Today we are going to talk about a very important new product, Snapdragon 8Gen3. I believe everyone has been looking forward to it for a long time, right? We feel that this year's A17 Pro energy efficiency improvement is not as good as expected. This has made many of us digital enthusiasts place our hopes on Qualcomm's 8Gen3. After all, last year's 8Gen2 did bring us surprises . So can this time's 8Gen3 continue the bravery of 8Gen2? What about performance? We have now gotten the Xiaomi 14, the first mobile phone equipped with 8Gen3. Today, let’s use this much-anticipated new phone to show you the real performance of Snapdragon 8Gen3! There are quite a lot of interesting things about the Snapdragon 8Gen3 processor. It is the first chip to use the new version of the ARM v9.2 instruction set architecture. It is also Qualcomm’s first pure 64-bit mobile phone chip. Yes, we have finally completely eliminated it at the hardware level. Congratulations to you for being 32nd ! Although the big node of the process has not ushered in an update and is still built using TSMC’s 4nm , 8Gen3 still has some design innovations. Let’s start with the CPU and take a look at last year’s Snapdragon 8Gen2. Qualcomm broke the traditional 1 The +3+4 core configuration replaced a small core with a large core and changed it to 1+4+3. Even though ARM did not bring much innovation, Qualcomm still significantly improved the multi-core performance of 8Gen2 by adding large cores . Multi-core energy efficiency: With this year’s 8Gen3, Qualcomm saw the effectiveness of this trick and went a step further. This time, 8Gen3 replaced a large core. It is now 1+5+2 , which is a brand new X4 super large core + 5 new cores. In the end , the A720 only had two small cores of the A520. Then this time, the frequency of the X4 super core reached 3.3GHz. The large core of the A720 was divided into two groups. Three of them can actually reach the high frequency of 3.15GHz. The other two are limited to 3.15GHz. Below 2.96GHz, the small core of the A520 has also been increased to 2.27GHz, with a total of 6 large cores + such a radical peak frequency. I am very much looking forward to how strong its multi-core performance is. As we have already mentioned the X4 and A720, we should talk about it. Let’s talk about the CPU architecture. ARM’s new architecture this year has a lot more to talk about than last year. Especially the super-large core Cortex-X4 has changed a lot. Since the launch of the X1 core, the Cortex-X series cores have been getting wider and stronger. The gap between the core specifications of the A7xx series and the X4 series is getting wider and wider. In the X4 generation, its core size can even be said to be on par with Apple's A17 Pro large core . And you will see that the overall design ideas are very similar to the front end of the X4 . Unprecedented use of 10-width decoding, which is not only much wider than the X3, but even wider than the A17 Pro. The branch prediction has also been improved, and the corresponding ROB depth has also been increased from 320 of the X3 to 384. With the significant widening of the front end, the scale of the back end has also changed. The number of ALU units has increased a lot from 6 to 8 in the previous generation, which is on par with the size of the A17 Pro. In addition, the maximum supported L2 cache has been increased from 1M to 2M. Overall, X4 should bring considerable performance improvements. We ran SPEC. 2017 also seems to confirm this. Compared with last year's 3.2G X3, the integer performance of the 3.3G X4 increased by 18%, and the floating point was 8% stronger. Overall, it has exceeded double-digit IPC. The performance is indeed impressive , but it is so wide. It is difficult to control the power consumption. The integer power consumption of X4 has skyrocketed from 4.1W to 5.7W. The floating point power consumption has also increased from 5.4W to 6.7W . I think this single-core power consumption is really a bit embarrassing . It is said to be for The processors built for PC are not overrated. We still don’t know the results of its mid- and low-frequency energy efficiency. But judging from the high-frequency situation, the X4 core can be said to be high-performance at all costs. If you compare it with If compared with Apple, the power consumption is close to that of A17 Pro, and the performance is close to that of A15 ... If you use this term, doesn't it feel like it's not that bad? Anyway, the super-large core of ARM is following the same route as Apple, but there is still a slight gap in strength. Next, let’s take a look at the A720. In fact, from a micro-architecture perspective, the degree of evolution of the A720 is not as great as that of the X4. Compared to the A715, it is more Optimizations of microarchitectural details include improvements to branch prediction. The A720 branch prediction error penalty cycle has been reduced from 12 cycles to 11 cycles on the A715. In addition, the access latency of the L2 cache has also been reduced from 10 cycles to 9 cycles, as well as some other aspects. The overall design idea of minor improvements is still minor repairs to the A715. ARM itself has set the main direction of the A720 to reduce power consumption. Of course, this is exactly what mobile phones need. In SPEC 2017, we measured the integer of the A720 compared to the A715. In fact, the improvement in energy efficiency is limited. The biggest improvement this time is the floating point energy efficiency. The frequency of the 8Gen3 A720 is one step higher than the A715 of 8Gen2, but the power consumption is only more than 0.1W higher. At the same time, the floating point performance is 12% better than the A715. I say this A720 is quite competent. As for the A520 small core that only has two left now, the A5xx series, the most shameful series of the Arm family, has always maintained the tradition of sequential execution. The A520 is still no exception, so you don’t have to expect its performance at all , even to improve energy efficiency. It also sucked up the toothpaste! Its integer ALU units have shrunk from 3 in the previous generation to 2. Arm is confident that it can make up for the performance through optimization. However, we actually measured the performance of the A520 in SPEC 2017 as much as the frequency exceeded. IPC can be said to be the original It's hard to say there has been any improvement in terms of energy efficiency. Anyway, this small core is still familiar to us. No wonder it has been reduced. Generally speaking, the changes in the CPU part of 8Gen3 are still quite big. First, the comprehensive 64-bitization is completely abandoned . The second burden of 32-bit is the re-adjustment of the core layout. The weak small cores are further reduced to two, giving more positions to large cores and super large cores. The third is the upgrade of ARM's new architecture , including the crazy stacking of X4 and A720. The energy efficiency upgrade should be said that this CPU is quite interesting. In addition to the CPU, I think the GPU of 8Gen3 is the focus of everyone's attention. This time Qualcomm did not upgrade the GPU architecture. It still belongs to the Adreno 700 series , but 8Gen3 was already very good in the previous generation. On the basis of the huge GPU, the core scale has been increased again and an extra set of CUs has been added . This GPU not only supports hardware ray tracing, but also supports grid shading characteristics. It has evened the A17 Pro. It can be said that this is really a PC-level device. As for the GPU , considering that Apple’s A17 Pro GPU can’t even beat 8Gen2, I’m curious to see how far ahead this year’s 8Gen3 GPU can be. Including this time, the NPU’s computing power has doubled and can even run offline. We will actually verify the large language model in the future . The hard-core architecture analysis is almost done. It’s time to take out the most popular arena part of the audience. Whether it is a mule or a horse, you will know after a walk . Let’s first look at the performance of the CPU. In terms of peak performance in GB5, 8Gen3 is indeed very impressive. The single-core performance is close to 1700 points , which is very close to the level of Apple’s A15! The multi-core performance has soared to 6782 points, which is even higher than the A17 Pro. Wow, Qualcomm actually beat Apple in CPU multi-core. This is really unexpected. This stack-core gameplay is simply tried and tested. Of course, the peak performance is also I believe everyone here is more concerned about energy efficiency , so let’s release the most important energy efficiency curve! Since we have not yet obtained ROOT permissions , the energy efficiency of 8Gen3 is still measured at only two points this time. However, it is enough to show that the peak energy efficiency of 8Gen3 multi-core is very impressive. Although the motherboard power consumption is only 1W higher than that of 8Gen2, the performance has greatly increased. Its peak energy efficiency even beats Apple's A17 Pro! After who knows how many years, the Android camp has finally become king in terms of CPU multi-core energy efficiency. This is really not easy. This stack core strategy is very effective in the energy efficiency curve of GB5. However, the stack core is of little use or even less for low-frequency energy efficiency. The core is more beneficial to low-frequency energy efficiency, so the low-frequency energy efficiency of 8Gen3 here is generally not much improved compared to 8Gen2. However, this time 8Gen3 or this generation of ARM architecture has a more important innovation , which can partially turn off L3 or SLC . In this way, there is a new way to play in scheduling. We can turn off a few cores and then turn off part of the cache to obtain better low-frequency energy efficiency. I don’t know if it is actually good for manufacturers to implement and schedule, but this method is theoretically It is very useful for low frequencies. After we have ROOT permissions, we will also do further tests. In addition to GB5, we also did the energy efficiency curve of GB6. Both tests are worth watching because they have different focuses on GB5 . It has multiple cores running multiple loads separately, which can maximize all the power of the CPU. In GB6, multiple cores work together to run one load. This is more in line with the actual working method of mobile phones running various apps. In addition, GB5 will be more The runtime results of multiple calls to the system will be more affected by the OS of each manufacturer. GB6 rarely calls system libraries. It uses the native Clang16 compiler. The results of the same chip from different manufacturers will be closer, which can reduce the manufacturer's special adjustments for running scores. We will release these two tests of system impact for everyone to see. Okay, in GB6, Apple will have an advantage over GB5 . After all, for multi-core processors, especially Qualcomm’s stack-core method, the more cores there are, the more difficult it is. Using all the cores to run a load will definitely result in greater performance loss. However, 8Gen3 does not falter even in Apple's dominant GB6. Although the peak power consumption is much higher than that of 8Gen2 , the performance is also greatly improved. Finally , looking at the peak energy efficiency, 8Gen3 is actually quite good. At least it can beat the A16. In fact, the improvement in mid- and high-frequency CPUs from 8Gen2 to 8Gen3 is much greater than from 8+ to 8Gen2. Considering that the process has not even improved in the past few years, it is so The substantial CPU progress is quite remarkable in my opinion. To sum up, in terms of CPU architecture, the ARM public version is still trying to catch up with the Apple If you squeeze the toothpaste like this, you will be caught up sooner or later. In terms of multi-core, thanks to Qualcomm's stacking and the good performance of the A720, the energy efficiency of 8Gen3 is already very impressive. Compared with 8Gen2, its mid- and high-frequency energy efficiency has been greatly improved . It is even as good as the A17 Pro. I think the 8Gen3 CPU has significantly exceeded my expectations because of its relatively high capital. The CPU is already so powerful. Does Qualcomm’s traditional strength, the GPU, bring any surprises to us this time? We first ran GFXBench. 8Gen3 improved the performance by 20% based on the already strong GPU 8Gen2. Now the GPU is completely different from the A17 Pro. Let’s look at the energy efficiency. 8Gen3 does not use a lot of energy like Apple and Fa Ge. Power consumption is exchanged for GPU performance. The GPU of 8Gen3 is only 8W more when running at full speed. The power consumption of the motherboard is less than 9W. The power consumption is not much higher than that of the 8Gen2 model, but the performance is significantly improved. This GPU energy efficiency performance can be said to be the undisputed king of the mobile terminal . Let’s take a look at the performance of 3DMark Wlidlife Extreme. The focus of the two tests is also different. GFX’s Aztec is closer to the rendering pipeline of traditional mobile games. Its API version is also slightly older. WLE uses many advanced features. The rendering pipeline is relatively close to modern 3A masterpieces. These two different working conditions are worth taking a look at. Apple’s GPU is more dominant than Aztec in WLE. The performance of A17 Pro can be stronger than 8Gen2 , but it is not enough to stand up to 8Gen3. It’s so powerful. 8Gen3 easily defeated the A17 Pro , and its power consumption is much lower than that of the A17 Pro. It can be said to be a win. But do you know that this is not the limit of 8Gen3. Through special methods, we have actually surpassed the 8Gen3. The GPU was further overclocked from 770MHz to 903MHz. At this time, WLE ran an incredible score of 5170 points, which was nearly 38% faster than 8Gen2! Of course, the power consumption will definitely be higher at such a high frequency, reaching 12.6W, but I can already imagine what kind of work can be done with this terrifying extreme performance. This is really exciting. Let’s look at the stability of 8Gen3 After all, the power consumption is not high , so the GPU stability of Xiaomi 14 is still very good. Even after the frequency is reduced, it is stronger than 8Gen2. This GPU performance is quite satisfactory. Due to the cross-platform feature of 3DMark, we can even use the 8Gen3 GPU. Comparing PC performance, it is already equivalent to the performance of the low-power version of 1050Ti and AMD's previous generation 680M core display. The overclocked 8Gen3 is close to the notebook standard version of 1050TI and AMD's 780M core display , even if the heat dissipation is included. After frequency , it can still reach the notebook 1050-level performance, which is already equivalent to the flagship PC graphics card from 12 years ago. In addition to the traditional rasterization test, the 8Gen3 GPU also supports hardware ray tracing, so we will also run 3DMark. Solar Bay to see how its ray tracing performance is? As a result, we were all shocked that the light tracing performance of 8Gen3 was improved by 43% compared to 8Gen2 at a silent frequency of 770MHz ! Even the Apple A17 Pro, which is better at light tracing, has to be defeated by 8Gen3 , and the power consumption of 8Gen3 is also lower. If we further overclock to 903MHz, this GPU can run a light tracing score of more than 8700 points, which is far behind . Open any opponent , so this GPU is indeed the most powerful mobile GPU in all senses. The theoretical performance and energy efficiency of 8Gen3 look good. Then I believe everyone must be very curious about 8Gen3’s mobile gaming. What kind of experience is it? Not flowing smoothly? Are you having a fever? Let’s quickly serve up the famous roasting machine software-Genshin Impact! The resolution of Xiaomi Mi 14 in Genshin Impact this time is very interesting. The default standard mode provides a normal 720P resolution. If you choose high quality, you will see an unprecedented 1220P resolution that fills the entire Xiaomi Mi 14 point-to-point. Just look at the screen. The definition is almost PC-level gaming experience. No matter how smart or advanced the super-sampling technology is , it can never compare to the effect of violently increasing the physical resolution. It’s a pity that in this high-quality In mode, Xiaomi limits the frame rate to 30 frames. In fact, Genshin Impact running 1220P 30 frames on the 8Gen3 is so easy. We measured that the power consumption of the whole machine is only 3.9W, which seems perfect. You can let go of 60 frames and see. Maybe it’s true. If it can run well, let’s take a look at the game performance of Xiaomi Mi 14 under normal 720P . It is still the familiar Xumi Cheng Yelan running chart. The performance of the previous generation Xiaomi Mi 13 can barely conquer the 720P Genshin Impact with an average of 57.5 frames in 30 minutes. Already good So what is the performance of Xiaomi Mi 14 this time? 30 minute average... 59.2 frames! 8Gen3 can run the game very smoothly throughout the process with very few frame drops. At the same time, the power consumption of Xiaomi Mi 14 is a little lower than that of Xiaomi Mi 13. It only needs 5.7W to achieve this performance. It can be said that this time the CPU energy efficiency is better. The upgrade means that Xiaomi 14 no longer has to be as meticulous about performance as Xiaomi 13, trying to achieve a balance between frame rate and power consumption. Xiaomi 14 can more easily approach full frames without worrying too much about power consumption. Looking at the CPU scheduling, we will find that Xiaomi allows three One large core runs above 2.2GHz. The remaining two large cores and the ultra-large core are all operated at relatively low frequencies. It should be said that 8Gen3 only used a few successes to run Genshin Impact but still left a lot of margin . So Xiaomi Mi 14 How does the performance compare with the iPhone? Since Genshin Impact has made a lot of special optimizations for Apple's Metal API , Apple is still better in terms of energy efficiency. The power consumption of A17Pro is much lower than that of 8Gen3 when running 60 frames, but it cannot withstand the heat dissipation of the iPhone. It is really terrible, especially when playing. After a long time, even if the A17 Pro is still close to full frame , you can still feel the small frame drops caused by the frequency reduction. We also tested the surface temperature of the body after playing games for 30 minutes. Since our Xiaomi Mi 14 is a plain leather version, the surface temperature It is higher than the Xiaomi Mi 13 tested before, with a maximum temperature of 45.4 degrees, which is still well controlled. In contrast, the iPhone 15 Pro, even though the 17.1 system has been updated to improve heating, it still reaches a body temperature of 47 degrees. It was once known for its low heat generation. The iPhone has become the hotter one now. It’s really a turn of events. Now that we are talking about heat dissipation , shouldn’t we sacrifice the big devil? The Wi-Fi environment of 25 degrees is too easy. It is really awesome to conquer 30 degrees 5G. It seems that this environment is still a bit harsh for Xiaomi Mi 14. After running for 8 minutes, it will start to reduce the frequency. The final average frame rate is 45.7 frames . It’s a slight improvement compared to Xiaomi Mi 13. Of course, it’s much better than the iPhone. The body temperature in the Devil’s test further soared to 47.5 degrees. It seems that this is the limit for a small-screen machine. I’m a little curious about the machine. Will the larger Xiaomi Mi 14Pro perform better in the devil test? Having said that, Genshin Impact is no longer the most stressful game on the mobile terminal . Who is more demanding on the configuration may depend on Honkai Impact: Star Dome Railway. This game puts a great test on both the CPU and the GPU . 8Gen3 represents my generation of CPU. If the GPU and GPU are upgraded together, are you worried that you will fail? Sure enough, Xiaomi Mi 14 achieved an unparalleled average score of 55 frames while playing Bengtie! It is far stronger than Xiaomi 13 and iPhone 15 Pro. At the same time, the power consumption can still be controlled at the level of 6W. With the help of 8Gen3, non-gaming mobile phones can finally run high-definition games smoothly. It can be said to be gratifying. If MiHoYo's games are too feverish, we also tested Honor of Kings with a relatively lower load. The results of Xiaomi Mi 14 also brought us surprises. The power consumption of this generation of Honor of Kings at 60 frames has been It has been reduced to 2.3W for the whole machine. Compared with Xiaomi Mi 13, it has been reduced a lot. Mid-load games are more energy-saving. In 120 frame mode, Xiaomi Mi 14 has reduced the king's power consumption to less than 3W for the first time ! The power consumption of King 120 is close to that of King 60. This is really powerful! Finally, we also tested the heat dissipation capability of the Xiaomi Mi 14. This generation of Xiaomi has made a lot of improvements to heat dissipation. It uses a new annular cold pump to separate gas and liquid in the vapor chamber, which further improves the heat dissipation efficiency. We actually measured it . The heat dissipation of Xiaomi Mi 14 is better than that of the previous generation Xiaomi Mi 13 Pro, which can handle up to 6.2W of continuous heat. As a small-screen mobile phone, this heat dissipation design is quite good . Of course, it is limited by the lack of ROOT. Permissions We still have a lot of things that we can’t show you this time, but it’s enough to give 8Gen3 and Xiaomi 14 their personality. It can be said that the performance of 8Gen3 this time really exceeded our expectations. I originally thought that 8Gen2 was good enough and 8Gen3 was a little crowded. Toothpaste will do. After all, Qualcomm will develop its own CPU next year. This last generation of ARM public version does not need to work so hard. As a result, 8Gen3 has brought considerable improvements without reservation. Both the CPU and GPU have continued to improve in energy efficiency. It feels like this generation of Xiaomi 14 It’s going to be another year. To be honest, after testing the A17 Pro, I don’t have much expectations for this year’s new products. However, 8Gen3 now makes me look forward to next year’s 8Gen4. Qualcomm has launched its own products two days ago. The Orion core developed by the company is used in notebooks. Judging from the official running scores, the single-core is even much better than the M2 Max. Then next year's 8Gen4, the only single-core energy efficiency that lags behind, may be able to overtake Apple. By then, the mobile phone industry will change. What kind of pattern will it become? What can Xiaomi Mi 15 do? Wow, these are exciting just thinking about them! I wonder if everyone is satisfied with the performance of 8Gen3? We will do more tests on Snapdragon 8 Gen3 and new Xiaomi phones in the future, including the whole life. Students who are looking forward to it must continue to pay attention to our Geek Bay channel. If you think this program is well done, be sure to reward us. You can also like, collect and repost for free. You can also come to our Taobao store in Geek Bay to buy some peripherals to support us. I am your brother Piao . See you next time.
