Construction of large model computing power system and key technology analysis

doi:10.12267/j.issn.2096-5931.2025.11.012

Abstract

Abstract:

Firstly, the working principle and key technologies of the large model; Secondly,the challenges faced by large model computing power in terms of computation, networking, and storage were analyzed, and further research was conducted on the key technological advancements of large model computing power in storage, high-performance ensemble communication, networking, and infrastructure. Once again, a large model computing power system include collaborative construction of computing, storage, and network was constructed, as well as software and hardware collaborative optimization. The three key elements of computing power, storage power, and network power in the large model computing power system have been clarified; Finally, the development trends such as the demand for large computing power clusters, the development of end-to-end large model, and the empowerment of new quality productivity by artificial intelligence were analyzed.

Key words: large language model, LLM computing power system, collaborative computing, storage, and network

CLC Number:

TP18
F49

GUO Liang, WANG Yue, LI Jie. Construction of large model computing power system and key technology analysis[J]. Information and Communications Technology and Policy, 2025, 51(11): 81-88.

Add to citation manager EndNote|Ris|BibTeX

URL:

http://ictp.caict.ac.cn/EN/10.12267/j.issn.2096-5931.2025.11.012

http://ictp.caict.ac.cn/EN/Y2025/V51/I11/81

Figures/Tables 1

References 21

[1]	KAPLAN J, MCCANDLISH S, HENIGHAN T, et al. Scaling laws for neural language model[J]. arXiv Preprint, arXiv:2001.08361, 2020.
[2]	VASWANI A, SHAZEER N, PARMAR N, et al. Attention is all you need[C]// Proceedings of the 31st International Conference on Neural Information Processing systems. Long Beach: NIPS, 2017: 5998-6008.
[3]	OpenAI. Improving language understanding by generative pre-training[R], 2018.
[4]	RADFORD A, WU J, CHILD R, et al. Language models are unsupervised multitask learners[J]. OpenAI Blog, 2019, 1(8): 9.
[5]	BROWN T B, MANN B, RYDER N, et al. Language models are few-shot learners[J]. Advances in Neural Information Processing Systems, 2020, 33: 1877-1901.
[6]	DEEPSEEK-AI D A, GUO D, YANG D, et al. DeepSeek-R1:incentivizing reasoning capability in LLMs via reinforcement learning[EB/OL]. 2025[2025-07-08]. https://arxiv.org/pdf/2501.12948.
[7]	SCHICK T, DWIVEDI-YU J, DESSI R, et al. Toolformer: language models can teach themselves to use tools[EB/OL]. 2023[2025-07-08]. https://doi.org/10.48550/arXiv.2302.04761.
[8]	WAN Z, WANG X, LIU C, et al. Efficient large language models: a survey.[J]. arXiv Preprint, arXiv: 2312.03863, 2023.
[9]	王帅, 李丹. 分布式机器学习系统网络性能优化研究进展[J]. 计算机学报, 2022, 45(7): 1384-1411.
[10]	JIANG Z, LIN H, ZHONG Y, et al. MegaScale: scaling large language model training to more than 10,000 GPUs[C]// The Proceedings of the 21st USENIX Symposium on Networked Systems Design and Implementation. Santa Clara: USENIX Association, 2024: 16-18.
[11]	Team L, AI @Meta. The Llama 3 herd of models[EB-OL]. (2024-07-23)[2025-07-08]. https://Llama.meta.com/.
[12]	徐勇. 先进存力在三个方面展现六大新特征[N]. 人民邮电, 2024-10-31(005).
[13]	郭亮, 王少鹏, 权伟, 等. 面向大模型的智算网络发展研究[J]. 电信科学, 2024, 40(6): 137-145. doi: 10.11959/j.issn.1000-0801.2024147
[14]	中国信息通信研究院. 中国运力发展报告(2024年)[R], 2024.
[15]	JLL. Data centers 2024 global outlook[R], 2024.
[16]	The International Energy Agency (IEA). Data centres and data transmission networks[EB/OL]. (2025-01-03)[2025-07-08]. https://www.iea.org/energy-system/buildings/data-centres-and-data-transmission-networks?itid=lk_inline_enhanced-template.
[17]	ZHANG Z X, WEN Y B, LYU H Q, et al. AI computing systems for large language models training[J]. Journal of Computer Science and Technology, 2024, 40(1): 6-41. doi: 10.1007/s11390-024-4178-1
[18]	SemiAnalysis. 100,000 H100 clusters: power, network topology, ethernet vs infiniBand, reliability, failures, checkpointing[R], 2024.
[19]	Apple. Apple intelligence foundation language models[R], 2024.
[20]	Yale University. Scaling laws for post training quantized large language models[R], 2024.
[21]	DeepSeekAI. DeepSeek-V3 technical report[R], 2024.