Chinese New Supercomputer bid farewell to import China-made CPU chip is ready to compete with foreign
https://www.laoyaoba.com/html/news/newsdetail?source=pc&news_id=651997
Recently, the development work has been completed soar FT-2000plus server CPU, soar company's partners are actively developing the appropriate machine products. FT-2000plus The chip is based on an improved version of FT2000 is based, though, and Intel are still some gaps in the single-core performance, but on multi-core performance, has reached the level of mainstream Intel CPU E5 server product.
Reportedly, National Defense Science Supercomputer being developed 1000P 3 Milky (performance eight times prowess Tai light), which will use the prototype CPU or FT-2000 or its successor as a master CPU. Taking into account the Milky Way 2A, National Defense has been calculated matrix 2000 Xeon Phi card Intel replaced with self-developed accelerator. From now on, the Milky Way Series Supercomputer will also count as super series and divinity, to bid farewell to foreign or CPU accelerators to achieve national production of chips. (Network transmission Tianhe-3 prototype compute nodes, the board has two FT-2000, industry experts told the author that this is misinformation) FT-2000 series chip is the result of profound knowledge in recent years, some domestic companies are keen to "go shortcut ", met with Intel, IBM, ARM, Qualcomm and other chip giant foreign technical cooperation or joint ventures, some foreign companies to buy back authorization to do CPU IP integration, it is propaganda independent intellectual property rights, and even the foreign company's chips packing look into homemade chips. While this approach can cut corners short period of time to come up with a good performance of the chip, but the foreign companies have always had "church disciples, starved to death master" mentality, it does not authorize the most advanced technology to China, but also in domestic units in getting foreign technology licensing, vulnerable to know these, I do not know why their plight, sustainability do not have technological development, after foreign companies to develop next-generation products, only the introduction of foreign technology again.
In contrast, soar choose a more difficult road of independent research and development, although the development will be slower, but more stamina. Than a decade ago, soar based on IA-64 instruction set CPU design, then again open source Sparc designed FT-1000 and FT-1500 is based on two CPU, FT-1000 is used Milky No. 1 Supercomputer , FT-1500 are used for No. 2 Milky supercomputer. After 2012 years ago, the soar select the ARM camp, won the ARM instruction set authorization, based on ARM 64 instruction set design their own CPU. And in a few years time, we have developed a FTC660, FTC661 two CPU cores, and FT-1500A / 4, FT- 1500A / 16, FT-2000, FT-2000plus and other CPU.
FT-1500A / 4 is used for a notebook PC and, using 28nm manufacturing process. Integrates four FTC660 CPU core, clocked at 1.5GHz-2.0GHz, integrated 2MB secondary cache and 8MB L3 cache, two DDR3-1600 integrated memory controller to provide 25.6GB / s memory access bandwidth, maximum power consumption is 15W.
FT-1500A / 16 is a server, using 28nm manufacturing process. Integrated FTC660 CPU core 16, clocked at 1.5GHz, 8MB integrated secondary cache and cache 8MB three, four DDR3-1600 integrated memory controller to provide 51.2GB / s memory access bandwidth, maximum power consumption 35W.
FT-2000 using 28nm manufacturing process, can be used for ultra high-performance compute nodes and the server computer. 64 FTC661 integrated processor cores. Work clocked at 1.5GHz-2.0GHz. Two integrated 32MB cache, 128MB support extended three cache. Support DDR3-1600 memory controller 16, may provide 204.8GB / s memory bandwidth access.
The latest FT-2000Plus using TSMC 16nm 2.0GHz-2.4GHz manufacturing process, clocked, integrated core processor 64 FTC661. In order to increase the degree of integration, abandoned L3cache, the DDR4 memory controller integrated into the chip. Currently, soar series CPU has formed a family, we can meet the actual needs of the military and political fields.
Enough to compete with mainstream server CPU and Intel in the lead position in the domestic ARM server CPU in
accordance with the actual test FT-2000 clocked at 2GHz, using GCC4.8 compiler, SPEC2006 test scores were SPECint_rate2006: 570, SPECfp_rate2006: 482, and intel Xeon E5-2695v3 quite.
The latest FT-2000Plus since deleted the FT-2000 performance L3 cause some children fall in SPEC test scores, and other improvements to enhance the integration of DDR4 brings smooth, and the FT-2000Plus at the same frequency with no big difference. Why is that soar company to do it? The reason should be to commercialization, improve the integration of FT-2000plus can easily OEMs to reduce the cost of developing FT-2000plus whole product. However, a better use TSMC 16nm manufacturing process after the highest frequency can be stabilized to 2.4G, the FT-2000plus with respect to the FT-2000 is not a small performance to meet high-end server and the host CPU Supercomputer performance requirements.
In recent years, many manufacturers have started developing ARM server CPU. In China, Huawei and China is also working on the development of core ARM server CPU. However, for now, it soars ARM server CPU is the absolute leader in terms of performance.
Sinochip through mainland China and Qualcomm joint venture company, based on access to the source code and the source code to design their own CPU from Qualcomm hands. However, the core of China through Beijing R & D center opened in time is too short for commercial products come out officially take time.
By contrast, Huawei R & D ARM server CPU time much earlier, but also in Beijing and the US research and development center, has come out Hi1612 product. Although the publicity, Huawei claimed Hi1612 is "independent research and development", "In addition to the memory cell has a complete knowledge of the product," but in fact on this server CPU using TSMC 16nm manufacturing process, its core is the ARM Cortex A57. And Hi1612 performance is also more general, perhaps because of a lack of Huawei server CPU design experience, this relatively limited server CPU performance to FT2000 soar as a reference, Huawei this server CPU manufacturing process in the next two generations ahead of the situation, FT2000 half the power to achieve a performance less than half of the FT2000. However, according to the grapevine, Huawei under an ARM servers being developed, and ramp door of Huawei R & D center in Beijing Godson has ARM Cortex A72 take at least two modified version (Huawei money, buy ARM "big gifts "), together with US-based R & D center can provide technical support, in case there are ARM Cortex A72-based Huawei's performance under an ARM server CPU's worth the wait. Conclusion Indeed, FT-2000 / FT-2000plus there are some shortcomings, such as single-threaded performance weak. Whether FT2000, or FT-2000plus, although to achieve multi-threaded performance on the performance of Intel's high-end server CPU, but in the single-threaded performance, there are still a large gap - even clocked control in 2.5GHz, Intel's SPECint_base2006 server CPU scores are above 25, while the FT2000 SPECint_base2006 only 12.4 (@ 2.0GHz), FT- 2000plus expected only 14+ minutes (@ 2.4GHz).
Software ecosystem is also a problem, because the ARM ecosystem soar chose to follow the path, the soar of commercial promotion depends largely on the ARM ecosystem of the perfect situation. And on the server ecology, ARM and X86 larger than the gap, which is not conducive to commercial FT-2000 / FT-2000plus's.
However, things are difficult, so far, Hunan Great Wall Galactic, Hunan meter Sea Shield companies, space 706 has introduced a new generation soar microprocessor-based self-controlled products. (FT-CPU machine products) Wall Galaxy EF111 company has developed a series of compute servers, Chi Heng AF21B machine, causing NF15N Xiang Shi Heng KF510 Series laptop and desktop computers desktop machine products. Which, EF111 series of high-performance server-based computing servers whole new generation of domestic FT2000 Plus processors can be used in the field of ERP, high-performance database, virtualization and business intelligence, big data analysis. Aerospace Science and Industry two homes 706 days Yue also launched 64-core high-end rack-mounted servers. According to public information, prototype No. 3 Milky CPU FT-2000 has been used in the chips calculated on the selected nodes. This is not only to get rid of imported chips Supercomputer Tianhe constraints is a good news for the FT-2000, it is also a very good advertising.
Reportedly, National Defense Science Supercomputer being developed 1000P 3 Milky (performance eight times prowess Tai light), which will use the prototype CPU or FT-2000 or its successor as a master CPU. Taking into account the Milky Way 2A, National Defense has been calculated matrix 2000 Xeon Phi card Intel replaced with self-developed accelerator. From now on, the Milky Way Series Supercomputer will also count as super series and divinity, to bid farewell to foreign or CPU accelerators to achieve national production of chips. (Network transmission Tianhe-3 prototype compute nodes, the board has two FT-2000, industry experts told the author that this is misinformation) FT-2000 series chip is the result of profound knowledge in recent years, some domestic companies are keen to "go shortcut ", met with Intel, IBM, ARM, Qualcomm and other chip giant foreign technical cooperation or joint ventures, some foreign companies to buy back authorization to do CPU IP integration, it is propaganda independent intellectual property rights, and even the foreign company's chips packing look into homemade chips. While this approach can cut corners short period of time to come up with a good performance of the chip, but the foreign companies have always had "church disciples, starved to death master" mentality, it does not authorize the most advanced technology to China, but also in domestic units in getting foreign technology licensing, vulnerable to know these, I do not know why their plight, sustainability do not have technological development, after foreign companies to develop next-generation products, only the introduction of foreign technology again.
In contrast, soar choose a more difficult road of independent research and development, although the development will be slower, but more stamina. Than a decade ago, soar based on IA-64 instruction set CPU design, then again open source Sparc designed FT-1000 and FT-1500 is based on two CPU, FT-1000 is used Milky No. 1 Supercomputer , FT-1500 are used for No. 2 Milky supercomputer. After 2012 years ago, the soar select the ARM camp, won the ARM instruction set authorization, based on ARM 64 instruction set design their own CPU. And in a few years time, we have developed a FTC660, FTC661 two CPU cores, and FT-1500A / 4, FT- 1500A / 16, FT-2000, FT-2000plus and other CPU.
FT-1500A / 4 is used for a notebook PC and, using 28nm manufacturing process. Integrates four FTC660 CPU core, clocked at 1.5GHz-2.0GHz, integrated 2MB secondary cache and 8MB L3 cache, two DDR3-1600 integrated memory controller to provide 25.6GB / s memory access bandwidth, maximum power consumption is 15W.
FT-1500A / 16 is a server, using 28nm manufacturing process. Integrated FTC660 CPU core 16, clocked at 1.5GHz, 8MB integrated secondary cache and cache 8MB three, four DDR3-1600 integrated memory controller to provide 51.2GB / s memory access bandwidth, maximum power consumption 35W.
FT-2000 using 28nm manufacturing process, can be used for ultra high-performance compute nodes and the server computer. 64 FTC661 integrated processor cores. Work clocked at 1.5GHz-2.0GHz. Two integrated 32MB cache, 128MB support extended three cache. Support DDR3-1600 memory controller 16, may provide 204.8GB / s memory bandwidth access.
The latest FT-2000Plus using TSMC 16nm 2.0GHz-2.4GHz manufacturing process, clocked, integrated core processor 64 FTC661. In order to increase the degree of integration, abandoned L3cache, the DDR4 memory controller integrated into the chip. Currently, soar series CPU has formed a family, we can meet the actual needs of the military and political fields.
Enough to compete with mainstream server CPU and Intel in the lead position in the domestic ARM server CPU in
accordance with the actual test FT-2000 clocked at 2GHz, using GCC4.8 compiler, SPEC2006 test scores were SPECint_rate2006: 570, SPECfp_rate2006: 482, and intel Xeon E5-2695v3 quite.
The latest FT-2000Plus since deleted the FT-2000 performance L3 cause some children fall in SPEC test scores, and other improvements to enhance the integration of DDR4 brings smooth, and the FT-2000Plus at the same frequency with no big difference. Why is that soar company to do it? The reason should be to commercialization, improve the integration of FT-2000plus can easily OEMs to reduce the cost of developing FT-2000plus whole product. However, a better use TSMC 16nm manufacturing process after the highest frequency can be stabilized to 2.4G, the FT-2000plus with respect to the FT-2000 is not a small performance to meet high-end server and the host CPU Supercomputer performance requirements.
In recent years, many manufacturers have started developing ARM server CPU. In China, Huawei and China is also working on the development of core ARM server CPU. However, for now, it soars ARM server CPU is the absolute leader in terms of performance.
Sinochip through mainland China and Qualcomm joint venture company, based on access to the source code and the source code to design their own CPU from Qualcomm hands. However, the core of China through Beijing R & D center opened in time is too short for commercial products come out officially take time.
By contrast, Huawei R & D ARM server CPU time much earlier, but also in Beijing and the US research and development center, has come out Hi1612 product. Although the publicity, Huawei claimed Hi1612 is "independent research and development", "In addition to the memory cell has a complete knowledge of the product," but in fact on this server CPU using TSMC 16nm manufacturing process, its core is the ARM Cortex A57. And Hi1612 performance is also more general, perhaps because of a lack of Huawei server CPU design experience, this relatively limited server CPU performance to FT2000 soar as a reference, Huawei this server CPU manufacturing process in the next two generations ahead of the situation, FT2000 half the power to achieve a performance less than half of the FT2000. However, according to the grapevine, Huawei under an ARM servers being developed, and ramp door of Huawei R & D center in Beijing Godson has ARM Cortex A72 take at least two modified version (Huawei money, buy ARM "big gifts "), together with US-based R & D center can provide technical support, in case there are ARM Cortex A72-based Huawei's performance under an ARM server CPU's worth the wait. Conclusion Indeed, FT-2000 / FT-2000plus there are some shortcomings, such as single-threaded performance weak. Whether FT2000, or FT-2000plus, although to achieve multi-threaded performance on the performance of Intel's high-end server CPU, but in the single-threaded performance, there are still a large gap - even clocked control in 2.5GHz, Intel's SPECint_base2006 server CPU scores are above 25, while the FT2000 SPECint_base2006 only 12.4 (@ 2.0GHz), FT- 2000plus expected only 14+ minutes (@ 2.4GHz).
Software ecosystem is also a problem, because the ARM ecosystem soar chose to follow the path, the soar of commercial promotion depends largely on the ARM ecosystem of the perfect situation. And on the server ecology, ARM and X86 larger than the gap, which is not conducive to commercial FT-2000 / FT-2000plus's.
However, things are difficult, so far, Hunan Great Wall Galactic, Hunan meter Sea Shield companies, space 706 has introduced a new generation soar microprocessor-based self-controlled products. (FT-CPU machine products) Wall Galaxy EF111 company has developed a series of compute servers, Chi Heng AF21B machine, causing NF15N Xiang Shi Heng KF510 Series laptop and desktop computers desktop machine products. Which, EF111 series of high-performance server-based computing servers whole new generation of domestic FT2000 Plus processors can be used in the field of ERP, high-performance database, virtualization and business intelligence, big data analysis. Aerospace Science and Industry two homes 706 days Yue also launched 64-core high-end rack-mounted servers. According to public information, prototype No. 3 Milky CPU FT-2000 has been used in the chips calculated on the selected nodes. This is not only to get rid of imported chips Supercomputer Tianhe constraints is a good news for the FT-2000, it is also a very good advertising.