It should be pointed out that although the “full blood version” AD103 GPU has 64MB L2 cache + 224 ROP raster units, the reduced specification may be changed to 48MB L2 + lower ROP.
The GPU frequency has not yet been determined, but TSMC’s 4N process can easily be set in the 2.0 – 3.0 GHz range (the previous generation Ampere GPU was affected bySamsung10nm optimized version of the “8nm” process drag).
For reference, the NVIDIA Pascal architecture also achieved a huge leap in inter-generational performance through TSMC’s 16nm Dongfeng.
In terms of video memory, the RTX 4080 may use 256-bit @ 16GB 21Gbps GDDR6X specifications. In addition to the bandwidth of up to 672 GB/s, the graphics card is also expected to squeeze higher potential through GPU frequency tuning.
(screenshot viaWCCFTech）
In terms of benchmark performance, it is conservatively estimated that the 3DMark “Time Spy” project of the RTX 4080 can be rated at 15,000 points – about 35% higher than Ampere’s flagship RTX 3090 Ti.
In terms of functionality, it is expected that the RTX 4080 will integrate all modern features – including 4th generation tensor / 3rd generation ray tracing core, NVENC / NVCDEC codec, and support for the latest API and other RTX features.
Finally, NVIDIA may officially announce the starting lineup of the RTX 40 series around the fourth quarter of 2022 – the first quarter of 2023.
In terms of pricing, the RTX 4080 is expected to be between $699 – $799 (about 4715 ~ 5390 RMB).
