New RTX GPUs
- Thread starter Tyres O'Flaherty
- Start date
Which part?
There are 24Gb GDDR7 ICs in production, which means 24GiB of memory with eight ICs in a 256-bit wide configuration, which matches the GB103 (5070 Ti and 5080) parts. So, it seems almost a given that there will be 24GiB incarnations of these parts at some point in the not too distant future. Likewise, 18GiB GB205 parts are also very likely.
As for the idea that 16GiB is not enough for 4k...that's mostly false, currently. More VRAM, all other things being equal, is better, but once you're at 16GiB paying more just for more VRAM is highly dubious. It might be justifiable for a 5080, or other near fully enabled GB103 part, but anything less than that will almost certainly be running out of raw GPU performance before VRAM contention becomes a major issue. I don't expect more than 16GiB of VRAM to be broadly useful until at least 2027, but there are always exceptions.
More memory doesn't make a GPU faster, except in areas where it's actually running out of memory...which is quite rare with 16GiB cards.
Additionally, the RTX 5080 is already a fully enabled GB203 part, so about the only thing they could do to improve performance without using an entirely different GPU and PCB (i.e. a heavily cut down GB202) is a modest clock speed bump and power limit increase. This is no where near enough for the anticipated RTX 5080 Super to match RTX 4090 performance in any apples-to-apples comparison.
Additionally, the RTX 5080 is already a fully enabled GB203 part, so about the only thing they could do to improve performance without using an entirely different GPU and PCB (i.e. a heavily cut down GB202) is a modest clock speed bump and power limit increase. This is no where near enough for the anticipated RTX 5080 Super to match RTX 4090 performance in any apples-to-apples comparison.