When the Intel Ruixuan Pro B60 with 24GB of video memory meets the NVIDIA RTX A1000 with 8GB of video memory, how many unknown industry battles are hidden behind this seemingly disparity duel? Can Intel CEO Chen Liwu’s declaration that “if you want to return to the peak, you need to tell the truth” shake NVIDIA’s dominant position in the AI chip market?
Performance duel: The technology breakthrough behind the 3x graphics memory gap
Under the spotlight of Computex 2025, Intel showed industry-shocking data: Ruixuan Pro B60 is equipped with 24GB GDDR6 graphics memory, which is three times that of NVIDIA A1000. This is not only reflected in the hardware parameters, but also directly transformed into practical application advantages – when running large models such as Llama3, the generation speed of B60 is up to 2.7 times ahead of its rivals.
What is more worthy of attention is the multi-card interconnection scheme. Through the Project Battlematrix platform, eight B60 combinations can provide 192GB of video memory, which is enough to control a medium-scale AI model with 150 billion parameters. This kind of “stacked innovation” just hits the pain point of the current AI reasoning market – the thirsty demand of large models for graphics memory.
Market chess game: Intel’s two-line battle strategy
Chen Liwu’s “to tell the truth” strategy is reshaping Intel’s product route. On the one hand, use the B60 series to attack the high-end AI reasoning market held by NVIDIA, and on the other hand, harvest cost-effective users through the $299 B50. This approach of “high-end branding and mid-end grabbing share” is the same as AMD’s strategy against Intel.
But the real killer lies in the layout of the Chinese market. Due to export controls, it is difficult for NVIDIA’s latest flagship chip to enter China, and the B60/B50 series, which meets the regulations, just fills this vacuum. The relevant person in charge of Intel said bluntly that this will “cater to the market demand for domestic AI reasoning model computing power”.
Industry changes: from single-point breakthrough to ecological war
NVIDIA’s moat is never just hardware. The software barriers built by CUDA ecology make challengers break back again and again. But this time, Intel is obviously prepared – AIAssistantBuilder open source software stack, containerized deployment scheme, ISV certification system, these supporting measures directly point to NVIDIA’s life.
The management reform of “directly going deep into the seventh and eighth floors to listen to the opinions of engineers” revealed by Chen Liwu at the dinner may be the key to Intel’s revitalization of engineering culture. When the chip war enters the three-dimensional confrontation stage of “hardware + software + ecology”, execution will become the X factor that determines victory or defeat.
The duel is far from over. NVIDIA’s next-generation products are on the way, and Intel needs to prove that the success of the B60 is not a pasy. But in any case, 2025 is destined to be a watershed in the AI chip market – when the challenger takes out three times the memory configuration of the king, the rules of the whole industry are being rewritten. As Chen Liwu said, “Let the results speak”, consumers will eventually vote with their feet, and the balance of the market has begun to tremble slightly.
发表回复