back to the public version of the architecture, ARM now launch Cortex-X1 is also a kind of adjustment, equivalent to an official pre-reform version, Qualcomm ,Huawei, Samsung and other customers can continue to change, customized different features.
Comparison between cortex-x1 and cortex-a78
Compared with cortex-a78, the differences of cortex-x1 are almost the same as before, mainly higher peak performance and more flexible customization.
Cortex-X1 and Cortex-A78 are both under ARMv8.2 instruction set in architectural detail, the instruction set is compatible, but Cortex-X1 is a custom CPU core, the decoding bandwidth is raised from 4 to 5, an increase of 25%, and the floating point is raised from 2 128 to 4 128 b, which is equivalent to doubling floating point performance.
In terms of cache, the L1 cache of cortex-x1 can reach 64KB, the L2 cache can reach 1MB, and the L3 cache can reach 8MB, twice that of cortex-a78.
Cortex-x1 performance improvement level: 30% for single core and 100% for AI
How about the performance improvement of cortex-x1? According to the data released by arm, compared with the a77 architecture, the integer performance is improved by 30%, compared with the A78 architecture, by 22%, and the ML machine learning performance is improved by 100%.
Of course, those performance improvements are theoretical values. In SPEC2006 testing, the single core performance increases by 4% to 30%, and it seems that there is no awesome force. But now it is the performance of ARM Public Edition Cortex-X1. Maybe other manufacturers can customize stronger performance.
Cortex-X1 flexible collocation: acting as a super-large core 3 4 cluster cores
With the Cortex-X1, the first thought is actually super-nuclear, because Qualcomm ,Huawei in the Cellon 865, Kirin 990 design three core, unlike the previous 4 4 size match, but become 2 2 4 or 1 3 Four, one or two of the large cores are independent, and the frequency is the highest.
Now, the advent of cortex-x1 represents that arm officially recognizes this multi-core configuration. Cortex-x1 can be matched with cortex-a78 and cortex-a55 cores.
A typical combination is one cortex-x1 super large core, three cortex-a78 large cores, plus four cortex-a55 low-power small cores, with 1MB L2, 8MB L3, forming a dynamiq cluster.
With 4 cortex-a78 and 4MB L3 cache, the performance is 20% higher than that of the previous generation, while the core area is reduced by 15%. With 1 cortex-x1 and 3 cortex-a78 and 8MB L3 cache, the core area will increase by 15%, but the peak performance will increase by 30%.
The rest depends on which company can launch cortex-x1 CPU, Qualcomm, Samsung, apple or Huawei?