the first step to reducing costs for large models is to start with the intelligent computing center
2024-09-23
한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina
after the first stage of competing in computing power and focusing on the leading edge of basic models, the development of big models has now entered a more pragmatic second stage, namely commercial application. how to reduce the cost of big model applications has become the focus of industry attention.
jd cloud has taken the first step towards reducing the cost of large models, starting with the intelligent computing center. it helps customers reduce the cost of using large models by using liquid cooling technology to reduce refrigeration expenses and introducing ai for intelligent energy management.
jd cloud (langfang) data center
keep cooling costs down
in the operation of intelligent computing centers, electricity costs account for more than 60% of operating expenses, a large part of which comes from refrigeration costs. optimizing the refrigeration system has become an important part of improving energy efficiency and reducing costs in data centers.
liquid cooling is an efficient, energy-saving and stable heat dissipation method. the coolant transfers heat more than 6 times faster than air, and is becoming the "standard configuration" of intelligent computing centers.
jd cloud's independently developed alpha intelligent computing power module uses innovative liquid cooling technology to meet the deployment requirements of computing power density ranging from 20kw/cabinet to 100kw/cabinet. the computing power module cooling factor clf is as low as 0.07, which means that the energy consumption of the cooling system is only 7% of that of it equipment, helping customers significantly reduce cooling energy consumption.
in addition, by adopting evac radiator and liquid cooling technology, jd cloud's new generation of liquid-cooled servers can improve heat dissipation efficiency by 50%; and based on a diversified power supply solution, under the condition of a single cabinet of 20kw, the power consumption of a single cabinet can save 8,500 kwh of electricity in a year.
jd cloud alpha intelligent computing module
energy saving rate increased
intelligent computing centers often have higher power density, more than five times that of traditional data centers, and the requirements for operation and maintenance also grow exponentially.
to ensure that every watt of electricity can be used to its maximum effect, jd cloud launched the energy saver product, which uses high-precision modeling to achieve 1:1 simulation construction of the hvac or related electromechanical systems of the entire link in the infrastructure, and realizes the refinement of global energy saving through ai algorithms, with an energy saving rate of 10%-45%.
based on jd cloud's self-developed jingwei software platform, it integrates functions such as fault location, capacity management, dynamic environment monitoring, and intelligent energy-saving operation management of air conditioners, greatly improving the efficiency of power system use while ensuring availability.
cost reduction practice external service industry
in addition to serving itself, jd cloud is also actively exporting the intelligent computing center's capabilities to reduce costs and improve efficiency.
the qianhai intelligent computing center demonstration base, invested and built by broad group data, uses jd cloud's "alpha" computing power module products. it took less than 90 days from planning to project delivery. with the help of liquid cooling technology, the infrastructure energy consumption was saved by more than 30%.
in addition, jd cloud also supports the construction and operation of qinghai telecom's intelligent computing center. by jointly building a stable and reliable commercial intelligent computing cluster, it exports universal high-performance ai computing power to the society and provides strong momentum for the application of large models.
at present, with the continuous evolution of technology, intelligent computing centers are gradually moving towards a higher level of automation and intelligent upgrades. jd cloud has a digital computing power matrix from maas services to intelligent computing clusters to hardware infrastructure, and based on a mature intelligent computing cloud operation system, customers can enjoy more efficient, economical and stable intelligent computing services.
