Monday, September 5th 2022
NVIDIA GeForce RTX 4080 Comes in 12GB and 16GB Variants
NVIDIA's upcoming GeForce RTX 4080 "Ada," a successor to the RTX 3080 "Ampere," reportedly comes in two distinct variants based on memory size, memory bus width, and possibly even core-configuration. MEGAsizeGPU reports that they have seen two reference designs for the RTX 4080, one with 12 GB of memory and a 10-layer PCB, and the other with 16 GB of memory and a 12-layer PCB. Increasing numbers of PCB layers enable greater density of wiring around the ASIC. At debut, the flagship product from NVIDIA is expected to be the RTX 4090, with its 24 GB memory size, and 14-layer PCB. Apparently, the 12 GB and 16 GB variants of the RTX 4080 feature vastly different PCB designs.
We've known from past attempts at memory-based variants, such as the GTX 1060 (3 GB vs. 6 GB), or the more recent RTX 3080 (10 GB vs. 12 GB), that NVIDIA turns to other levers to differentiate variants, such as core-configuration (numbers of available CUDA cores), and the same is highly likely with the RTX 4080. The RTX 4080 12 GB, RTX 4080 16 GB, and the RTX 4090, could be NVIDIA's answers to AMD's RDNA3-based successors of the RX 6800, RX 6800 XT, and RX 6950 XT, respectively.
Sources:
Wccftech, MEGAsizeGPU (Twitter)
We've known from past attempts at memory-based variants, such as the GTX 1060 (3 GB vs. 6 GB), or the more recent RTX 3080 (10 GB vs. 12 GB), that NVIDIA turns to other levers to differentiate variants, such as core-configuration (numbers of available CUDA cores), and the same is highly likely with the RTX 4080. The RTX 4080 12 GB, RTX 4080 16 GB, and the RTX 4090, could be NVIDIA's answers to AMD's RDNA3-based successors of the RX 6800, RX 6800 XT, and RX 6950 XT, respectively.
61 Comments on NVIDIA GeForce RTX 4080 Comes in 12GB and 16GB Variants
but system memory is just so cheap these days, it makes very little sense to not go for at least 32gib unless again it's like a hard budget build. but if you're like buying a 3070 you might asw spring for the 32gib of RAM.
It's not on a different die, there is no way AD102 based on the 12GPC rumor to be cut-down so much in order to fit AD103 performance, they would have to fully disable 4GPC which ain't happening.
Probably it's the same AD103 but I doubt the name would be the same, it's just there to confuse whoever want to figure out Nvidia model placement.
Probably a model sitting above Full AD104 (If Full AD104 is 4070 then this is 4070Ti for example)
According to rumors up till now, 4080 was a cut-down AD103 with 9728 Cuda cores with 256bit bus and 21Gbps GDDR6.
Before one-two weeks we had the rumor regarding 23Gbps memory, which correlates very good with a cut-down 6GPC AD103 model because the extra memory bandwidth (23/21) will be needed with a 192bit bus since the ROPs cutting is only 15% (7/6).
Of course it can have the same 7 GPC 9728 Cuda cores as the previous rumor (or 7 GPC with 10240 Cuda cores) supported by 192bit bus with 23Gbps GDDR6X instead of 256bit bus with 21Gbps but will be bandwidth limited.
Below what the rumors were suggesting regarding Ada architecture/AD103 model:
BTW for RDNA3 AMD is increasing bus width after telling us IC was the answer. 7900 394 bit, 7800 320 bit, 7700 256 bit. Nividia is cutting bus width other than on 4090 series. AMD realised with 4K gaming IC wasn't enough as they watched often large 1440p leads reversed in 4K.
RDNA3 is looking better all the time.
I'm of much the same nature, the fact some idiots overpay for GPUs doesn't mean I somehow need to feel forced to do the same. I just get those off their hands at half price a bit later. NP, ty. Its the same in every market, those who want the latest greatest are paying premium. To each their own.
A good deal is one where you're deciding what it is, not someone else deciding it for you or 'circumstance' somehow giving you FOMO. Zero chance of whatever being there at what price? We'll see. People are just going to sit on their money, it has happened in the PC and in the gaming world quite a few times already, alongside all those who overpay for product just the same. Both groups are relevant.
And its fun looking back on those good deals too. I'm still using a 2017 bought GTX 1080; playing anything I really want to play to date, even at 3440x1440 as the GPU survived the move from 1080p to it just fine. Still smooth frames and even > 80 FPS in most stuff I find worthwhile. And then realizing I paid under 500 for that GPU. That's value ;) And its still running as it did on day one, too - and what did I miss over the years since 2017? A few ray traced light sources in a handful of utterly forgettable titles? Its hilarious what a little patience and restraint has on offer in that sense ;)
Another thing you might not realize proper is the fact that we're about to crash into a recession, inflation is high and any entertainment-based market is going to feel the biggest blow as large consumer groups are forced to devote time and money to survival instead of 'fun'. GL with 1K GPUs in that environment. The time of free money is behind us...
Super uses 256 bus and 8GB
12GB uses 192 bus and clearly, 12GB.
Both have the same specs in terms of clock speeds (1470/1750) and memory types (GDDR6) but the Super is only around 6% faster.
I couldn't imagine a 4080 12GB being that far behind a 4080 16GB. I think it would just be stupid if they dropped both models. How are they going to leave room for a Ti or Super model down the road if they're hogging all the small stepping stones of performance right out of the gate?