Radeon RX 7950 XTX a third faster than Radeon RX 7900 XTX? Unreal

GPU released last year Navi 31, which is the basis of the Radeon RX 7900 XT and 7900 XTX, did not reach the expected performance. There are two reasons. On the one hand, unrealistic expectations that came from the misinterpretation of data on the increase in arithmetic performance (mistakenly interpreted as an increase in gaming performance) and on the other hand, from the fact that not everything turned out as AMD expected.

source: AMD

Two elements seem to have gone as planned. On the one hand, the clock frequencies, which even according to reports from AMD’s partners should have been higher, and on the other hand, the performance per clock. Every GPU (and this applies to chips in general to some extent) is put to work in some way after the first samples are made, and the drivers are tuned so that the combination of hardware and software works stably and at the highest possible performance. Rumor has it that AMD went the classic way with is Navi 31, solved the problems from the apparently most fundamental, and when most were solved, proceeded to mass production. But hey, it turns out that a seemingly minor problem that didn’t affect stability, just manifested itself with occasional artifacts, cannot be solved by software. In the end, the solution came, but higher percentages of performance fell on it than AMD would have liked. It was the involvement of the entire software department to discover a solution that would have a lower performance impact that should have caused the failure of a whole series of drivers for older generations of Radeons. A solution that would not impact performance was apparently not found, so it was released with the performance of the Radeon RX 7900s. While no specific number has been released, there are indications that the patch could cost around 10% of the performance.

In the shortly announced AMD ROCm 5.6, one of the users noticed the names of graphics cards not yet released. Most of them are known, but Radeon RX 7950 XT and Radeon RX 7950 XTX are new:

• AMD Radeon™ RX 7950 XTX | RDNA3 | gfx1100
• AMD Radeon™ RX 7950 XT | RDNA3 | gfx1100
• AMD Radeon™ RX 7900 XTX | RDNA3 | gfx1100
• AMD Radeon™ RX 7900 XT | RDNA3 | gfx1100
• AMD Radeon™ RX 7800 XT | RDNA3 | gfx1101
• AMD Radeon™ RX 7700 XT | RDNA3 | gfx1102
• AMD Radeon™ RX 7600 XT | RDNA3 | gfx1102
• AMD Radeon™ RX 7500 XT | RDNA3 | gfx1102
• AMD Radeon™ RX 7600M XT | RDNA3 | gfx1102
• AMD Radeon™ RX 7600M | RDNA3 | gfx1102
• AMD Radeon™ RX 7700S | RDNA3 | gfx1102
• AMD Radeon™ RX 7600S | RDNA3 | gfx1102

It is true that it was originally reported that AMD would prepare a new kernel revision Navi 31which will fix some problems, however, news soon came that the company would not waste the time, money and manpower that it would have to transfer from RDNA 4 development. However, a Radeon RX 7950 XTX performance graph appeared in the TechPowerUp database, which the editors of this site build 32 % over current Radeon RX 7900 XT:

It’s hard to say how she arrived at that, but she probably speculated based on supposedly higher rates and added some percentages on top. In reality, however, there are no sources from which it would be possible to draw similar conclusions. AMD is not known to have dedicated resources to addressing 3GHz+ clocks. It is not known to devote resources to a hardware solution to the bug with artifacts, the removal of which claimed a significant part of the performance. And it also cannot be said that even if AMD dealt with both, that the problems would have been 100% solved with the limited human resources in the specified time.

Therefore, this graph seems more like wishful thinking. It cannot be ruled out that AMD will release Radeon RX 7950 XT and RX 7950 XTX, but rather than a complete redesign of the chip, it would be a matter of new binning (selection of cores according to stricter parameters that can be afforded with regard to tuning the production process), at most an improvement of the situation by modifying the design of the metal layers. Perhaps the application of a better software solution that may have appeared eventually. Such changes can result in a performance shift of somewhere between 5-15%. If AMD had a product in its pocket that would bring a third increase in performance, it would be a marketing mistake to present it other than as a new generation (Radeon RX 8000), because fifty extra would not really correspond to it.

But as said, such a large-scale solution would be risky (requirements for human resources, time and finances could negatively affect the development of RDNA 4) and, moreover, in the current economic situation, when sales of graphics cards with prices over $1000 are falling, it would not make sense to preparations >$1000 product to invest.