On the basis of the first 19 models, AMD brought a new Xiaolong 7Fx2 series today, which has higher running frequency and larger third-level cache capacity, and the price and cost performance are as attractive as ever.
Xiaolong 7F72:
24 cores and 48 threads, the reference frequency is 3.2GHz, the highest acceleration is about 3.7GHz, the third-level cache is 192MB, the thermal design power consumption is 240W, and the wholesale price is 2450 USD.
Compared with the previous Xiaolong 7402, its reference/acceleration frequency is increased by 400/350MHz, and the third-level cache is expanded by half, at the expense of thermal design power consumption increased by 60W.
The closest competitor is Xeon Gold Medal 6248R, with a frequency of 3.0-4.0GHz, a three-level cache of 35.75MB, a thermal design power consumption of 205W and a price of 2,700 US dollars.
Xiaolong 7F52:
16 cores and 32 threads, the reference frequency is 3.5GHz, the highest acceleration is about 3.9GHz, the third-level cache is 256MB, the thermal design power consumption is 240W, and the wholesale price is 3100 USD.
Compared with the previous Xiaolong 7302, its frequency is greatly increased by 500/600MHz, the third-level cache is doubled (even more than the 24-core one), and the thermal design power consumption is also greatly increased by 85W.
The closest competitor is Xeon Gold Medal 6246R, with a frequency of 3.4-4.1GHz, a three-level cache of 35.75MB, a thermal design power consumption of 205W and a price of $3,286.
Xiaolong 7F32:
8 cores and 16 threads, the reference frequency is 3.7GHz, the highest acceleration is about 3.9GHz, the third-level cache is 128MB, the thermal design power consumption is 180W, and the wholesale price is 2100 USD.
Compared with the previous Xiaolong 7262, its frequency is increased by 500/500MHz, the third-level cache remains unchanged, and the thermal design power consumption is increased by 35 W.
The closest competitor is the Xeon Gold Medal 6250, with a frequency of 3.9-4.5GHz, a three-level cache of 35.75MB and a thermal design power consumption of 185W W.
In terms of performance, AMD also cited several examples. For example, Dell PowerEdge R6525 is equipped with Xiaolong 7F72 and a two-way four-node system, which set a world record for the performance of VMmark 3.1, which is up to 47% higher than before.
Microsoft SQL Server 2017 used two-way Xiaolong 7F52, and the number of transactions per minute increased by 17%, while using two-way Xiaolong 7F32, the number of transactions per minute increased by 35%.
At the same time, AMD also announced a number of new Xiaolong solutions.
Among them, Chaowei’s SMC SuperBlade is the world’s first blade server based on AMD Xiaolong, which was launched in the second quarter; HPE/Nutanix HCI certification has supported AMD Xiaolong, and related applications were launched in the third quarter; IBM Cloud newly added 48 core instances of Bare Metal of Xiaolong, which went on sale in the second quarter.
After two generations of hard work, AMD Xiaolong has gained extensive ecological support in the fields of servers and data centers. For example, Amazon AWS, Google Cloud, Microsoft Azure, IBM Cloud, Oracle Cloud and Tencent Cloud in the cloud are all major customers, not to mention top OEM customers such as Dell, Lenovo, HPE, Chaowei and Baidu.
In the field of supercomputers representing the ultimate performance, AMD Xiaolong has also gained a lot, including the US Air Force, Indiana University, Oak Ridge National Laboratory, Lawrence Livermore National Laboratory, Aragon National Laboratory, San Diego Supercomputing Center, French Meteorological Bureau, etc. AMD Xiaolong has been or will be deployed soon, and some will also be equipped with AMD Radeon Instinct computing cards.
According to the previously announced roadmap, AMD will launch the third-generation Xiaolong "Milan" based on 7nm process and Zen 3 architecture, and then the fourth-generation Xiaolong "Genoa" based on 5nm process and Zen 4 architecture.
Where does the computing power come from? Naturally, it is a chip, especially a chip based on x86 architecture. According to the global market share data of x86 architecture processors in 2019 released by Mercury Research, a well-known data research company, 99.9% of computers are using x86 architecture now, and the remaining 0.1% is divided by many chip manufacturers including VIA. In fact, as long as you look at the computers used by friends and relatives around you and the mainstream computing equipment in the data center, this result is not difficult to imagine.
That is to say, x86 undertakes most of the current computing tasks, and CPU plays a particularly important role in it. For data center users, perhaps everyone has begun to adopt acceleration chips such as GPU, FPGA and ASIC, but the core and most common computing power device is CPU. It can be said that the computing power of CPU largely determines the computing power of data center, and then determines the efficiency of data processing.
Especially for the current high-performance computing, big data and hyper-converged applications, they all put forward higher requirements for the specifications of processors-high clock speed, multi-core and large-capacity cache have become the favorite of these applications, and the multi-core platform built by this will also get better data processing capabilities. This also provides a great market opportunity for AMD. On April 14th, Beijing time, AMD released three new specifications of the second-generation EPYC processors to the world. They have 8 cores, 16 cores and 24 cores respectively, designed for mainstream applications, and all of them have high frequency to meet enterprise-level core business applications.
Debut is the peak, and many partners support AMD.
So, can the newly released AMD 7Fx2 series processors be recognized by the industry? The answer is yes. In fact, as early as last year, when the second-generation EPYC processor was released, AMD welcomed the support of many OEM companies including HPE, Dell Essence, Lenovo, Ultramicro and many cloud service providers such as Microsoft, IBM and Tencent. This time, the release of 7Fx2 series processors has also attracted widespread attention in the industry.
The first to respond is Dell EasyTrust. In fact, Dell EasyTrust has always been a close partner of AMD, and the two sides have a history of cooperation for many years. This time, Dell EasyTrust used four 1U dual-socket servers (model PowerEdge R6525) based on AMD 7F72 processor to form a cluster, and achieved the first good result in the VMmark3 virtualization vSAN test, with a performance improvement of 47%. This further proves the advantages of AMD 7F72 processor in hyper-convergence project, which is suitable for the current mainstream application environment.
Another close partner of AMD, HPE, also joined hands with Nutanix to launch the ProLiant DX server, an operating system based on AMD 7Fx2 series processors. This server is also oriented to hyper-converged applications. Thanks to the multi-core and maximum memory capacity support of 7Fx2 series processors, it can adapt to more virtual machines in hyper-converged applications, and it is expected to be launched in the third quarter.
Have you seen the blade server of all AMD platforms? It came today. Ultramicro announced that it will launch a high-performance blade server with the second generation EPYC processor in the second quarter. This service area will be oriented to data centers and high-performance computing applications, and can take into account both virtualization and VDI applications. What’s more worth mentioning is that apart from the core of AMD, it also supports 200G HDR InfiniBand network, and its high bandwidth and low latency are more suitable for users who are demanding performance.
Why did these partners give support and response at the first time? This is also due to the excellent performance of the new product. The high frequency of AMD 7Fx2 series processors is especially suitable for high performance computing applications, especially in graphic simulation, aerodynamic simulation and industrial design. According to authoritative tests, in the above applications, the new 7F72 and 7F52 processors have shown strong performance, which has improved graphic simulation and industrial design by more than 30%, while the improvement for aerodynamic simulation is more obvious, nearly doubling.
In addition to the support of OEM partners, cloud service providers have also shown strong interest in AMD 7Fx2 series processors. For example, IBM announced that it will take the lead in adopting 7F72 processor in its own cloud platform and apply it in its own bare metal cloud to cope with high-load high-performance computing, databases and other businesses. More importantly, IBM has started to take action, and soon we can experience the excitement brought by new products.
Above, we introduced the excellent performance of AMD 7Fx2 series processors in HPC applications. In fact, the database is also its strength. For example, in the test of Microsoft SQL Server 2017, the performances of 7F52 and 7F32 are also very eye-catching, both in terms of performance (tpm) and cost.
Of course, the above are only some cases of AMD 7Fx2 series processors. With the release of new products, more and more users will start to contact and use them, and more and more application cases will appear. In fact, with the number of digital applications, our demand for high-performance computing is becoming more and more urgent, and big data analysis has become an indispensable part of artificial intelligence, which requires the support of super computing power and is also the target market of AMD 7Fx2 series processors.
From the release of the second-generation EPYC processor last year to the launch of three new 7Fx2 series processors, AMD’s layout in the data center market has become more and more perfect, and its product classification has become more and more detailed. We can foresee that the market of the second-generation EPYC processor will grow with the development of applications and the improvement of customers’ computing power demand, and 7Fx2 is just the most mainstream enterprise data center market, which will naturally bring unprecedented pressure to Intel.
Editor in charge: gt