Before the start of sales of new game consolesGeneration Xbox Series X from Microsoft is only a few months away and developer representatives do not cease to provide more and more information about the future hit of sales. Another piece of technical information about the architecture of the Xbox Series X processors was provided at the HotChips 2020 conference, held in Silicon Valley this month.
It was stated that the console usesProject Scarlett chipset, manufactured by TSMC using 7nm process technology. The SoC used received a die area of 360.4 square millimeters and has 8 computing cores based on the AMD Zen2 architecture with support for 16 instruction streams. The processor uses 52 graphics computing units based on the AMD RDNA2 architecture.
The processor received 15.3 billion transistors, which is1.7 billion more than the NVIDIA Turing TU104 GPU, which is manufactured using 12nm technology and has an area of 545 square millimeters. For comparison, the Xbox One X processor had 6.6 billion transistors, and its area was 367 square millimeters.
Microsoft developers also said that inthe processor uses 28 Dual Compute Units, two of which are disabled. As a result, the total number of computing units is 4 less than originally implemented. The GPU occupies about 47.5% of the die area. The RDNA2 GPU block diagram confirms the presence of dedicated Ray Accelerators capable of four texture or four ray operations per clock cycle.
Developers also see significant growththe complexity of the production of new processors, which entails an increase in production costs. The use of innovative solutions allows us to solve emerging difficulties. In the process of improvement since 2013, Xbox consoles have been able to increase performance almost 10 times, while the increase in power consumption has been negligible.
Microsoft representatives told about supportray tracing technologies. The use of Microsoft DirectX Ray Tracing (DXR) is similar to NVIDIA RTX methods, which does not completely replace ray tracing, but implements a simplified hybrid design. This approach allows for better visualization using existing performance. According to Microsoft, this approach can be called "a cost-effective upgrade to traditional rendering."
It was also announced about the use of supportspeed up machine learning tasks. At the same time, the developers do not indicate whether the chip used tensor cores similar to NVIDIA cores. As a result, it can be assumed that the acceleration is provided by the INT4 and INT8 operations performed on the Stream cores.