The Major Regret of Kirin 990! What’s So Good About ARM Cortex-A77 Architecture?

Click the above Computer Enthusiasts to follow us

The latest Kirin 990 series released by Huawei is undoubtedly the hottest flagship SoC of 2019, with TSMC’s 7nm+EUV process, native integrated 5G baseband, dual large core NPU + single micro core NPU (Kirin 990 5G version), ISP 5.0 and other new features, making this chip’s overall strength surpass Qualcomm’s strongest Snapdragon 855 Plus, with only the Snapdragon 865, which will be launched next year, able to compete.

However, the biggest regret of the Kirin 990 series is that it continues to use the Cortex-A76 and Mali-G76 architecture for the CPU and GPU from the Kirin 980, while new products represented by Samsung’s Exynos 980, Snapdragon 865, and MediaTek’s 5G SoC will adopt ARM’s latest Cortex-A77 and Mali-G77 architecture.

So, as a brand new CPU/GPU architecture just released by ARM in May 2019, what changes have the Cortex-A77 and Mali-G77 actually made?

ARM’s Annual Iteration Strategy

ARM releases a new generation of IP (core licensing) every year.

For example, the Cortex-A73 released in 2016 (Kirin 970, Helio X30, and Snapdragon 835), the Cortex-A75/A55 released in 2017 (Snapdragon 845), and the Cortex-A76 brought in 2018 (Kirin 980, Snapdragon 855).

In 2019, ARM once again introduced the brand new Cortex-A77 CPU architecture, combined with the synchronously upgraded Mali-G77 GPU, what changes can they bring to smartphones?

Interpreting Cortex-A77 CPU

Before interpreting the Cortex-A77, we need to review its two predecessors to better understand this brand new CPU architecture.

Standing on the Shoulders of Giants

In ARM’s recent CPU architecture updates, Cortex-A75 and Cortex-A76 hold significant historical positions.

The Cortex-A75 introduces DynamIQ technology based on big.LITTLE, maximizing the flexibility and scalability of the architecture, allowing combinations like 2 (big cores, i.e., A75) + 6 (little cores), 3 + 5, 1 + 7 to become possible.

At the same time, the “little cores” synchronized to the Cortex-A55 architecture are no longer synonymous with weak performance. When different numbers of little and big cores are paired, a relatively perfect (performance) and (power) balance can be achieved.

As the successor to Cortex-A73 and A75, Cortex-A76 adopts a completely new CPU microarchitecture (based on ARMv8.2 instruction set design) from scratch, and collaborates with TSMC’s 7nm process, bringing the largest leap in performance and efficiency in ARM’s history—Snapdragon 855 has over 40% CPU performance improvement compared to Snapdragon 845, and Kirin 980 has over 40% CPU performance improvement compared to Kirin 970, while battery life has also been extended.

In ARM’s words, the Cortex-A76 is a “laptop-level” high-performance processor architecture, which also has high efficiency. Qualcomm’s “Snapdragon laptops” focuses on the “Always Connected PCs” concept, which is a manifestation of this advantage.

ARM’s latest released Cortex-A77, codenamed “Deimos (the god of fear),” adopts the same CPU microarchitecture as A76 and has undergone certain optimizations based on it, ultimately achieving higher IPC performance improvements.It can be said that the A77 architecture is an iteration based on “standing on the shoulders of giants,” and its performance is naturally worth looking forward to.

Further Improved Performance

Since the Cortex-A77 inherits the microarchitecture from A76, there are many commonalities in design between the two generations of cores.In ARM’s words, chip suppliers (like Qualcomm, MediaTek, etc.) can easily upgrade the SoC’s IP design when building cores without spending too much effort and cost, thus shortening the development cycle.

Compared to Cortex-A76, the A77 core has a higher read bandwidth in the front end, and the branch predictor’s target buffer capacity has increased by 33%, thereby reducing branch mispredictions and improving prediction accuracy.

At the same time, A77 also brings a brand new Macro-Op cache structure, adds an additional integer ALU, and introduces a feature called “system-aware prefetching,” which can avoid unnecessary impacts on the shared cache, thus improving overall system performance.

According to data released by ARM, when using the same 7nm process technology and running at the same frequency of 3GHz (theoretical only, as actual flagship SoCs often run at frequencies between 2.6GHz and 2.8GHz), the new Cortex-A77 has a 20% increase in memory bandwidth, a 20% performance improvement in SPEC int2006 and Geekbench 4, and a 30-35% improvement in floating-point performance.

In the smartphone field, web browsing happens to be a killer application that tests floating-point calculations, so the significant improvement in A77’s floating-point performance should further enhance the actual experience.Based on the current performance of Snapdragon 855 and Kirin 980, the CPU performance of Cortex-A77 should be sufficient to surpass Apple’s A12, and is expected to compete with Apple’s yet-to-be-released A13 and Samsung’s next-generation self-developed Mongoose M5 core.

Understanding Mali-G77 GPU

Are you amazed that ARM Cortex-A77 has a 20%-25% IPC performance improvement over its predecessor A76 while maintaining the same power consumption?In fact, ARM’s latest Mali-G77 GPU has seen even greater improvements over its predecessor.

The Temptation of a New Architecture

Compared to the GPUs integrated into Qualcomm and Apple’s SoCs (such as Adreno GPUs), ARM’s own Mali series GPUs have always been in a lukewarm state; if it weren’t for ARM’s “own child,” its influence might be less than that of the PowerVR series GPUs.

To make up for its GPU’s shortcomings in the graphics field, ARM Mali-G77 abandons the patchwork strategy from Mali-G71 to G76 on the “Bifrost” architecture, and instead adopts a new generation architecture called “Valhall,” which supports a new ISA bus and computing core design, with independent units designed for AI computing, focusing on enhancing performance in anti-aliasing, high resolution, and AI-assisted computing, making it better suited for future 4K resolution screens and top devices and standards that support HDR gaming.

According to official data from ARM, Mali-G77 can improve performance and efficiency by 30% compared to Mali-G76, while AI performance has increased by 60%, with performance per square millimeter being 1.4 times that of G76.

Compared to the earlier Mali-G72, Mali-G77’s power consumption has also decreased by 50%.Of course, this power consumption part should not be overly concerned, as Mali-G72 was paired with a 10nm process, while Mali-G77 will soon use the latest second-generation 7nm EUV process, making power consumption reduction inevitable.

In other words, Mali-G77 finally has the foundation to compete with Qualcomm’s Adreno 640 GPU integrated into the 855.However, how much performance the ARM Mali series GPU can actually unleash will ultimately depend on how many computing cores the chip suppliers equip it with. For instance, Kirin 980, Samsung Exynos 9820, and Kirin 990 all integrate Mali-G76, but Kirin 980 only prepared 10 computing cores for it, which is Mali-G76MP10, Exynos 9820 uses 12 computing cores, i.e., Mali-G76MP12, while Kirin 990 is equipped with 16 computing cores of Mali-G76MP16, so in terms of 3D performance, the latter is more advantageous.

In addition, the competitors that Mali-G77 is about to face will also upgrade to Snapdragon 865 and Apple’s A13, and these competitors’ GPUs will only get stronger. When the time comes, who will be superior compared to Mali-G77? We will have to wait for time to verify.

Click “Read the Original” for more exciting content

Leave a Comment Cancel reply