The competition between Qualcomm and Huawei has always been, and now there are two hottest chips in Kirin and Xiaolong. How about their performance? Especially the Kirin 970 and the Dragon 835 who are mainly engaged in AI artificial intelligence. Huawei's officially announced Kirin 970 compared to the performance data of the Kirin 960, CPU efficiency increased by 20%, GPU performance increased by 20%, power consumption reduced by 50% (converted, energy efficiency increased by 140%). The biggest highlight of Kirin 970 is the introduction of the concept of NPU (full name "neural network computing unit"), which claims to be the world's first AI chip with a computing power of 1.92 TFP 16OPS. Kirin 970's AI technology comes from the famous Chinese Academy of Sciences, Cambrian Co., Ltd., which launched the Cambrian A1 processor last year. Huawei officially commercialized its chip into the Kirin 970 and named it NPU. For traditional CPUs and GPUs, the NPU has outstanding performance in terms of computing power and power consumption ratio. In other words, the NPU has less power consumption and smaller area under the premise that the performance is significantly higher than the traditional CPU and the mainstream GPU. Qualcomm Snapdragon 835 does not have the concept of an AI chip, but it also has a similar functionality of the Zeroth SDK, which can greatly improve the computing speed of AI requirements such as image and speech recognition. Let's take a look at the overall comparison: Kirin 970: 2.4GHz 4x A73 + 1.8GHz 4x A53, and Kirin 960 architecture, the frequency is exactly the same. According to the Kirin 960 test, Ann Bunny runs about 33268 points (CPU single item), GeekBench single core 1933 points, multi-core 6203 points. Xiaolong 835: 2.45GHz 4x Kryo 280 + 1.9GHz 4x Kryo 280, is considered to be the magic version of A73+A53, Ann Bunny runs about 38,039 points (CPU single item), GeekBench single core 1944 points, multi-core 6180 points. According to Master Lu's performance on the Android camp chip performance, this basis is Lu Master's Android v8.4 test run, Huawei Kirin 970 118758 points to the top, Qualcomm Xiaolong 835 112462 points closely followed, although the gap Not big (about 6%), but Huawei once again achieved a transcendence of Qualcomm's top flagship. Taking into account, Huawei's Kirin 970 and Snapdragon 835 CPU performance is very close, all in the early 41,000 points, GPU is Kirin 970 lead by about 9%, the first commercial 12-core Mali-G72 made a big contribution, surpassing Adreno 540 . The previous generation Huawei Kirin 960 also led the Dragon 821, but the gap was only 3.5%. In 2017, the mobile phone flagship processor was on the 10nm process. The Huawei Kirin 970 uses the TSMC 10nm FinFET process; the Qualcomm Snapdragon 835 uses the Samsung 10nm FinFET process. The Kirin 970 and the Snapdragon 835 are also 10nm processes. In fact, Samsung and TSMC still have some differences in power consumption. At present, it seems that Samsung's 10nm process is more mature than TSMC, and its production capacity is better. After all, Xiaolong 835 began to supply very early. And before the Huawei Qilin 960 adopted a relatively "bold" configuration, although the performance is strong, but the power consumption is not low, it was praised by some netizens as "fire unicorn." Looking forward to the adoption of TSMC's latest 10nm manufacturing process, Kirin 970 has a better performance in power control. However, if Samsung considers the water injection in the manufacturing process in the past, Samsung's 14nm manufacturing process is not as good as TSMC's 16nm manufacturing process. Therefore, if Samsung maintains the consistent style of the past, in the manufacturing process, the Kirin 970 will slightly occupy a certain advantage. . The CPUs of Kirin 970 and Xiaolong 835 are both large and small nuclear structures, and the Kirin 970 is basically not upgraded. In 2017, Qualcomm, Samsung, and Huawei's flagship SoC all use the "four big four small" size core structure in the CPU part. Compared with Qualcomm and Samsung, Huawei still uses the ARM Cortex-A73 public version core, while Qualcomm and Samsung select the Kryo 280 self-developed core and the Exynos M2 semi-self-developed core. The Kirin 970 integrates the quad-core Cortex A73 and the quad-core Cortex A53. The performance is not much improved. The Qualcomm Snapdragon 835 uses the 8-core Kryo 280. Although naming, the Kryo 280 is much better than Kryo. However, in terms of actual performance, the Kryo 280 is very limited in its improvement over Kryo. The Kryo 280 cuts off the more powerful floating point performance of Kryo, and the fixed-point performance is not much different. Similarly, the Cortex A73 is also designed to maintain point performance relative to the Cortex A72, cut the floating point performance, and strive to improve the performance and power ratio. Therefore, Cortex A73 and Kryo 280 can be considered to be roughly the same level of CPU core. Considering that the Kryo 280 is much better than the Cortex A53, the Kirin 970 and the Qualcomm Snapdragon 835 are generally similar in terms of CPU. However, in some scenarios, when the Kirin 970's four Cortex A73s and the high-frequency quad-core Kryo 280 are offline, Qualcomm's quad-core low-frequency Kryo 280 is superior in performance to the Unicorn 970's quad-core. Cortex A53. The CPU gap is not too big, and the possibility of being flat is very large. The Kirin 970 is basically no upgrade in the CPU part compared to the Kirin 960. In the GPU part, Qualcomm still integrates its own Adreno series, while Samsung and Huawei both use ARM's Mali core. The Snapdragon 835 is still an absolute advantage in terms of GPU performance, and the Samsung Exynos 8895 and Kirin 970 are likely to be very close on the GPU. GPU: Adreno 540 vs. Mali-G71 and G72 brothers Snapdragon 835: Adreno 540 with a frequency of 710MHz. Ann Bunny runs 74325 points (3D performance single item), GFXBench Manhattan off screen 1080P best score of 42fps. Bilge float alarm,Bilge float alarm price,Bottom floating alarm Taizhou Jiabo Instrument Technology Co., Ltd. , https://www.taizhoujiabo.com
