Just three weeks after AMD's release of Xiaolong Milan, the birth of Intel Ice Lake made the third generation of the fastest play a peak confrontation with the third generation of Xiaolong.
According to the Hyperion Research prediction, in March 2020, the COVID-19 great popularity basically stopped the world, and the global high-performance computing server market is expected to increase from US $ 13.7 billion in 2019 to 2020 in 2020 $ 14.5 billion. Since then, of course, the situation has changed. In the latest forecast of Hyperion in November 2020, Hyperion stated that the global HPC server revenue in 2020 will reach $ 11.9 billion, a decrease of 17.9%from previous forecasts. The business shrinks.
However, with the gradual control of the epidemic, the global economy gradually recovered, and the growth of high -performance computing markets will gradually rise. Hyperion Research's judgment growth mainly comes from two reasons: the world ’s new demand (drug research and development field) in order to fight COVID-19; the HPC market demand from the cloud is faster than the original expected growth.
As the earliest country to effectively control the epidemic in the world, China is also the only country in the world's major GDP in 2020. As domestic high -tech enterprises' investment in R u0026 D personnel and funds continues to increase, it is not difficult to see that the domestic high -performance computing market growth rate is expected to be higher than the global average.
On the other hand, from the perspective of the semiconductor industry, the horn of competition has already sounded. Just three weeks after AMD's release of Xiaolong Milan, the birth of Intel Ice Lake made the third generation of the fastest play a peak confrontation with the third generation of Xiaolong. Intel claims that each processor of the Ice Lake-SP platform contains up to 40 u0026 ldquo; Sunny Cove u0026 RDQUO; of which built-in acceleration functions and new instructions can significantly improve the performance of artificial intelligence, high performance computing, network and cloud workload.
In addition to increasing the core number from 28 from the previous generation Cascade Lake to 40, ICE LAKE also provides 8 DDR4-3200 memory channels per slot. 64 fourth-generation PCIE channels, while the previous generation only provided only 6 DDR4-2933 channels, each slot support up to 48 third-generation PCI channels.
Intel revealed that with these enhanced functions and the AVX-512 and for calculating accelerationFor deep learning acceleration for artificial intelligence acceleration, compared with the previous generation, ICE LAKE has increased the data center load performance by 46%, and the average performance of high -performance computing has increased by 53%. In the early internal benchmark test*, Intel ICE LAKE processors also exceeded the main high -performance computing, artificial intelligence and cloud applications, and the third -generation Xiaolong processor codenamed Milan, which was recently released.
In an interview, Trish Damkroger, Vice President of Intel and General Manager of High Performance Computing Department, emphasized the improvement of Intel's core and high -performance computing platform for the core of Sunny Cove. She said: u0026 ldquo; For the workload of limited memory, it is critical to have 8 memory channels. In addition to 40 cores and AVX-512 instructions, ICE LAKE can provide excellent performance under the computational dense workload. u0026 rdquo; Damkroger further emphasized the Intel Speed u200bu200bSelect Technology (SST), which can finely control the frequency, core number, and power consumption of the processor. In ICE Lake, it can make the dynamic adjustment of core numbers and power more flexible.
DAMKROGER pointed out that (ICE Lake, the third -generation Intel Solidarity processor) in Monte Carlo algorithm simulation, which is 70 % higher than the previous generation, and competitors products (That is, AMD's 7-nanometer Milan CPU) is 50 % higher than that of ICE LAKE's L1 and L2 cache size, eight faster memory channels, and AVX-512 instructions.
Dan Olds, chief research officer of the interest360 Research, said: u0026 ldquo; Intel is definitely better than AMD than before. Before the release today, the choice of AMD was almost on the board, and now this situation no longer exists. Judging from the benchmark test of the previous generation of products in Intel, WRF is nearly 60 % higher, Monte Carlo algorithm is 70 % higher, Linpack is 38 % higher, HPCG is 41 % higher u0026 mdash; u0026 mdash; HPCG is a stability test, for. Customers are of great significance. u0026 rdquo;
The figure below shows the top ICE LAKE processor (40 cores) and top AMD Xiaolong Milan processor (64 cores) under the dual slot configuration. , Comparison of cloud computing and artificial intelligence. According to Intel's test, ICE LAKE's performance is 18%higher than Milan when running LinPack, 27%higher when running NAMD, and 5 high when running the Monte Carlo algorithm 5 50%.
Since AMD re -entered the data center field in 2017, Intel has firmly held about 90 % of the server market share. Hyperion Research senior adviser Steve Conway told HPCWIRE: u0026 ldquo; Needless to doubt that Intel X86 can easily lead the type of processor in the global high -performance computing market. u0026 rdquo; Research by this market survey company shows that Intel X86 may maintain a dominant position within its predicted period.
SuperMicro's senior vice president of on -site application engineering Vik Malyala told HPCWire that their customers are eager to make full use of the fourth -generation PCIe and higher core density in Ice Lake. He said: u0026 ldquo; For our customers, it has been optimized for many workloads for the Intel architecture for a long time. This is why many of our customers are willing to wait, not to adopt other products. u0026 rdquo;
At a pre -communication meeting recently, Intel revealed that the rise in ICE LAKE production capacity in the fourth quarter of last year has risen smoothly. The company had shipped more than 200,000 in the first quarter of 2021, and was widely used in all market segments. Performance computing customers. So far, the high -performance computing of related products has been received, including Germany's LRZ and Max Planck, Cineca in Italy, the South Korean Meteorological Administration, and the National Institute of Science and Technology (AIST) of the National Advanced Industrial Science and Technology of Japan, the University of Tokyo and Osaka University.