Nvidia unveils RTX 4090 – but it's the 4080 to watch out for

And don't forget AMD's RDNA 3 GPUs are right around the corner

GTC Nvidia CEO Jensen Huang unveiled his GPU giant's flagship RTX 40-series graphics cards at GTC today.

Powered by Nv's Ada Lovelace microarchitecture and a TSMC 4nm process, the RTX 4090 and 4080 are said to offer more than twice the performance of the previous 3090 TI and 3080 TI flagships.

Kicking things off with the RTX 4090, the unit bears more than a passing resemblance to 3090 TI Huang was comparing it to during his keynote presentation. The card features the same 24GB compliment of GDDR6X memory, including a 384-bit interface, and 450W TDP as the 3090 TI.

Beyond that, the new card’s specs suggest it will deliver substantially higher performance, presumably thanks to architectural and process node improvements. The 4090 packs 16,384 CUDA cores, up from 10,752 in the 3090 TI, while simultaneously, Nvidia has pushed base clocks from 1.67 GHz to 2.23 GHz.

In addition to more CUDA cores and faster clocks, the card — and the 4080 for that matter — also features Nvidia’s third-gen real-time ray-tracing cores and fourth-gen Tensor cores, as well as support for the Nvidia's DLSS 3 AI supersampling tech.

When Nvidia claimed the units will deliver about twice the performance of their predecessors, it was basing that on a carefully cherry-picked selection of games, which included Microsoft Flight Sim and Warhammer 40K Darktide.

The chip designer further claimed performance was four times greater when trying out a pre-release version of Cyberpunk 2077 using its next-gen AI supersampling tech. That's not to say these cards are only aimed at games: as GPU devices, they can accelerate plenty of other workloads.

But that oomph won’t come cheap. Nvidia is claiming an MSRP of $1,599 for the card, though it’s still cheaper than the 3090 TI when it shipped this spring with an MSRP of $1,999.

Of course, what with the dwindling world of GPU-based cryptocurrency mining, thanks to the Ethereum changes, and all this economic uncertainty causing people to buy less stuff, customers may stand a better chance of finding one of these 4090 cards at that suggested price when it arrives in mid-October. With that said, we’re not holding our breath.

Be careful not to buy the wrong 4080

Things get a bit more interesting when it comes to Nvidia’s mini-boss or rather mini-bosses: the 12GB and 16GB 4080s.

Nvidia compared the RTX 4080-series cards favorably to the outgoing 3080 TI, touting twice the performance in games over the Ampere-based card.

That's despite the 3080 TI holding a narrow CUDA core lead at 10,240 versus the 4080 16GB’s 9,728 CUDA cores. The gap grows even larger in the case of the 4080 12GB variant, which features 7,680 CUDA cores, roughly 2,000 fewer than its predecessor.

And that’s not the only way Nvidia has seemingly kneecapped the 4080. While the cards use the same GDDR6X memory as their predecessors, Nvidia has cut the memory interface down to 256 bits on the 16GB model and 192 bits on the 12GB version. Even the 10GB 3080 had a 320-bit interface.

Nvidia appears to have made up for this by making architectural changes and boosting the 4080’s clock speed, which now tops out at 2.51 GHz on the higher-end 16GB model, and 2.61 GHz on the 12GB one, all while claiming to have done so while using less power. Yay, TSMC?

The 12GB version claims a TDP of just 285W, while the 16GB version is rated to consume 320W. That’s roughly 9-19 percent less power than the 3080’s 350W TDP.

If Nv's numbers are to be believed – and some preliminary specs indicate the 4080 12GB at least does beat the 3080 and 3080 TI on FP32 performance and memory bandwidth – then the 4080 series should outperform the 3080 TI despite the reduced CUDA core count and memory bus width.

Just beware that there's a bit of a gap between the 4080 16GB and 12GB models. Both the 12GB and 16GB 4080s will launch in November at a suggested price of $899 and $1,199 respectively.

Nvidia waits on mid-tier 40-series cards while AMD preps RNDA 3

As with previous years, Nvidia is sticking with its flagship cards out of the gate. The GPU giant will presumably begin rolling out 40-series updates to the 3070, 3060, and 3050 cards over the next year. But for now, Nvidia’s mid-tier 30-series cards are here to stay.

Meanwhile, we’ll have to wait a few more weeks to see how Nvidia’s Ada Lovelace-powered cards stack up against AMD’s RDNA 3 architecture. While little is known about AMD’s upcoming consumer cards beyond a few seconds of gameplay demoed during the Zen titan's Ryzen 7000 desktop launch last month, we do know it’ll be based on a 5nm TSMC manufacturing process.

And if AMD follows a similar partner as its RDNA 2 launch in 2020, we’ll won’t be getting any mid-tier cards from AMD out of the gate either. ®

PS: Nvidia also announced the Lovelace-based RTX 6000 GPU, which is aimed at workstations doing heavy graphics work and similar computation.