AMD released its RDNA 3-based RX 7900 series cards based on the Navi 31 die on November 3. The RX 7900 XTX and 7900 XT cards are priced $999 and $100 cheaper than Nvidia’s $1,199 RTX 4080 16GB model, which is now just known as RTX 4080 because Nvidia discontinued or rather “unlaunched” the 12GB edition. According to tipsters, the less powerful 4080 is purportedly changing into the RTX 4070.
Since then, AMD has released additional information about the RX 7900 series’ performance, including benchmarks for the 7900 XT and additional jabs at Nvidia. Using that information, we estimated the Radeon RX 7000’s performance level and compared it to the RTX 4080 and 4090.
In summary, the predicted performance was right where you would have expected it to be, with the 7900 XTX’s raster output falling between the 4080 and 4090, but its performance in ray tracing lagged both of the RTX 40-series GPUs.
Today, we have the first leaked synthetic benchmark results for the RX 7900 XTX, which will be available in a week. However, given that the 7900 XTX appears to have produced very dismal results, AMD fans will undoubtedly be extremely upset. The leaked data is in the form of Geekbench scores, showing that Nvidia’s RTX 4090 dominates the OpenCL computing test by a huge margin of 143.6%. The RTX 4080, the 7900 XTX’s major rival, is currently 65.5% ahead. The GPU from AMD has scored 150,372 points.
The test systems, which were both powered by a Ryzen 9 7950X CPU, are quite comparable and nearly identical, as can be seen in the photographs below.
The RX 7900 XTX’s Vulkan score is also not particularly impressive. The RTX 3090 Ti, 3090, and AMD’s own RX 6900 XT and 6800 XT from the previous generation all scored more than the XTX, which scored 91,652 points.
These are, of course, pre-release figures, so they might not indicate much, and the outcomes on launch day might be entirely different. The data, however, is alarming, particularly if you support AMD or desire competition in the GPU market. The next RX 7900 XTX and 7900 XT are expected to be quite good, so let’s hope the sample(s) tested here are merely flawed engineering samples or the drivers are messed up.