On Intel’s 2021 Architecture Day, the silicon maker announced the company’s upcoming latest graphics architecture which is the Xe HPG. The company also announced that the first GPUs which will be released will be using the TSCM’s N6 process for manufacturing and will be first made available for launch in the first quarter of 2022.
With the launch of the Xe HPG GPUs in the market, this will be Intel’s re-entry into the independent GPU market after more than 20 years since it first released its i740 in 1998.
The GPU market is currently running very good and according to recent sources, the global GPU shipments reached 119 million in the first quarter of 2021, which is a year-on-year increase of 38.78% and a month-on-month decrease of 3%.
currently, Intel has a 68% market share in overall GPU shipments, and in the discrete GPU market, Nvidia has an absolute lead of 81% share, and AMD ranks second with a 19% share.
In terms of cloud, the edge, or the terminal, various electronic systems require high-performance image processing capabilities, GPUs have entered into an accelerated stage. Currently, the discrete GPU has become the leader in a wide range of uses and has reached the peak of the chip industry.
Nvidia released its flagship GeForce RTX 30 series GPU for the consumer market released in 2020 and the GPU uses Samsung’s 8nm process. Nvidia’s RTX3080 and RTX3090 contain 28 billion transistors on the other hand AMD’s RX 6000 series uses TSMC’s 7nm process and has reached 26.8 billion in transistors.
NVIDIA is currently reigning supreme with its Ampere architecture that includes Stream Processor (SP), Texture Mapping Unit (TMU), Tensor Core (Tensor Core), and ray tracing.
Currently, the GPU market is in fierce competition between NVIDIA and AMD, and the demand for GPUs is extremely high in the market. This is the current time for Intel to enter the market and get a piece of the pie.
Intel has been polishing its Xe architecture and has become solidly popular for many years, and to add to the cherry, it will use TSMC’s 6-nanometer process, which is fully capable of competing with NVIDIA and AMD.
However, “At the moment when ASIC and FPGA are competing with GPU, Intel chose GPU, indicating that GPU may still be the best choice for general AI.”