Rumor: AMD reduces RDNA 3 architecture GPU specifications
As the Radeon RX 7000 series approaches, more news about RDNA 3 architecture GPUs flows out. Recently, multiple sources have shown that AMD has reduced the specifications of the Navi 3x series GPUs, and with the reduction of core specifications, the related expected performance will also be reduced.
Twitter user @Redfire75369 revealed that the number of WGPs of Navi 31 will be reduced from the previous 60 to 48, and the number of corresponding stream processors will also be reduced from 15,360 to 12,288, which means that the core size is reduced by 20%. The same is true for Navi 32 using the MCM multi-chip package. The number of WGPs will be reduced from the previous 40 to 32, and the number of corresponding stream processors will also be reduced from 15360 to 12288. The single-chip Navi 33 has not escaped the fate of being cut. The number of WGPs will be reduced from the previous 20 to 16, and the number of corresponding stream processors will also be reduced from 5120 to 4096.
Correspondingly, the expected performance of Navi 3x series GPUs will also be adjusted down. Previously, it was reported that the Navi 31’s FP32 computing performance will reach 92 TFLOPs at 60 WGP and 3GHz frequency, but with the reduction of the number of stream processors, the FP32 computing performance has also been reduced to 73 TFLOPs.
Recently, Twitter user @kopite7kimi said that the GeForce RTX 4090 based on the Ada Lovelace architecture may be able to achieve FP32 computing performance of 100 TFLOPs. To achieve such a performance metric, an AD102 with 18,432 CUDA cores must be clocked at least 2.7 GHz. Since the GeForce RTX 4090 is likely to use AD102 which is not a full core, the core frequency will be higher.
It is difficult to judge how real this leaked news is, and it also involves the use of ray tracing, super-resolution technology, and other functions, but both Nvidia and AMD seem to want to target 100 TFLOPs for their next flagships. It is said that the graphics cards based on the new architecture of both parties will have a considerable increase in performance.
It is difficult to judge how real this leaked news is, and it also involves the use of ray tracing, super-resolution technology, and other functions, but both Nvidia and AMD seem to want to target 100 TFLOPs for their next flagships. It is said that the graphics cards based on the new architecture of both parties will have a considerable increase in performance.
