Full-blooded 18,000-core AMD RX 8000 graphics card will bet on ray tracing and AI: tough RTX 50

Quick Technology News on June 8th, AMD’s RDNA architecture GPU has developed for three generations. Last year’s RX 7000 series used the latest RDNA3 architecture, claiming that the performance per watt has increased by 50%, but the final performance is hard to describe.

The performance of the flagship RX 7900 XTX is not only incomparable with the RTX 4090 of its competitors, but also cannot even beat the RTX 4080. If light tracing is turned on, the gap will be even greater. It is normal to be ahead by more than 50% in performance.

The FSR 3 technology, which is regarded as the killer feature, was originally designed to benchmark DLSS 3, but it has not been officially released until now, and the number of games supported is incomparable with N cards.

In addition to game performance, RDNA3 has no presence in AI performance. The energy efficiency is said to be greatly improved. The TDP power consumption of 335W is 95W lower than the 450W of RTX 4090. In practice, there is no advantage.

In short, RDNA3 is very good and powerful in theory, but compared with the RTX 40 series, it is really hard to accept. N-cards are criticized for being expensive, but besides being expensive, they have other advantages.

A fan’s hope can only be the next generation RDNA4 architecture, which is the RX 8000 series. According to the schedule, it should be released in 2024. It is unlikely that the process technology will be upgraded to 3nm. It should be an improved version of the 5nm 4nm process. , but it does not rule out AMD’s radical 3nm process.

More importantly, the design ideas of RDNA4 will change. NVIDIA has successfully proved that light tracing and AI are important. AMD has to follow. The new architecture will be greatly changed, the RT light tracing unit will be reset, and the AI ​​performance will also change. It will be the highlight - after AMD acquired Xilinx, there is no shortage of advanced AI technology and ecology.

RDNA4 will continue to use the small chip design of RDNA3, and the scale of the computing core GCD and IO core MCD will be increased. Up to 144 sets of CU units, 18432 cores at full health, and the video memory can reach 32GB, or 48GB , depends on the MCD quantity.

The calculation scale of RNDA4 has been increased by 50%, and the frequency has been optimized to 3.5GHz. Theoretically, the performance has doubled. However, the figures in this regard are not accurate. Some say that the performance has been improved by at least 50%. There is also a 50% improvement in energy efficiency to ensure that power consumption does not get out of control.

If you remember the early revelations of RDNA3 last year, you should understand that this RDNA4 is the full-blooded version of the original RNDA3. We can only hope that AMD will really make up its mind to develop high-performance graphics cards.

Nvidia will also launch the RTX 50 series in 2024, codenamed Blackwell. It is said that the performance will be greatly improved. Compared with the current improvement of 60%, it will not be a problem. The pressure of AMD RX 8000 to surpass the RTX 50 is not small at all.

The above is the detailed content of Full-blooded 18,000-core AMD RX 8000 graphics card will bet on ray tracing and AI: tough RTX 50. For more information, please follow other related articles on the PHP Chinese website!