Suspected public RTX 4080 spy photos
We’ve heard about the “official” SKU of the RTX 4080 with 16GB of video memory before, but the 12GB version remains a mystery. For reference, the RTX 3080 is also available in 10/12GB versions with 384/320-bit video memory widths, respectively.
It is rumored that the memory bit width of the RTX 4080 16GB graphics card is 256-bit, so it can be speculated that the 12GB version will be cut to 192-bit. For the new high-end graphics card, Lao Huang’s knife method has really made many potential consumers. Feel helpless.
However, in order to make up for the GPU’s bandwidth shortcomings, NVIDIA may use higher memory frequency + memory compression technology-recent rumors that the RTX 4080 uses 23Gbps GDDR6X memory.
Of course, it’s not ruled out that Nvidia is feinting a shot with so-called “engineering samples” – like the 16/20GB RTX 3080 models that have been seen in the wild before – they are all manufactured by the company, but ultimately not into the retail market.
In terms of specific specifications, the RTX 4080 is expected to use the shrunk AD103-300 GPU, with 76 SM units / 9728 CUDA cores, while the full-blooded version is 80 SM / 10240 cores.
Also, while the full chip has 64MB L2 and up to 224 ROPs, the RTX 4080 may only have 48MB L2 and lower raster units with a “PG136/139-SKU360” PCB.
In terms of video memory, according to the 23Gbps GDDR6X specification of 16GB @ 256-bit, the bandwidth can reach 736 GB/s – lower than the 320-bit wide RTX 3080 10GB graphics card, but it is expected to introduce next-generation video memory compression technology.
In terms of power consumption, it is reported that the nominal TBP of the RTX 3080 is 340W, an increase of 20W from the earlier rumored 320W, and it is at the same level as the RTX 3080 (up to 350W). (Micron24Gbps GDDR6X memory in full mass production)
As for the RTX 4090, it will enable 128/16384 CUDA cores out of 144 SM units, complemented by 96MB L2 cache/384 ROP raster units.
As for the GPU frequency, although it has not been disclosed, given that Nvidia has switched to TSMC’s 4N process, we speculate that it will fall between 2.0 – 3.0 GHz.
In addition, the RTX 4090 is expected to be paired with 384-bit @ 24GB of 21Gbps GDDR6X memory, and the bandwidth is comparable to the RTX 3090 Ti (up to 1 TB/s).
In terms of power consumption, it is said that the nominal TBP is 450W (the previous generation RTX 3090 is 350W), which means that the TGP may be lower than this value.
The newly introduced single 16-pin connector can theoretically withstand a maximum power of 600W, which means that the non-public version may exceed 500W like the RTX 3090 Ti.
Finally, Nvidia may launch the RTX 4090 around October 2022, but the RTX 4080/4070 may have to wait until early 2023.