AMD Radeon fans could be in for a massive surprise in a good way as the company may be working on some insanely powerful next-gen RDNA 3 GPUs. According to the latest report by Twitter leakster Greymon55, the top of the line RDNA 3 (Navi 31) graphics card could come in with close to 92 TFLOPs of single-precision (FP32) compute power. This number is nearly four times more than AMD"s current best offering, the RX 6900 XT, which is capable of delivering 23.04 TFLOPs of FP32.
≈92T FP32
— Greymon55 (@greymon55) April 30, 2022
The 92 TFLOP number is reached using a 3GHz boost clock for next-gen flagship Navi 31 GPU, which might be the Radeon RX 7900 XT. Navi 31 is rumored to feature a total of 120 60 Workgroup Processors (WGPs) or 240 120 Compute Units (CUs) for a total of 15,360 Stream Processors. This is how the rumored specifications compares with the RX 6900 XT:
| RX 6900 XT | RX 7900 XT | |
|---|---|---|
| Workgroup Processors (WGP) | 40 | |
| Compute Units (CUs) | 80 | |
| SIMD32 | 160 | 480 |
| Stream Processors | 5,120 | 15,360 |
| Boost Clock | 2,250 MHz | 3,000 MHz (?) |
| FP32 TFLOPs | 23.04 | ~92 |
Additionally, earlier today, a driver leak suggested that AMD"s RDNA 3 architecture is moving to a four SIMD32 per CU design up from two SIMD32. Hence, the total final performance of the 7900 XT may even end up being more than four times that of the RX 6900 XT.
Source: Greymon55 (Twitter)
Edit: The original article says that the next-gen Radeon RX 7900 XT could be four times faster than the 6900 XT. However, this was made with a wrong information about the Workgroup Processors (WGP) being 120 instead of 60. Multiplying this with the new four SIMD32 per CU approach, we get the 15,360 Stream Processor figure in the above table for RDNA 3.
The error has been fixed and as it stands, the specs suggest that the card will be twice as fast if not more than the current RX 6900 XT.