The rise of MediaTek has broken the dull SoC market pattern of mobile platforms. The performance testing of Tianji 9000 and Xiaolong 8 Gen 1 has been lifted. Now let's analyze these two flagship SoCs in 2022.
Specification comparison and analysis
Qualcomm Snapdragon 8 Gen 1 was released on December 1, with Samsung 4nm process, X2 super core (1MB L2)+3 A710 large cores (3x512KB L2)+4 A510 small cores (2x256KB L2), with frequencies of 3.0GHz+2.5GHz+1.8GHz respectively. Although the number suffix is no longer used, the GPU has finally crossed generations to the Adreno 700 series.
MediaTek Tianji 9000 released on November 19 that TSMC's 4nm process is also X2 super core (1MB L2)+3 A710 large cores (3x512KB L2)+4 A510 small cores (2x256KB L2), with frequencies of 3.05GHz+2.85GHz+1.8GHz respectively. The GPU also uses the new architecture Mali-G710 MC10. The core size of this generation of GPU has doubled and is actually close to the previous MC20.
All the architectures, including the A55, which is "one forever", have been updated this time, marking a new era of Armv9 instruction set. Both of them use the new features of A510. The four small cores are 2+2 structures and share 256KB L2 caches respectively.
Key differences:
Large core frequency: The frequency of the three A710s of Tianji 9000 is as high as 2.85GHz, which is higher than the super core frequency of most SoCs. Qualcomm has "only" 2.5GHz, with a frequency difference of 14%;
Memory support: Tianji 9000 launched 7500Mbps LPDDR5x memory support for the first time. Snapdragon 8 Gen 1 is still 6500Mbps LPDDR5, with a frequency gap of 15.4%. For the time being, there is no LPDDR5x memory for mass production, but it is hard to say in 2022;
Cache: Snapdragon 8 Gen 1 is a 6MB L3 cache+4MB SLC (system level cache), while Tianji 9000 is a more exaggerated 8MB L3 and 6MB SLC, 33% and 50% larger respectively. As a reference, the era of Snapdragon 845 and 855 was 2MB L3+3MB SLC, and the era of Snapdragon 865 and 888 was 4MB L3+3MB SLC. The L3 of Tianji 9000 is 1MB larger than the L3+SLC of 888
APU： Qualcomm is a hybrid architecture, without a separate APU. The related load is completed by the Hexagon DSP/CPU/GPU. The seventh generation AI engine only publishes data that improves performance by 4 times and energy efficiency by 1.7 times. MediaTek is the fifth generation of independent APU 590, 6 core, with mixed precision design, which claims to improve performance and energy efficiency by 4 times.
ISP： There are three 18bit ISPs on both sides. Qualcomm Spectra also hides the digital suffix, supports up to 200 million pixels of photo shooting, 3.2 billion pixels per second of processing power, and adds functions such as always online low-power ISP responsible for facial unlocking. The Tianji 9000 is the Imagiq 790. Three 18bit ISPs all support triple exposure, and can process 18bit HDR video at the same time. It can support up to 320 million pixel cameras. The processing speed is 9 billion pixels per second, and the data processing speed is more than 2.8 times that of Snapdragon 8 Gen 1.
In addition to the GPU, which has no detailed data and cannot be directly compared, MediaTek Tianji 9000 is a textbook style "miracle with great efforts". Almost all the specifications of the energy stack have exploded. It has more advanced processes, high frequency, large cache, and the most exaggerated ISP throughput is 1.8 times different.
Performance measurement
The Tianji 9000 engineering machine does not have a complete heat dissipation design. It uses 12GB LPDDR5 6400Mbps memory+256GB UFS 3.1 flash memory and LCD screen (claimed that the power consumption is 0.6W higher than that of A screen). The Qualcomm QRD quoted below is a complete testing machine, while the moto edge X30 is a mass production machine. In terms of test conditions, MediaTek itself has made a concession, and the performance rate of mass production machines will probably be better than that of engineering machines.
GeekBench, GFXBench and Anthare for MediaTek engineering machines are offline versions, but there is no big difference between the subproject and the latest version, which can be compared horizontally.
In the GeekBench 5 test, we added the comparison of Snapdragon 888+, Kirin 9000, Apple's A13, A14, and A15 generations. The A15 also includes the full blood version on the iPhone 13 Pro series, the youth version of the iPhone 13/13 mini 4-core GPU, the iPad mini 6 CPU, and the Max-Q version of the GPU frequency reduction.
As a result, two X2 super nuclei with similar specifications and frequency difference of 1.67% had a single core score difference of 3.35%, which was not significant. However, in terms of multi-core performance, Tianji 9000 surpassed A14 with a score of 4291 points, pushing the score of Android camp forward by a big step and leaving Qualcomm by 14.1%.
Compare the sub projects of GeekBench 5, and join the companion run of Snapdragon 888 and 888+. Even the X2 super core with similar specifications and performance, Qualcomm and MediaTek have different tuning and focus. Tianji 9000 is significantly stronger in PDF rendering, Gaussian blur, image compression and other projects, with a range of 10% or more.
For the multi nuclear project, Tianji 9000 led by 21% in floating point score when its total score was 14% strong. Looking at the ratio change between single core and multi-core, the 2.85GHz UHF A710 of Tianji 9000 seems to be weak in HTML5 and Clang tests. The former is 9%, but it is more than 30% stronger in ray tracing and machine learning projects. The position of "Android CPU boss" is solid.
The CPU frequency difference between the two is concentrated on the A710 core. The three A710 cores of Snapdragon 8 Gen 1 are 2.5GHz, and the Tianji 9000 is 2.85GHz, with a frequency difference of 14%. However, the difference between GeekBench's "multi-core single core" performance and GeekBench's "multi-core single core" performance is 19.3%, which is obviously larger than the 14% frequency difference. We speculate that the gap between the two is mainly due to the large cache of Tianji 9000. After all, L3 and SLC have a gap of 33% and 50% respectively, and sub projects with huge differences are often tests of buffer sensitivity.
What should be noted here is that the multi-core performance of the Snapdragon 8 Gen 1 mass production machine has made progress, from the first defeat to the Snapdragon 888+to the present small super. Other media teachers' Tianji 9000 engineering machines also reached 4317 points, and the test results of Geek Bay even reached 4400 points. The peak values on both sides have not yet been completely determined, which is only for reference.
In terms of GPU performance, we use GFXBench for comparison. Here, we introduce the Snapdragon 8 Gen 1 QRD (fan cooling) data unique to Teacher Feiwei as a reference. This is the strongest part of Snapdragon 8 Gen 1, which is 42% to 52% stronger than Snapdragon 888+. QUALCOMM can finally play back and forth with Apple A15, and the gap with Tianji 9000 is about 11.4%.
However, the progress of Tianji 9000 is more exaggerated, and its performance is far stronger than expected. In the high pressure Aztec test, it even slightly outperformed the "Youth A15" (6% stronger) of the 4-core GPU on iPhone 13, which is very close to the "A15 Max-Q" (4% weaker) of the iPad mini 6, while in the low pressure Manhattan and Manhattan 3.1, it is even closer.
In recent years, the GPUs of Android and iOS have made rapid progress, which is almost blindfolded. Tested in Manhattan 3.1, the performance of this generation of GPU is 4.2 times that of Snapdragon 835 (flagship in 2017); Tested in Manhattan, it is 4.4 times higher than Snapdragon 835. It is equivalent to an average annual performance increase of 43% to 45%, which is simply terrible (of course, power consumption is also different).
On rabbit side, have a ritual entertainment (Anthare is all about Tuyile, and the real running score software is Taobao) . Here we borrow the data of WHYLAB's Snapdragon 8 Gen 1 refrigerator, and Tianji 9000 is still passive cooling. The CPU and MEM (memory+flash memory) of Tianji 9000 are 13% and 12% stronger respectively, the UX (data security/data processing/strategy games/image processing and I/O performance and other projects) is 2% stronger, and only the GPU is 12% weaker, the total score is 102W. The GPU weight of Entertainment Rabbit is not right (manual funny. jpg).
The PCMark test is seldom used. The score of Tianji 9000 is 18162, and that of Snapdragon 8 Gen 1 is about 1.7W. There is no big difference between the two sides.
Qualcomm has never made an independent APU/NPU, resulting in a significant gap in AI testing. According to ETHZ AI benchmark V5 test, Tianji 9000 is about 1058K, and Snapdragon 8 Gen 1 is about 538K, with a gap of 96.7%.
Power consumption gap
The mobile platform needs battery power, and its size is limited, so it is impossible to rush like the PC platform regardless of power consumption. For one thing, the battery life cannot be stretched, and for another, the cooling conditions are not allowed. Mobile SoC itself is an art of balance. In addition to performance, flagship SoC should also focus on power consumption and heating. Two generations of fire dragons have taught us that the peak performance is as strong as a tiger, and the heating frequency is reduced by 250.
On Tianji 9000, MediaTek still emphasizes power consumption as usual. At the beginning of the conference, "global energy efficiency optimization technology" was highlighted. MediaTek said that it can cover different IP blocks such as CPU/GPU/APU/ISP, and provide different optimization schemes according to the light, medium and heavy loads of mobile phones. Web browsing, video viewing, standby and other light loads claimed to save 38% of the power compared with the flagship of the previous year; Power can be saved by 9% and 12% in heavy scenes such as photographing and video recording; Reduce power consumption by up to 25% in heavy loads such as games.
The APU of Tianji 9000 claims to be able to participate in the game, providing AI super score, AI-VRS variable rendering technology, CPU thread optimization, intelligent dynamic frame stabilization and other features. With HyperEngine5.0 intelligent control engine, AI-VRS can reduce power consumption by 15%; By analyzing the APP, the CPU operation is broken down into multiple threads, which can optimize 5% of the power consumption; Intelligent dynamic frame stabilization technology can save 9% power consumption. And how about the actual effect, we need to get the mass production machine for testing.
On the basis of basic energy consumption, AnandTech has previously conducted SPEC 2017 test on Snapdragon 8 Gen 1 QRD, and compared with the CPU of Snapdragon 888, it has improved the integer performance by 8% and the peak power consumption by 2%; Floating point increased by 19%, and floating point peak power consumption changed little. Almost all tests have higher instantaneous power consumption, and the peak value can reach 14W. AnandTech speculates that the reason is that L3 and SLC are called, which leads to a sharp increase in power consumption (there is no full cache, maybe not because of cutting, but because of heat...).
For Tianji 9000, we refer to the exclusive data of Mr. Feiwei: in the AndSPEC test, the X2 super large nuclear energy consumption ratio of Tianji 9000 is 18.54, which is 49% higher than the 12.43 of Snapdragon 8 Gen 1; The energy consumption ratio of A710 core is 22.25, which is nearly 40% higher than 15.94 of Snapdragon 8 Gen 1 A710.
You should know that the frequency of the A710 core of the Tianji 9000 is 14% higher, and the cache of the power consuming major L3 and SLC is 33% and 50% higher respectively... In the case of large cache, high frequency, and stronger performance, the energy consumption ratio of the Tianji 9000 is 40% to 49% ahead. There is a huge gap between the energy consumption ratio of the Tianji 9000 and the Snapdragon 8 Gen 1, at least two generations or more. TSMC Yes! Tianji 9000 Yes!
Summary and analysis
Even for engineering machines, the CPU performance of Tianji 9000 is already the strongest in Android, and it is still far away from the second place in Chang'an Street. The GPU has also been greatly improved, which is indeed the flagship level.
Compared with the Snapdragon 8 Gen 1 of "strong show weak U", the configuration of Tianji 9000 is obviously more balanced. Regardless of the daily use, CPU performance is more needed. Even in modern mobile games, the bottleneck is the CPU, not the GPU. The higher the energy consumption ratio of modern CPUs, the more obvious the benefits. Don't talk about next generation SoC Score running software In the original God, the current GPU is not satisfied with running, and the problem of energy consumption ratio makes the CPU slow down early, which is even worse.
The era of Xiaolong 835 and double baking 3.6W is gone forever, but MediaTek has always emphasized the right direction of power consumption control. No matter how large the heat dissipation area is, no matter how large the soaking plate is, the surface heat dissipation space of the mobile phone is so small. The power consumption of 10W has already exceeded the carrying capacity of the mobile phone. This is not a notebook, this is a mobile phone!!!
In terms of progress, the moto edge X30 of the first Snapdragon 8 Gen 1 has been sold. The first batch of Tianji 9000 terminals will be launched in the first quarter of 2022, OPPO、vivo、 Both Xiaomi and Glory officially announced the first launch or the first launch. This is interesting. How will mobile phone manufacturers arrange two flagship SoC models?
The mobile SoC market has not been so busy for a long time. It's a bit like AMD beat Intel before. It's also the leader in absolute performance and energy consumption ratio. There is no doubt that Tianji 9000 is the strongest SoC in the history of MediaTek, as well as the SoC with the strongest comprehensive Android platform and the highest energy consumption ratio.
Follow our Weibo @ Love Computer
Follow our WeChat official account: playphone
Of course, we also pay attention to our Bilibili account: love computer
