Russian company wants to develop AI Accelerator to compete with Nvidia

Russian company Baikal Electronics wants to develop its own AI Accelerator to provide an alternative to Nvidia. But the budget and schedule don’t even match the current task.

According to Aroged, AI Accelerator is computer hardware that handles the specific requirements of artificial intelligence (AI) that accelerates processes such as artificial neural network (ANN), machine learning (ML), and marketing tasks. machine sense. This is different from the deep learning that ordinary users exploit on graphics cards and is now focused on by chip manufacturers

In order not to be left behind, Baikal has established a new division to develop the AI ​​Accelerator for AI to meet the requirements of deep learning applications. Not only will it help compete with Nvidia, but the new division will allow Baikal to provide the Russian government with corresponding products to support the development of defense technology.

Still, there are question marks about the success Baikal’s division has achieved. Industry analysts say Baikal will need about three years for development purposes. In addition, the budget of 2 billion rubles (nearly 20 million EUR) is tight. By comparison, Nvidia has planned to spend about $6.7 billion in research budget on AI Accelerator in fiscal year 2023 alone.

In fact, Baikal admits that the AI ​​Accelerator it has developed cannot compete with products from the West, but the effort will help the company close the gap. However, analysts say that with a short time frame and limited budget, Baikal’s AI Accelerator power can be questionable, like what the Baikal-S processor encountered before

In addition to the development process, the production of the AI ​​Accelerator also poses many challenges to Baikal’s efforts. Russia itself has only one semiconductor factory for 90nm chips – which is completely unsuitable for current computing applications. Previously, foundries outside of China had stopped cooperating with Russia, so Baikal could not rely on TSMC as before. foundry The last option is probably China’s SMIC, which is also fearing further sanctions for such cooperation. Therefore, it is not clear whether the production target has been achieved as expected by Baikal or not.

Related posts

Google launches Gemini 2.0 – comprehensive AI that can replace humans

NVIDIA RTX 5090 can be 70% more powerful than RTX 4090?

iOS 18.2 launched with a series of groundbreaking AI features