The ThunderX3 processor of the ARM chip explains the king of the ARM chip in detail:
Recently, Amazon released the 64-core Graviton2 and Ampere of the ARM architecture.
now, Mavell has presented the third generation ARM chip "ThunderX3".
many people may not be familiar with Marvell's name, but this semiconductor company with a long history is always by your side, such as mechanical hard disk master, most of which are written by Marvell, and SSD master also has a wealth of products.
Marvell introduction:
1. Founded in 1995, it is headquartered in Santa Clara, California, USA.
2. There are R&D centers in the United States, Israeli, Indian, German and China.
3. There are more than 5, employees worldwide, more than 1, patents worldwide, and the revenue in fiscal year 22 is 2.7 billion US dollars.
4. The core business covers three major areas: storage, network and computing, especially with more than 15 years of research and development experience in high-performance multi-core CPU.
5. Mavell owns a complete product line of CPU processors, including four series: Armada, OcteonTX, OcteonFusion and ThunderX.
6. Especially, it has quietly become the world's largest supplier of ARM server processors, and its application scope and ecological support are unique in the world, with cumulative shipments exceeding 1 million.
7. in 217, Marvell also acquired Cavium (kaiwei semiconductor), an American factory-free semiconductor company, for about $6 billion, further enhancing the design capability of the ARM platform.
x86 architecture and ARM architecture:
For a long time, the fields of servers and data centers have been dominated by x86CPU architecture. However, with the diversification of market demand and application load, and with the continuous evolution of cloud computing, edge computing and high-performance computing, RISC-V, ARM and other architectures have also ushered in new opportunities, especially ARM. Many giants have participated in it, and ecological construction is also in full swing.
now that we have mature x86, why should we vigorously promote ARM? What attracts the whole industry?
Marvell believes that ARM has shown better quality in architecture, technology, performance, energy efficiency, ecology and other aspects. Although it will not and cannot completely replace x86, it has obviously surpassed x86 in some specific fields, especially cloud computing and high-performance computing.
in terms of architecture, x86 has always been dominated by one architecture. From the server data center to the desktop notebook consumer level, different products are derived from the same architecture, which lacks flexibility, especially in parallel processing, and the scale is getting bigger and bigger, which also leads to the cost and power consumption becoming more and more unable to meet the demand.
in terms of technology, Intel has lost its leading position. Xeon series is still at 14nm. Although AMD has reached 7nm, it uses chiplet multi-chip design in architecture, which leads to obvious lack of memory bandwidth and memory delay. ARM has followed TSMC all the way to 7nm and will soon enter 5nm.
In terms of performance and energy efficiency, Marvell thinks that the single-thread performance of ARM architecture is first-class, and the multi-thread and platform performance are far ahead. memory bandwidth and delay are equally excellent, and it has low power consumption and high energy efficiency.
in terms of ecology, x86 has a long history, but it is also a heavy burden. For example, considering various histories and even consumer compatibility, ARM is not too much drag, and the architecture is directly born for server optimization.
At the same time, the hardware and software industries are increasingly supporting ARM. For example, Marvell HunderX 2 has been deployed by more than 2 end customers in the large-scale cloud computing and high-performance computing markets, including Microsoft Azure, HPE, Cray, Atos, Los Alamos National Laboratory, Sandia National Laboratory, Lawrence Livermore National Laboratory, Oak Ridge National Laboratory, Bristol University, Leicester University and so on.
even some hardware and software technologies of Intel also support the ARM platform, and so do NVIDIA and AMDGPU.
MarvellThunderX series products:
Next, let's talk about specific products. MarvellThunderX series strictly follows the roadmap of upgrading every two years. ThunerX2 series was born in 218, and after today's protagonist Thurner X3, ThunerX4 will be seen in 222, and the process is also upgraded all the way. The 28nm, 16nm and 7 nm
Thurner X3 are manufactured by TSMC's 7nmDUV process. Based on Marvell's self-developed architecture, the instruction set is compatible with ARMv8.3, with a maximum of 96 threads, and it continues to support 4 threads, that is, 384 threads at most, which is three times that of the previous generation, and it also supports two-way parallelism. At this time, a single system can provide 192 cores and 768 threads.
at the same time, the highest frequency benchmark is 2.4GHz and the highest acceleration is 3.1GHz, which is 1MHz higher than the previous generation.
the new kernel integrates four 128-bitNeonSIMD cells, which is equivalent to an x86AVX-512 in width, thus greatly improving the floating-point performance. Intel Xeon has two top models, and AMD Xiaolong has two 256-bitSIMD units per core, which are basically equivalent.
the memory supports eight channels of DDR4 with a frequency of 32MHz, and supports 64 PCIe 4. (with 16 controllers), which is further than the previous generation PCIe3. and adds 8 channels, and will be upgraded with the evolution of PCIe5./6. standards in the future.
of course, enterprise-level RAS and virtualization are all essential elements.
next, let's talk about performance. Of course, it's all official data from Marvell, and the specific performance depends on the specific workload.
in terms of comprehensive performance, the performance of ThunderX3 is improved by more than 25% compared with the previous generation IPC (instructions per clock cycle), which can basically be regarded as the progress of the architecture itself. Combined with higher running frequency, the performance of a single core is improved by more than 6%, and with the greatly increased number of cores, the overall performance can be improved by more than 3 times!
compared with x86 duo, ThunderX3' s advantage lies not in absolute performance, but in better energy efficiency, which is 3% higher than AMDRome II Xiaolong and 1.2 times ahead of Intel II Scalable Xeon.
ThunderX3 is not greedy for perfection, and there are only two main markets. One is cloud computing, including big data, databases, streaming media, Web, search, storage, mobile application development, mobile games, etc. Its highly parallel, rich I/O, ultra-low latency and native ARM are all unique advantages.
in terms of performance, single core and four threads can bring considerable performance improvement, which can be up to 8% in common applications and even 3-4 times improvement under specific load.
Compared with AMD's second-generation Xiaolong and Intel's second-generation scalable Xeon, ThunderX3 has not shown weakness in the cloud, and it has won all the fields it serves, with lower latency and supporting more virtual machines.
The second main direction is high-performance computing, including government departments, weather forecasting, oil and gas exploration, computer-aided engineering, biological genes, electronic design automation, etc. The characteristics of high memory bandwidth, high energy efficiency and high parallelism can be said to be very suitable.
comparing the two x86 schemes, ThunderX3 has a good performance in floating point, memory bandwidth index, quantum chemistry, computational fluid dynamics, computational chemistry, vector physics and other applications.
Marvell is also very positive and optimistic about the market prospect of ThunderX3, revealing that more than 2 customers who have deployed ThunderX2 are preparing to upgrade ThunderX3, and the subsequent ThunderX4 will continue to bring substantial improvement.
that's all about the ThunderX3 processor of the ARM chip.