The technical specifications of the NVIDIA GH100 GPU with 80,000 million transistors come to light
The technical information of the next NVIDIA GH100 based on Hopper architecture. This new GPU is made with a node TSMC’s 4N featuring the most advanced EUV technology. This is how he got to 80 billion transistors along with up to 6 HBM3 memory stacks. This transistor count is 50% higher than the previous generation GA100, but in a size less than 814mm2. This is possible because the previous generation is manufactured with a 7nm node from TSMC.
Inside it counts 18,432 FP32 CUDA cores and 9,216 FP64 double-precision CUDA cores, it also has 576 Tensor Cores. It also has basic GPU functions that will be able to offer output to connect a monitor or do some simple graphics tasks. has an interface PCI Express 5.0 x16 and a clock speed of 1.8 GHz. We have already seen how this new card has been on sale in Japan for about 30,000 euros in exchange.
This new NVIDIA unit will feature a 80GB HBM3 6144-bit base memory with a bandwidth will exceed 3TB/s. The GH100 unit will also feature NVLink technology to deliver 900 GB/s interconnect bandwidth. It is also expected that its consumption will increase with respect to the current 400W to 700Wa consumption that is also expected to is greatly increased in versions for home computers of NVIDIA GPUs.