跳到正文
The Elec· JEONG IL JOO·· 17 天前AI 評分53

AMD稱Ryzen AI在CPU效能、遊戲續航及價格方面勝過RTX Spark

AMD Says Ryzen AI Outperforms RTX Spark in CPU Performance, Battery Life and Price

AI 導讀

AMD在Advancing AI Seoul 2026表示,其AI PC在Cinebench 2024多核心測試中的CPU效能較Nvidia RTX Spark高14%,遊戲續航近50%更長,相關系統約便宜17%。

正文

Alexey Navolokin, senior director at AMD, speaks at the event. (Photo: THE ELEC)

Alexey Navolokin, senior director at AMD, speaks at the event. (Photo: THE ELEC)

Alexey Navolokin, senior director at AMD, speaks at the event. (Photo: THE ELEC)
Alexey Navolokin, senior director at AMD, speaks at the event. (Photo: THE ELEC)

AMD highlighted the gaming efficiency of its artificial intelligence (AI) PCs compared with rival products and unveiled AI optimization results achieved in collaboration with South Korean companies.

Alexey Navolokin, AMD’s senior director of commercial sales for Asia-Pacific (APAC), said at the AMD Advancing AI Seoul 2026 event held at the Westin Seoul Parnas in Samseong-dong on Nov. 4 that AMD’s central processing unit (CPU) delivered 14% better performance than Nvidia’s RTX Spark in benchmark tests. “Battery life during gaming was also nearly 50% longer,” he said.

CPU performance was measured using the Cinebench 2024 multi-core benchmark, while battery life was evaluated using the PCMark 10 Gaming benchmark.

Nvidia’s RTX Spark is a system-on-chip (SoC) for AI PCs launched in October 2026. AMD did not explicitly identify the product used in the comparison, but it is believed to be the Ryzen AI Max+ 395, codenamed Strix Halo, which the company introduced at the event as a chipset for high-performance AI PCs.

The Ryzen AI Max+ 395 features a CPU with up to 16 cores, a Radeon 8060S GPU based on the RDNA 3.5 architecture and up to 128GB of LPDDR5X memory. The RTX Spark CPU has 20 cores.

“AMD systems, including finished products such as laptops and workstations, are also about 17% cheaper,” Navolokin said. “Moreover, RTX Spark does not properly support popular online games.” This is because the anti-cheat software required to run those games does not support RTX Spark’s Arm-based CPU architecture.

AMD, however, did not compare the graphics performance of the two chipsets. The Radeon 8060S offers performance comparable to Nvidia’s laptop RTX 4060 to RTX 4070 GPUs.

Kim Tae-ho, CTO of Nota AI, presents a Strix Halo optimization case study. (Photo: THE ELEC)
Kim Tae-ho, CTO of Nota AI, presents a Strix Halo optimization case study. (Photo: THE ELEC)

South Korean companies also presented examples of optimizing AI processing performance on Strix Halo. Nota AI used its proprietary Nota Optimizing Agent to iteratively optimize AI models, runtime engines and code-level kernels for Strix Halo.

As a result, Strix Halo’s GPU and neural processing unit (NPU) were able to share a single AI model, completely eliminating duplicate memory usage. Previously, running separate instances of AI models on each processor resulted in wasted memory. The company also achieved inference performance improvements of up to 30% across several large language models (LLMs).

Clevi developed Uniform, a Strix Halo-based document processing data engine. Uniform incorporates a concurrent acceleration pipeline that enables simultaneous NPU and GPU processing.

The NPU converts user-provided documents and images into data that AI can process, while the GPU generates HTML text suitable for AI agents from the NPU-processed data. As the GPU generates a response, the NPU simultaneously processes the next documents and images. This approach reduced the waiting time before the first response text appears by about 40%.

來源:The Elec · thelec.net