Huawei’s Kirin 980 Features Two Clusters Of A76 Cores, Mali-G76, And A Dual NPU
IFA 2018 is in full swing and we’ve got lots of new product launches. For smartphone hardware, the world’s first 7nm processor is finally official. Huawei’s Kirin 980 is with us and it’s officially the world’s first 7nm smartphone processor. The Kirin 980 is manufactured on TSMC’s N7 node – the last Deep Ultraviolet Lithography based manufacturing process from the fab, as it will make the shift to EUV next year and effectively start producing the N7+, which will be referred to as 7LPP (Low Power Plus).
With the Kirin 980, Huawei’s introduced its best smartphone processor in recent history. The company hit a speed bump (literal and figurative) with its predecessors that utilized either low-performance GPUs, CPU cores – or both. But the Kirin 980 finally makes up for all of this and Huawei should be proud. Take a look below for more details.
Huawei’s Kirin 980 Is 7nm, Features 2xA76 Cores @ 2.60GHz, 2xA76 Cores @ 1.92GHz, And 4xA55 Cores @ 1.80GHz
Starting from its basic block design, Huawei’s the first with the Kirin 980 at two ends. Not only does the company utilize TSMC’s 7nm process but it also becomes the first mobile processor to use ARM’s freshly launched Cortex A76 cores. The Cortex A76 is a one of a kind and mark ARM’s first crucial step at trying to hit two birds with one stone, or should we say, core.
The Cortex A76 is capable of reaching frequencies of up to 3.3GHz, features a brand new micro-architecture with a new branch prediction unit, reduced instruction latency comparable to Samsung’s Mongoose M!, a 33% wider decode width and super-fast memory fetch. The Huawei Kirin 980 features four Cortex A76 cores in two clusters, medium, and high-performance.
These are clocked at 2.60GHz and 1.92GHz respectively and effectively allow Huawei to leverage the A76’s architectural improvements and reduce its power penalty by effectively dividing the cores into two different power brackets. This will ensure that programs that don’t ‘make the cut’ for the high-frequency cluster are not pushed down to the low-frequency cluster instead. ARM’s DSU (DynamIQ Shared Unit) will ensure that the two clusters perform as needed.
Moving towards the Kirin 980’s third core cluster, the processor utilizes four of ARM’s Cortex A55 cores that are clocked at 1.80GHz. The problem of scheduling that will result from three separate core clusters on the Kirin 980 will be managed by a new flexible scheduler. The Cortex A76 cores each feature 512KB of L2 Cache and the Cortex A55 cores feature 128KB of L2 Cache memory. The Kirin 980’s DSU features 4MB of L3 Cache.
Huawei’s numbers for the Kirin 980, especially for power efficiency of the GPU either demonstrate the true strengths of ARM’s designs or a serious measuring error. Before we get to the company’s power estimates, it’s better to start off with its claims of performance and power efficiency improvements of the CPU.
According to Huawei, the Kirin 980 is able to achieve a 75% performance gain over the Kirin 970. The 7nm processor can also reduce power consumption by 58% over its predecessor and according to Huawei, the Kirin 970 is also 50% more efficient over 2018’s processors – by which the company most likely means Samsung’s Exynos 9810 and Qualcomm’s Snapdragon 845. The Kirin 980 also has 6.9 billion transistors over a smaller die. For comparison, Apple’s A11 has 4.3 Billion transistors. Goes to show how much of a difference a new manufacturing process can make.
The Kirin 980 Features ARM’s Mali G76MP10 GPU, A New ISP, And A Dual NPU That Promises A 134% Improvement In AI Performance
Moving towards the GPU, it’s another component that Huawei needs to focus on if it wants its processors to become competitive. The Kirin 980 will feature ARM’s Mali G76MP10 GPU, becoming the first processor with the unit. The Mali G76 represents a massive increase over the G72 and can do much more in a smaller package. According to Huawei, the Kirin 980’s G76 will offer a 46% performance boost and a staggering 176% power-efficiency gain over its predecessor.
The gold standard for GPU performance is still Qualcomm’s Adreno lineup and only if the G76 can outperform the current Adreno 630 or the upcoming Snapdragon 855’s chip, will Huawei be able to stand out in the market. We don’t expect it to and you shouldn’t either – but of course, concrete conclusions can only be reached once real-life performance tests are available.
Speaking of standing out, Huawei’s building on its NPU with the Kirin 980. The processor increases the NPU’s size over the previous generation and Huawei believes that this will introduce some important improvements that will bring the ball right up in Apple and Qualcomm’s court. Huawei believes that the Kirin 980 will improve Neural AI performance by up to 134% and bring down power consumption by 88%.
As GPU workloads start to increase, manufacturers are finding it easier to relegate image recognition tasks to separate cores and NPUs are definitely here to stay. Given the excitement Apple created with the A11’s NPU last year, we’re really looking forward to the company’s upcoming launch event next month.
Thoughts? Let us know what you think in the comments section below and stay tuned. We’ll keep you updated on the latest.