The GH100 Hopper is probably the largest GPU in history

Nvidia GH100 Hopper Datacenter GPU can be called the largest GPU to date with its 1000 square millimeters.

HPC Accelerator, one of Nvidia’s most well-known whistleblowers on Twitter, reports The new company It is called GH100 and its surface area is only slightly less than 1000 square millimeters. This is 20% larger than the GA100 GPU based on the Amp architecture, which is currently the largest GPU It is considered a market.

Yesterday, news broke that Nvidia was suing Hopper for trademark It was suggested, of course, that this would not be an obstacle to the chip’s technical development. Apparently Nvidia and company Dish Network They used the same name on their products, which has caused a number of legal issues. Negotiations between the two companies now appear to take several weeks, which could prevent the unveiling of a new GPU. GH100 at the March GTC 2022 event Be.

It goes on to say that rumors about the GH100 GPU’s Multi-Chip-Module design may not have been true, and the chip still has an integrated design. Now two well-known data disclosors, including kopite7kimi and greymon55, have also claimed that the GH100 will be an integrated chip with a single-die design. However, there seems to be an MCM chip in the Hopper series, apparently called the GH102.

As mentioned, Nvidia Ampere GA100 is currently the largest graphics processor in production, with a die size of 826 square millimeters. The GH100 GPU is likely to be a high-end chip. Kopite had previously reported that the power consumption of this GPU would probably reach 1,000 watts, which should have a very efficient cooling system. Of course, the company already has a history of introducing liquid cooling systems for its data center GPUs, called HGX. Therefore, there does not seem to be a particular problem in this regard.

Unlike AMD, which will use MCM style in its chips, Nvidia is said to intend to use the new generation of Nvidia GeForce RTX 40 series graphics cards in NVIDIA Ada Lovelace graphics processors, while still adhering to the traditional and integrated design of its products. However, some rumors suggest that this method may be used in Hopper GPUs, especially for data centers and artificial intelligence (AI) applications. But what is certain is that the MCM architecture is not used in Ada Lovelace gaming products for the general consumer.

AMD, on the other hand, will introduce new products in the RDNA 3 and CDNA 2 families, based on the MCM and Monolithic chip architecture. CDNA 2 GPUs will use exclusively MCM chips and RDNA 3 models will have a combination of MCM and integrated design.

It is not yet clear which of the two companies’ products will be based on what manufacturing technology. But AMD’s new RDNA 3 and CDNA 2 families are said to use a combination of the company’s upgraded 6 and 5-nanometer chips. Nvidia, on the other hand, will use TSMC’s Nod 5nm process in the Ada Lovelace series. Of course, it is not clear here whether these products are based on N5 or N5P.

In the end, you can share your opinions about this with Zomji and other users.