AMD has made many comparisons slides between its Radeon RX 7900 flagship graphics card and NVIDIA’s upcoming RTX 4080 16GB GPU available (thanks to Tom’s Hardware via VideocardZ). This not only demonstrates that AMD is aiming at the performance of the RTX 4080 with its Radeon RX 7900 graphics cards, but it also indicates that the GPU wars are heating up to a level unseen since the Hawaii GPU era.
The Radeon RX 7900 XT and XTX are compared to the forthcoming RTX 4080 on the first presentation by AMD. It is comparing RDNA3 computing units to NVIDIA SMs based on the slides, which is a fascinating comparison to do. AMD is also claiming that 1 of their shader units equals 2 CUDA cores because RDNA3 has 64 shading units per compute unit and Ada Lovelace (RTX 4080) has 128 CUDA cores per SM.
Next, a comparison of GDDR6 memory is made, with AMD easily winning with 20 and 24 GB of GDDR6 memory compared to the RTX 4080’s 16 GB. In contrast to the RTX 4080’s 256-bit memory bus width, AMD GPUs have wider memory bus widths of 320 bits and 384 bits. Additionally, we compare the raw performance metric, where shader TFLOPs on the RTX 4080 are 49, whereas they are 52 on the RX 7900 XT and RX 7900 XTX. Following a price comparison of the SEPs, AMD reminds everyone that their cards enable DisplayPort 2.1 while NVIDIA only supports 1.4.
The Radeon RX 6950 XT and RX 7900 XT are both shorter and smaller than the RTX 4080, according to AMD.
While AMD will continue to use the common 2x 8-pin connectors that have been the backbone of GPU power for years, the RTX 4080 requires the new and potentially troublesome 12VHPWR socket.
AMD’s RX 7900 XT and XTX will also enable a maximum video bandwidth of 54 Gbps using UHBR 13.5 while NVIDIA will max out at 32.5 Gbps, and they are keen to point out that NVIDIA does not. In contrast to NVIDIA’s ceiling of 8K at 60 Hz and 4K at 300 Hz, AMD’s RX 7900 line of graphics cards will be able to support 8K at 165 Hz and 4K at 480 Hz.
Additionally, AMD released some performance benchmarks, showing that it was able to outperform the RX 6950XT by 67% in rasterization loads and up to 82% in ray tracing loads. They have also included the FSR ON/OFF numbers as stacked bar graphs in their performance chart, taking a cue from NVIDIA.
Also read:
