매출 감소세를 피하지 못한 퀄컴이 온디바이스 AI에서 해결책을 찾고 있다. 퀄컴은 메타의 라마 2 (Llama 2) 대형 언어 모델(LLM)을 클라우드에만 의존하지 않고 온디바이스에서 실행할 수 있도록 최적화할 예정이다.
Qualcomm Terminal Deficit Continues… AI Portfolio Diversification
Meta-Qualcomm, Rama 2-based On-Device AI Hand-in-Hand
Snapdragon X75 Supports Industry-Leading 5G Speeds at Sub-6GHz
Unable to avoid declining sales, Qualcomm is looking for solutions in on-device AI.
Qualcomm announced its second quarter 2023 performance on the 3rd, reporting second quarter sales of $8.45 billion (about 11.18 trillion won), down 23% from $10.93 billion in the same period last year. Net profit was reported to be $1.8 billion (about 2.38 trillion won), down 52% from the same period last year.
The poor performance is closely related to the global smartphone terminal market. Apple, Huawei, etc. were major smartphone chipset customers of Qualcomm, but since Apple developed its own chips and was hit, the smartphone market down cycle has continued, and Qualcomm's sales have been on a downward trend since 2022. In addition, performance in other sectors such as IoT and automobiles is also showing poor results.
Samsung Electronics' Galaxy Z Flip, Fold 5, and Galaxy Tab 9, released on the 11th, are equipped with the Snapdragon 8 2nd generation. The Snapdragon 8 Gen 2 showed excellent performance compared to competitors in performance tests.
Therefore, Qualcomm's sales are likely to be influenced by market conditions rather than chipset performance. There is a positive outlook as Samsung Electronics announced that it has surpassed 1 million Galaxy units in pre-orders. However, if smartphone demand recovery is slow, Qualcomm's poor performance is likely to be prolonged.
■ Hidden Powerhouse in On-Device AI Applications Qualcomm is looking for opportunities in AI. Qualcomm is working on developing acceleration solutions such as AI engines along with key IP innovations and technologies. As the movement from cloud to edge devices requires power efficiency and high performance, Qualcomm is expected to be able to take a position in the transition to a hybrid AI powerhouse.
In particular, the Snapdragon platform currently supports generative AI models with over 1 billion parameters, and is expected to support over 10 billion parameters.
To this end, Qualcomm has joined hands with Meta to implement on-device AI. In July, Qualcomm announced that it would optimize Meta's Llama 2 large language model (LLM) to run on-device rather than rely solely on the cloud.
Running generative AI models like Rama 2 on-device—including smartphones, PCs, VR/AR headsets, and vehicles—allows developers to save on cloud costs while providing personalized experiences.
Powered by Snapdragon, these new on-device AI experiences can be used even when there’s no internet connection or in airplane mode.
“To effectively democratize generative AI, AI must be able to operate not only in the cloud, but also on edge devices such as smartphones, laptops, vehicles, and IoT,” said Durga Malladi, senior vice president and general manager of technology, planning and edge solutions, Qualcomm Technologies.
Qualcomm plans to support Rama 2-based AI implementation on Snapdragon-based devices starting in 2024.
“As AI use cases proliferate across devices, we expect on-device AI to drive performance recovery across all our products,” said Cristiano Amon, President and CEO. Qualcomm said it will deliver platform power efficiency and accelerated computing performance.
■ Snapdragon X75, Sub-6GHz top 5G speed Qualcomm announced that the Snapdragon X75 supports the fastest 5G download speeds, achieving 7.5 Gbps downlink speeds in sub-6 GHz bands with the Snapdragon X75 5G Modem-RF System.
Unveiled at MWC Barcelona in February, Qualcomm’s Snapdragon X75 5G Modem-RF System is designed to power next-generation networks, including 5G-Advanced and 6G. The modem features the Qualcomm 5G AI Processor Gen2 to deliver groundbreaking 5G performance. The introduction of AI-enhanced Qualcomm 5G Suite Gen 2 further improves performance in speed, range, mobility, and robustness.
These results were achieved using 4x carrier aggregation (4xCA) with four TDD channels and 1024-bit quadrature amplitude modulation (QAM) on a single downlink connection at speeds performed through 5G standalone (SA) device testing, utilizing a total of 300 MHz of spectrum.
The explanation is that aggregating four TDD channels can achieve higher data rates by combining different spectrum assets. Additionally, 1024 QAM integrates more data into each transmission compared to 256 QAM, ultimately increasing data throughput and improving efficiency.
Qualcomm expects these Snapdragon X75 features and 10 other modem-RF system features to improve download speeds, increase network capacity, and improve spectral efficiency. They are expected to support smooth progress in future data applications such as video streaming and downloads, online gaming, and across industries such as smartphones, mobile broadband, computing, automotive, IIoT, and 5G specialized networks.
“The Snapdragon X75 5G Modem-RF System is our smartest wireless modem to date, designed with our 5G Advanced Ready Architecture to prepare for the next generation of networks,” said Sunil Patil, vice president of product management, Qualcomm Technologies, Inc. “We look forward to continuing to collaborate with industry leaders to deliver best-in-class connectivity experiences and transform industries across consumer, enterprise, and industrial use cases.”
The Snapdragon X75 is currently sampling to customers, with commercial devices expected to be available in the second half of 2023.