The latest graphics processing units or GPUs of NVIDIA are said to be undergoing testing already and the company is rumored to be gearing up for the unveiling of its next-generation GPUs at the upcoming GPU Technology Conference in April. According to a report in Ecumenical News, the Pascal GPUs, which are the successors of the Maxwell GPUs, will be introduced during the conference slated for April 4 to 7 in Silicon Valley. The report also claims that the GPUs will be available starting June.
NVIDIA announced that it was working on its next-generation GPUs during the 2014 GPU Technology Conference. According to the company’s blog, the new GPUs will have three new key features: 3D memory, which allows the GPUs to “get data from memory more quickly” as well as allow for more compact GPUs for smaller devices and better energy efficiency; unified memory allows both the GPU and the CPU to access each other’s memory and lastly, the NVLink which is “a fatter pipe” between the CPU and the GPU and makes for faster data flow at more than 80GB per second. The Pascal GPUs are named after the mathematician Blaise Pascal.
According to an earlier report in Tech Times, the GPUs are already undergoing testing and validation. The GPUs are said to be undergoing testing at the company’s testing facilities in India after being transported from Taiwan, where the GPUs were fabricated by the Taiwan Semiconductor Manufacturing Company.
The report also noted that the GPUs will be featuring new technologies such as Mixed Precision Support and High Bandwidth Memory 2 (HBM2). Four GPUs were reportedly undergoing testing, with serial numbers starting with 699. According to the Tech Times report, the first three units in question looked like “derivatives” or “variants” while the fourth unit is said to be the largest and “most powerful” among the four. While the Tech Times report claims that this may be the GP100 GPU, other reports are saying that there is no way to know for sure if this is the case. The GP100 GPU looks to be the flagship GPU of NVIDIA.
Among the highlights of the GP100 GPU are improved performance per watt compared to Maxwell, DirectX 12 feature level 12_1 or higher, twice the number of transistors compared to GM200, and it is also built using the faster 16nm FinFET manufacturing process of TSMC.
The Tech Times report likewise mentioned that NVIDIA will be revealing more about their Pascal GPUs at the GPU Technology Conference.