Strong demand for 5nm wafers coming out of TSMC has pushed AMD to manufacture a trim on their RX 7000. Specifically in its specifications. The reason behind this is that they are sharing space with NVIDIA and it is important to have enough units produced to be able to sell them later. How have the plans been affected compared to the ones you had a few months ago?
If a few hours ago we told you that the RX 7900 XT could reach 92 TFLOPS, now we have to contradict ourselves with this news. And it is that the latest information on the RX 7000, based on the RDNA 3 architecture, indicates significant cuts compared to what was initially planned by Lisa Su’s company regarding its next generation of graphics cards. Of course, it should be clarified that until now the information that appeared had no reliable authority. The reason? Easy, it does not come from AMD itself, but from the various information leakers.
Although the RX 7000 will receive cuts from what was originally planned, they will be a significant leap over the current generation. In addition, the two best performing configurations will be the first two chip graphics cards. Ahead of NVIDIA and Intel when launching a graphics card disaggregated into several chips. Something inherited from their desktop Ryzen CPUs, but we haven’t seen how it affects desktop GPU performance yet.
Changes in the configuration of future AMD RX 7000 graphics
First of all, keep in mind that months have passed since the leak of the AMD RX 7000 specifications based on its future RDNA 3 architecture. At the moment, we have no evidence of a graphics card finished, so a much later version than the NVIDIA RTX 40 is expected.
What was the surprise? Well, the fact that the various GPUs in the line, Navi 31, 32 and 33, received a considerable cut. In what? Well, in the number of ALUs in FP32 in each of the shader units. Which means that:
- top of the line, Navi 31is passed from 15,360 ALU FP32 in total at 12,288. This number means 48WGP throughout the processor spread over 12 shader matrices in 6 shader engines.
- As for the mid-range, Navi 32the cut went from the 10,240 cores of the specs leaked last year to the 8.192. Which translates to 32 WGPs in 8 shader matrices inside of 4 shader engines.
- And to finish with the low range, Navi 33 would have half the units of the previous case. So, 4.096 ALU FP32 in 16 WGPs inside of 4 shader matrices who are inside 2 shader engines.
This cut in the RX 7000 means the RX 7900 XT has gone from being rated to 92 TFLOPS to 73 TFLOPS. This if the speed of 3GHz and AMD decides not to make any last-minute changes to the GPU, which isn’t finished yet. We should therefore expect changes in this aspect as well. Has AMD ceded the throne to NVIDIA for the umpteenth time or can there still be surprises?
