Rumors about Nvidia’s Ampere and the proposed family of consumer GPUs based on it have been circulating for months. The latest leaks claim Nvidia is prepping some significant improvements at the top end, along with higher power consumption.
According to Igor’s Lab, which also leaked supposed images of the new Ampere GPUs this week, the upcoming top-end cards will be the RTX 3090 (possibly branded as Ti or Super), followed by the RTX 3080 (Ti/Super), followed by the RTX 3080. All three of these GPUs would be based on the GA102 die, and all of them would use 300W+ of TDP — 350W for the 3090, and 320W for both of the 3080 models below it. RAM would be GDDR6X in all cases, with a 384-bit, 352-bit, and 320-bit interface on each GPU respectively. The top-end 3090 would carry NVLink, but the lower-end GPUs wouldn’t.
The framing of these GPUs makes me wonder if Nvidia is launching its absolute top-end market stack first. With Pascal, Nvidia led with the GTX 1080 and 1070, with the 1080 Ti debuting months later. For Turing, Nvidia launched the RTX 2080 Ti, 2080, and 2070 simultaneously, but used a different GPU for each. This positioning sounds like Nvidia will lead with what we’d have typically called a “Titan / xx80 Ti / xx80” positioning as opposed to “xx80 Ti / xx80, xx70.”
The RAM loadout is also interesting. With consoles now packing 16GB of unified RAM and some high-end GPUs like the Radeon VII already featuring 16GB, I think there’s been a certain amount of assumption that 16GB would be the RAM capacity of choice next generation. This data suggests otherwise. The 24GB of VRAM on the 3090 Ti/Super is a nod to the card’s datacenter/workstation roots, not an attempt to move the market towards higher VRAM loadouts.
Typically Nvidia and AMD match each other on RAM capacity fairly closely, though it isn’t unusual for AMD to offer a bit more memory bandwidth and capacity depending on the SKU. Team Red has been warning about small VRAM buffers of 4GB and below, which both plays to its strengths in the market and aligns with trends showing that lower RAM levels can now meaningfully impact 1080p play at a high detail level.
If the TDPs are to be believed, Nvidia is also finally leaving the 250W TDP point behind at the high end. Both Nvidia and AMD have flirted with higher-power GPUs before, but 250W has been an anchor point for GPUs in much the same way that 125W TDPs were an anchor for consumer CPUs for many years. Intel and AMD have both exceeded that mark in recent years, and if these rumors are accurate, we should expect GPUs to do so as well. This would free AMD to essentially pursue the same path.
An increased TDP isn’t necessarily surprising. Nvidia may have chosen to maximize performance at the top end, gambling that high-end gamers who would consider these cards in the first place have systems powerful enough to handle them. If you have an 850W – 1.2kW PSU and adequate cooling, a 250W CPU and 350W GPU won’t be anything you can’t handle in the first place.
No word on pricing, but the one thing you can bet these cards won’t be is cheap. Nvidia may position them competitively relative to where Turing or Pascal debuted if it feels AMD is a threat or if it’s worried about the impact of coronavirus on GPU sales, but I’d expect the company to hold the line on pricing to the greatest degree possible.
- AMD Declares 4GB of GPU VRAM ‘Not Enough’ for Today’s Games
- Level Up: Nvidia’s GameGAN AI Creates Pac-Man Without an Underlying Game Engine
- Nvidia Sued by Angry Investors for Misclassifying Crypto Revenue