News

NVIDIA’s next-generation B300 GPUs will have 50% more AI power

A few months have passed since we saw how NVIDIA managed to reach the first place of the companies that have the most value in the world, all thanks to the creation of a specific type of hardware for a technology that for some reason or another , we all know. The importance it has had NVIDIA in the artificial intelligence sector is extremely large, meaning that its main competitors did not even have the option of entering this market, which they also intend to dominate next year with their next GPU next generation for AI, B300.

Hardware for AI is one of the largest implementations that we have seen in recent years, there are many people and companies that have looked for a way to obtain a series of specific components that allow them to make the most of the power that it is capable of offering. this technology. It is here where NVIDIA stands out, since they have not only focused on developing components for large servers, but also their traditional graphics cards along with the hardware for AI oriented for consumers that they have developed have allowed anyone to take full advantage of the potential that the company offers.

NVIDIA would have managed to improve the performance of its GPUs for AI by 50%

One of the objectives that practically all brands that manufacture components such as processors or graphics cards have had in recent years is to try to offer the greatest possible capacity to use artificial intelligence, either through an NPU or directly due to the capacity they have. For example, GPUs calculate millions of operations per second. In this aspect, being a company that has been dominating the graphics card sector for many years, NVIDIA has had great importance in the development of these components, but apparently things do nothing but improve for the brand.

We know various information about the new B300 that the company plans to launch next year, some GPU designed for servers that offer extremely high power, and which could be up to a top 50% to what the generation offered B200. All this comes at a cost, as we well know, and that is that one of the leaks that we have known in addition to this increase in power is also the increase in consumption, which would be about 200 W higher than the previous generation, that is, a total 1400 W.

NVIDIA B300 GPU NIC ConnectX-8

But compared to the fact that it greatly increases performance, the increase in consumption is actually not that high. One of the main reasons why they will be able to surpass the previous generation is in the use of 12-Hi HBM3E memory stacks, which will provide 288 GB of memory with 8 TB/s of bandwidth. On the other hand, they could also adopt the 800G ConnectX-8 NIC that they have developed themselves and which has twice the bandwidth of the current 400G ConnectX-7, in addition to offering 48 PCIe lanes compared to the 32 of their predecessor.

Related Articles

Leave a Reply

Your email address will not be published. Required fields are marked *