As early as this year's COMPUTEX 2024, AMD brought players preview information about the Zen 5 architecture and Ryzen 9000 series processors. It also officially released the third-generation Ryzen AI processor-Ryzen AI 300 series mobile platform. In addition, it also announced the AMD X870/X870E chip for the 800 series motherboard, which supports USB 4.0 and PCIe 5.0 standards to bring higher performance and better compatibility experience.
Recently, AMD held a technical sharing meeting to provide a more comprehensive technical analysis of these upcoming hardware. At the event, we also saw the latest Ryzen R9-9950X processor, which has really great performance! The wait has finally come! What are the improvements in the new processor series and how is the performance? What are the improvements in the Zen5 structure? Don't worry, let's analyze them one by one.
Zen 5 Architecture
AMD calls the Zen 5 architecture a “huge leap forward,” and compared to Zen 4, the Zen 5 architecture is a much larger upgrade, comparable to the changes in the Zen 3 era.The changes in the Zen 5 architecture compared to the previous generation mainly focus on optimizing computing units, improving bandwidth, enhancing instruction set throughput and AI performance.
Let's analyze it in more detail. First of all, the Zen 5 architecture has improvements in instruction fetching and decoding.The processor uses a 32KB first-level instruction cache, which significantly improves the processing efficiency of the instruction stream through 8-way parallel processing and 32-byte double data acquisition rate.
In addition, more advanced branch prediction technology has been introduced to guess the instructions that need to be executed next, so that the processor can prepare in advance and greatly improve efficiency.
In terms of integer execution, the processor has made a number of optimizations. The L1 instruction cache capacity remains unchanged at 32KB, but by increasing the width to 8, the parallelism of integer operations is improved. In addition, the ALU scheduler contains 6 ALU units and 3 multipliers, which not only improves the integer computing capability, but also enhances the flexibility of the execution unit.
The processor has also become more efficient in loading and storing data. It has a 48KB 12-way parallel design of the fast data cache area, which greatly speeds up the loading of data. At the same time, by increasing the data bandwidth and improving the prefetch algorithm, the processor can more effectively predict and load the data to be used, reducing the waiting time.
Finally, the processor has become faster at handling complex floating point and vector operations. As we all know, Intel has given up support for the AVX-512 instruction set due to its big and small core design, but AMD is different. The Zen 5 architecture also supports the AVX-512 instruction set, equipped with a full 512-bit data path and 8-wide scheduling capabilities. This also means that the processor can handle more complex mathematical operations at the same time.
In addition to the significant improvement in architecture, Zen5 has an average IPC performance improvement of about 16% compared to Zen4, which is quite impressive. The mathematical performance of the acceleration unit has been improved even more than Zen 4, with a single core improving by up to 32% in machine learning, and a 35% performance improvement in the AES-XTS encryption algorithm.
In general, the improvements of Zen 5 architecture are mainly reflected in data bandwidth, scheduling execution unit optimization, more accurate branch prediction and decoding cache. In addition, thanks to the use of TSMC's 3nm and 4nm leading processes, Zen 5 architecture has lower power consumption but higher performance. In the future, Zen 5 architecture will also be applied to multiple types of products.
Currently, the Ryzen 9000 series processors and Ryzen AI 300 series processors use the Zen 5 architecture, and the EYPC processor using the Zen 5 architecture will be launched in the second half of 2024, with a maximum of 192 cores and 384 threads. At the same time, the Zen 5c architecture will be launched in the future, both of which are built with TSMC 3nm and 4nm. As for the Zen 6 or Zen 6c architecture, AMD has not yet announced when it will be released.
Ryzen 9000 series desktop processors
There are four Ryzen 9000 series processors released this time, namely the flagship Ryzen 9 9950X, Ryzen 9 9900X, Ryzen 7 9700X and Ryzen 5 9600X. The detailed specifications are as follows, and I will not go into details here. It can be seen that the new Ryzen 9000 series has stronger performance while the TDP is also lower, and the energy efficiency ratio will be extremely amazing. It can be seen that Zen 5 is indeed full of black technology.
At COMPUTEX 2024, we showed you the performance of the Ryzen 9 9950X. In terms of performance, the Ryzen 9 9950X is far ahead. According to the data provided by AMD, the Ryzen 9 9950X is about 30% ahead of Intel's i9-14900K in the field of productivity creation, and can lead up to 56% in Blender. The performance of games is also good. For example, in Cyberpunk 2077, the Ryzen 9 9950X has a 13% lead. In Horizon: Zero Dawn, the performance can surpass the Intel i9-14900K by more than 23%.
At this technical sharing session, AMD brought us the performance of the remaining three processors. Ryzen 9 9900X can easily beat i9-14900K. In terms of productivity, it is basically more than 10% ahead of competitors, and the lead in the Handbrake project is an exaggerated 41%. The gaming field has also caught up. In Horizon: Zero Dawn, the performance can surpass Intel i9-14900K by more than 22%. Even in the online game Dota2, the performance is 7% stronger than the competitor. In the 6 games tested, the average performance of Ryzen 9 9900X is 14% ahead.
The high-end Ryzen 7 9700X is naturally competing with the i7-14700K. However, the i7-14700K does not seem to be a match for the Ryzen 7 9700X. In the GeekBench single-core test, the Ryzen 7 9700X is 17% higher than the i7-14700K. In the Puget Photoshop test, which also focuses on single-core performance, it is also 19% ahead. Looking at the game performance, in Horizon: Zero Dawn, the lead is 31%, and in the more stressful Cyberpunk 2077, the Ryzen 7 9700X leads by 12%, which is really amazing.
Finally, the mainstream players are paying attention to the Ryzen 5 9600X. Its productivity performance is far better than the i5-14600K. In the GeekBench single-core test, Puget Photoshop and 7-Zip tests, the performance lead is about 20%. In Handbrake, it is outrageous, with a 94% lead. In the 6 games tested, the Ryzen 5 9700X is also in the lead, with an average performance lead of more than 14%.
Outstanding performance is only one of the advantages of the Ryzen 9000 series processors, and energy efficiency is the highlight. The thermal resistance of the Ryzen 9000 series processors has increased by 15%, which means that at the same TDP, the temperature of the new processor can be reduced by about 7°C.
According to the information released by AMD, the gaming performance of the 65W Ryzen 7 9700X is much better than that of the 105W Ryzen 7 5800X3D, which is also known as the “game god U”. The actual test shows that the performance of the Ryzen 7 9700X in Warhammer 3 is 30% higher than that of the 5800X3D, and the average performance is about 12% higher.
As we all know from the specification configuration diagram at the beginning, the new Ryzen 9000 series processors, except for the flagship Ryzen 9 9950X, have reduced TDP, but the performance has increased instead of decreased. Lower power consumption achieves stronger performance. AMD Yes!
In addition, the Ryzen 9000 series processors have also been improved in overclocking, and the memory supports ultra-high frequency of DDR5-8000MHz.
The improvement of PBO is even greater. Since the default TDP of the Ryzen 9000 series processors is relatively low, the improvement is huge after turning on PBO. According to the information provided by AMD, the performance of the Ryzen 7 9700X is improved by up to 15% after turning on PBO, while the Ryzen 9 9900X and Ryzen 5 9600X also have a 6% performance improvement after turning on PBO.
In addition, AMD also performed manual overclocking on site, using the flagship Ryzen 9 9950X.
AMD staff overclocked the processor to 6.3GHz under liquid nitrogen! The multi-core score of CineBench R23 directly reached 52197!
800 series motherboard
The motherboard that matches the Ryzen 9000 series is the 800 series. Starting from the Ryzen 7000 series, AMD began to switch to the AM5 interface. The previous AM4 interface supported 145 CPU or APU processors, and the latest AM5 interface currently has 38 CPU or APU applications. AMD promised that the AM5 interface will be used until 2027. Players who are ready to replace new equipment can rest assured!
At COMPUTEX 2024, AMD only introduced the top-level X870E and X870 motherboards. Now AMD has announced the complete chipset for the 800 series motherboards, including X870E, X870, B850 and the brand new B840.
The top-level X870/X870E supports USB 4.0, graphics card and SSD interfaces support PCle 5.0, and support memory with higher operating frequency. The B850 is similar to the current B650/B650E, with two versions, PCIe 5.0 and PCIe 4.0, supporting USB 3.2 20Gbps, and both CPU and memory can be overclocked.
The latest B840 is a bit like a scaled-down version of B850, it does not support PCle 5.0, only supports PCle 4.0, and does not support processor overclocking. As for H810, although some motherboard manufacturers have leaked some information, it is not listed in AMD's information. If there is any update later, we will synchronize it as soon as possible.
Regarding the launch time of the motherboard, it is said that the high-end models X870E and X870 series motherboards in the 800 series motherboards will be launched in late September, that is, they will be released two months after the release of the Ryzen 9000 series processors. As for the release time of the B850 and B840 motherboards, it will most likely be even later.
Ryzen AI 300 Series Mobile Processors
Let’s continue to look at mobile processors. The Ryzen AI 300 series launched this time is also the third-generation AMD Ryzen AI. It adopts the Zen 5 architecture, has a built-in XDNA2 NPU, and the GPU is upgraded to an RDNA 3.5 graphics card. It is known as the most powerful processor for the next-generation AI PC/Copilot+ PC.
Currently, the Ryzen AI 300 series has widely supported the Windows ecosystem, with more than 100,000 games, 35 million apps, and 600 million devices compatible.
At COMPUTEX 2024, AMD announced that more than 100 OEM products equipped with AMD Ryzen AI 300 series processors will be launched one after another.
In addition to the common notebook form factor, the Ryzen AI 300 series can also be presented in other forms. Currently, AMD is continuously advancing end-to-end AI infrastructure, covering terminal products such as cloud, high-performance computing platforms, enterprise-level, embedded and personal computers, so that AI can better provide powerful power for billions of users in their daily lives and work.
Having said so much, let's talk about the hardware configuration. First of all, the Ryzen AI 300 series processors are also built on the Zen 5 architecture, with a maximum of 12 cores and 24 threads in the CPU cores, and further improved multi-threaded performance. The Ryzen AI 300 series GPU is built on the RDNA 3.5 architecture, with 16 computing units and further enhanced graphics performance.
The first models are Ryzen AI 9 HX 370 and Ryzen AI 9 365, both of which are high-end models. The top flagship Ryzen AI 9 HX 370 processor has 12 cores, 24 threads, 36MB L2+L3 cache, a maximum frequency of 5.1GHz, and the GPU is upgraded to Radeon 890M. The number of CU units has also increased from 12 to 16, and the NPU AI computing power reaches 50TOPS.
Ryzen AI 9 365 has 10 cores and 20 threads, 10MB L2 cache, 24MB L3 cache, and a maximum frequency of 5.0GHz. The GPU is Radeon 880M, with 12 CU units, and the NPU computing power remains the same at 50TOPS.
In terms of performance, the Ryzen AI 300 series is far ahead. In the field of productivity, the Ryzen AI 9 HX 370 leads in all aspects. The application startup speed is 1.07 times that of the Intel Core Ultra 9 185H. If it is web browsing, the performance of this processor is even stronger, reaching 1.3 times that of the latter.
In terms of video editing, multi-threaded performance, and 3D rendering performance, the Ryzen AI 9 HX 370 is even more powerful than the Core Ultra 185H processor. In the Blender BMW rendering project alone, the performance is 3.8 times higher.
In terms of game acceleration, the new RDNA 3.5 architecture iGPU brings stronger graphics performance. Compared with the Core Ultra 185H processor, Ryzen AI 9 HX 370 performs very well in games such as “Tomb Raider: Shadow”, “Assassin's Creed: Vision”, “Far Cry 6”, “F1 2022”, “Borderlands 3”, “Cyberpunk 2077”, “GTA5”, etc., with an average performance lead of more than 20-30%.
AMD also cooperates with many manufacturers to comprehensively expand the application scope of AI technology, penetrate into all fields, gradually build a complete AI environment, and create an unprecedented Windows AI PC ecosystem.
For example, AMD and Microsoft have cooperated deeply, and the Ryzen AI 300 series will bring new AI features to Copilot+, such as real-time subtitles, literary images, etc. In the future, more OEM manufacturers will realize more AI experiences based on the Ryzen AI 300 series.
At the same time, AMD also announced at the meeting that AMUSE 2.0 will be launched soon. Amuse is a professional and intuitive Windows UI. With the help of the ONNX (Open Neural Network Exchange) platform, it can realize various AI-powered operations such as text-to-image, image-to-image, redrawing, etc., and has model management functions.
In general, the Ryzen AI 300 series will bring you a complete ecosystem experience, rich and diverse product forms, powerful and extreme AI applications, excellent performance and ultra-long battery life.
AMD RDNA 3.5 architecture
We have already briefly seen the performance of the RDNA 3.5 architecture from the Ryzen AI 300 series processors. Below we will introduce the relevant information of the RDNA 3.5 architecture in detail.
The RDNA 3.5 architecture has three major advantages over its predecessor. The first is that the energy efficiency has been significantly improved. The second is that memory management has been optimized, which makes the frequency of memory access lower but can achieve higher efficiency. The last point is smarter GPU power management, which can save power consumption and achieve longer battery life.
Specifically, AMD has improved the performance and power efficiency of GPUs on mobile devices through improved schedulers and power management. This allows the GPU to maintain high performance while extending battery life. In addition, AMD has optimized the most commonly used texture sampling operations, achieving double the sampling rate, which helps improve the graphics quality and performance of games.
The vector instruction set architecture (ISA) has also been enhanced to increase the interpolation and comparison rates, thereby improving the efficiency of graphics rendering. Dramatic improvements have also been made in memory management, with AMD optimizing graphics data structures, including L2 cache and SoC (System on Chip) bus, to reduce memory access times, improve data compression technology and workload optimization.
Specifically in terms of practical applications, according to AMD, the new RDNA 3.5 architecture can improve performance per watt by more than 32% under the same TDP compared to the previous generation Ryzen CPU, and the energy efficiency ratio is quite amazing.
It has to be said that AMD does have a deep accumulation in graphics processing technology, and the RDNA 3.5 architecture will undoubtedly bring users a smoother and more efficient user experience.
AMD XDNA 2 Architecture
The last part is about the XDNA 2 architecture, which is mainly related to the NPU of the Ryzen AI 300 series. According to our introduction above, the Ryzen AI 300 series integrates the most advanced Zen 5 architecture CPU core, and also has an XDNA 2 architecture NPU and an RDNA 3.5 architecture GPU. With the support of the three core processing units, it brings powerful computing and AI computing capabilities.
AMD's XDNA architecture is completely different from traditional multi-core processors. As can be seen from the architecture diagram below, the XDNA architecture has a more flexible computing unit and memory hierarchy. Compared with traditional multi-core processors, the XDNA architecture can dynamically adjust computing resources and memory access according to the specific needs of AI workloads, thereby achieving better performance and lower latency.
In addition, the XDNA architecture has highly programmable interconnection and partitioning capabilities, which can adapt to various complex AI workloads and provide strong support for real-time video processing and content creation. Memory has also been optimized. The AMD XDNA architecture can effectively reduce the demand for memory bandwidth and eliminate cache misses, ensuring efficient multi-tasking and stable real-time performance.
The XDNA 2 architecture has made significant improvements in design compared to XDNA 1. First, the number of AI engine Tiles has increased from 20 to 32, so the AI computing power has increased fivefold, from 10 TOPS to 50 TOPS. In addition, the performance improvement of XDNA 2 also comes from a number of innovations, such as: expanding memory capacity, introducing efficient data transmission technology, and optimizing memory management.
Compared with the Ryzen 7040 series, the AMD XDNA 2 architecture provides up to 5 times the AI computing power. And at the same performance, the energy efficiency ratio is twice that of the previous generation, which means that its battery life is also greatly improved.
The Ryzen AI 300 series processors currently using the XDNA 2 architecture NPU have unparalleled advantages in terms of AI performance. Even the latest Apple M4 is not its opponent. With a computing power of 50 TOPS, it is definitely the most powerful NPU currently available.
In addition, the XDNA 2 architecture also innovatively uses the Block FP16 data model, which is a new data type designed specifically for AI. Unlike other data models, it does not require quantization. Therefore, while ensuring calculation accuracy, it can also greatly improve performance. Its appearance can be said to represent a major advancement in AI data processing.
According to the data provided by AMD, Block FP16 can provide higher throughput and smaller model size while maintaining high accuracy comparable to FP32, thus effectively reducing the storage requirements of the model.
In addition, the flexibility of Block FP16 allows it to serve as a direct replacement for FP32 with almost no loss of accuracy. Software vendors are fully able to seamlessly migrate existing models to the AMD platform. The efficiency and compatibility of Block FP16 will drive innovation and breakthroughs in AI technology in a wider range of application scenarios.
Currently, Block FP16 can be applied to the Stable Diffusion XL Turbo model. In addition, thanks to the high accuracy of Block FP16, in the response speed test of the Llamav2 7B model, the Ryzen 300 AI series with XDNA 2 NPU is 5 times faster than the Intel Core Ultra 7 155H. The excellent AI capabilities further prove the advanced technology and performance advantages of the XDNA 2 architecture in the field of AI.
Of course, AMD has not only innovated in hardware, it has also built a broad AI PC platform ecosystem. By working with more than 150 AI-driven independent software vendors (ISVs), AMD is reshaping multiple fields such as gaming, entertainment, personal AI assistants, innovative creation and editing, and enterprise productivity.
In general, the launch of the AMD XDNA 2 architecture not only demonstrates AMD's leading position in AI computing, but also provides strong hardware support for future AI applications, heralding a bigger step forward in the intelligence of personal computers.
With the continuous advancement of AI technology, the XDNA 2 architecture is expected to bring revolutionary experiences in multiple fields such as games, entertainment, and enterprise productivity.
Conclusion
With the launch of AMD Zen 5 architecture, Ryzen 9000 series processors, and Ryzen AI 300 series processors, we have once again witnessed a huge leap in personal computing performance.
These innovations not only achieve breakthroughs in traditional CPU performance, but also set new benchmarks in AI computing, graphics processing, and energy efficiency. In addition, the integration of AMD's XDNA 2 NPU and RDNA 3.5 GPU architectures heralds the arrival of a new era in which AI will be everywhere, bringing users a more personalized and intelligent experience.
By collaborating with many independent software vendors, AMD is building a powerful AI PC platform ecosystem that will not only reshape the gaming and entertainment industries, but will also trigger changes in enterprise productivity and personal creation.
In the future, with the continuous development and popularization of AI technology, we have reason to believe that AMD will definitely bring more convenience and surprises to our work and life.