At Intel’s FAD, CFO George Davis stated that the company’s 10nm node is going to be less profitable than the preceding 14nm process that sustained over four generations of consumer and server CPUs. At the same time, it’s being reported that the new node will never quite reach the same level of production as the existing 14nm or the 22nm processes. That leaves Intel’s upcoming Xe graphics cards in a precarious situation. As the 10nm Foundries won’t be ramped up to required levels and 14nm is still facing shortages, that leaves TSMC as the logical partner for fabbing Intel’s first proper graphics cards.
According to Technews Taiwan, Intel will leverage TSMC’s upcoming 6nm process for fabbing its post-DG1 consumer Xe graphics cards. That’s not the end though. Sources close to the channel report that the partnership might continue with Team Blue eyeing TSMC’s 3nm node as well.
DG1 is expected to perform on the same level as the GeForce GTX 1050 Ti with much better half-precision compute capabilities. Although Turing is NVIDIA’s first consumer GPU with uncompromised mixed compute capabilities, Xe might even outperform its latest GeForce competitors. DG1 is slated to launch at the end of 2020 with 96 EUs based on the Gen12 design. Each EU packs four ALUs with a base clock of 1GHz and a boost of 1.5GHz. It will be paired with 3GB video memory and feature a max TDP of just 25W. DG1 is going to be the only Intel GPU to be fabbed on Intel’s 10nm process, while DG2 will use TSMC’s 6nm node.
Intel Gen12 Xe Graphics Processors Will Have Potent FP16 Capabilities, Almost on Par with the NVIDIA GTX 970
DG2 is expected to be a high-performance graphics card competing with the likes of the GeForce RTX 2060 and the Radeon RX 5700. It has been reported multiple times that the GPU will be based on TSMC’s 7nm node but it turns out that it’ll instead be based on the more enhanced 6nm EUV process.
Ponte Vecchio expected to be based on Intel’s 7nm process and should come out by late 2021 or early 2022. While it’s being claimed that the accompanying consumer GPUs will leverage TSMC’s 3nm process, it’s not 100% certain.
Lastly, Intel is also planning to research chiplet based GPUs using TSMC’s 6nm node in 2021. Given TSMC’s history with NVIDIA and AMD, they’re the ideal partner for the job.