Semiconductor
Huawei Ascend 910C alleged specs suggest it a tough rival to Nvidia H100
Ascend 910C is the latest AI chip in the Huawei world, with improved specs over its predecessors. Now, a new leak suggests that the respective processor holds certain significant components that make it a tough contender against the Nvidia H100.
Huawei is speeding up the launch preparations of the Ascend 910C. The chip has already been used in the DeepSeek R1 and is now preparing to enter mass production.
X analyst @ohlennart reveals the alleged specs of the Huawei Ascend 910C. The company aims to build stronger alternatives to Nvidia chipsets for Chinese customers.
Specifications
Inputs reveal that Huawei will use two different silicon interposers for new AI chips. These interposers will interconnect via an organic substrate. Simply put, the company will merge two of the Ascend 910B SoCs to obtain a higher performance.
This approach will result in 10-20x less die-to-die bandwidth over the Nvidia H100 chips. In other words, Huawei is using a simple technique to produce AI SoCs.
“The 910C is basically two co-packaged Ascend 910Bs, China’s best current-gen accelerator. But there’s a twist: most (potentially all) of these chips weren’t produced domestically – they were illicitly procured from TSMC despite export controls.”
Huawei Ascend 910C alleged specs suggest it tough rival to Nvidia H100 (Image Credits: X)
Thus, the Ascend 910C will achieve 800 TFLOP/s at FP16 with around 3.2 TB/s memory bandwidth, which is quite equivalent (80%) to Nvidia H100 AI chipset.
Huawei will further use the 7nm process technology to build the Ascend 910C. The company has stockpiled a large amount of 7nm process from TSMC before the US restrictions came into action in 2019. It will eventually adopt this tech from the new advanced artificial intelligence semiconductor.
Meanwhile, SMIC is now capable of producing 50,000 wafers per month. As a result, the Chinese chipmaker will also contribute 7nm process tech in the making of the Ascend 910C.
Huawei is constantly striving to become the best alternative to Nvidia’s AI chipsets. It will be interesting to see how successful these efforts will be.
(via)
