This week, Microsoft released some specifications of Xbox series X, including a very surprising message: GPU with rdna2 architecture has the floating-point computing power of 12tflops.Although the specific number of stream processors is not mentioned, it will never be the 2560 stream processors previously rumored.
For the purpose of power consumption control, even if the same GPU chip is used, the GPU frequency of the host computer will be much lower than that of the desktop graphics card. For example, Xbox one uses the first generation of GCN architecture, with 768 stream processors, and the frequency is only 853mhz; Xbox one X has a GPU frequency of 1170mhz under the support of 14nm process, while the rx590 frequency on the desktop is as high as 1545mhz.
1. What is the number of Xbox series X stream processors?
To achieve the floating-point performance of 12 TFlops, the Xbox Series X can only choose the following frequencies to match the stream processor:
1.3584 1675 MHz frequency stream processors;
2.3 840 stream processors 1565 MHz;
3.4096 stream processor 1465 MHz frequency.
the desktop version of the rx5700 with 2304 stream processors at a frequency of 1725 mhz, the power consumption reaches 180 w. Considering that the Xbox Series X is also added to the hardware optical pursuit unit, it is absolutely impossible to control the power consumption frequency to reach 1675 MHz, otherwise so many stream processors, the power consumption will necessarily explode.
If it is 3840 stream processors, the frequency of 1565mhz is still high for the host. Then there is a considerable possibility:Xbox series X has 4096 stream processors, which can control power consumption at 1465mhz。
2. Is Xbox series X just twice as good as Xbox one X?
The floating-point performance of Xbox series X is as high as 12tflops, while the floating-point performance of Xbox one X is 6tflops, so it appears that the performance of Xbox series X is twice that of Xbox one X!
Is that true?
In fact, we all know that the floating-point performance is only a theoretical value. The IPC performance of rdna2 architecture is far better than that of the old GCN architecture. The performance gap between them is not only a simple addition and subtraction of floating-point data!
For example, Vega 64 water-cooled version has a floating-point performance of 13.7 tflops, but the game performance is not as good as RX 5700, which has a floating-point performance of only 7.9 tflops. The same radeon RX 580 floating-point performance is up to 6.2t, 27% worse than the RX 5700, but the game performance gap is 63%.
Simply put, under the same floating-point performance, the game performance of rDNA architecture is about 127% of that of GCN 4.0, considering that Xbox series X adopts rDNA 2 architecture, IPC performance should be a little higher.
Now let's convert,Xbox series X should have more than 2.6 times the real game performance of Xbox one X.
With 4096 stream processors and rdna2 architecture, this is where Microsoft dare to say that Xbox series X can run 3K @ 60fps!