GeForce RTX 40 Series Guide: Ada Lovelace Desktop and Laptop GPUs

The RTX 40 series is harder to navigate than a single generation should be. NVIDIA launched it, renamed one card after public backlash, refreshed the middle of the stack a year later with SUPER variants, and shipped a laptop lineup whose names map to completely different silicon than the desktop cards they echo. If you are buying into Ada Lovelace now, on the used market or in a discounted new laptop, the hard part is not performance, it is working out what each name actually refers to. This guide lays out the whole stack with verified specs so the names stop being a guessing game.

The desktop lineup

Every desktop RTX 40 SKU, with the silicon behind it:

The base RTX 4070 launched with GDDR6X; later board revisions shipped plain GDDR6 at the same effective bandwidth, and NVIDIA's own page now lists it as "GDDR6/GDDR6X." Treat that one as a production variance rather than two different products.

The 4080 12GB that never shipped

The clearest lesson in the whole lineup came before most of it launched. In September 2022 NVIDIA announced two RTX 4080s: a 16 GB model on AD103 and a 12 GB model at a planned $899. The catch was that the "4080 12GB" was an AD104 part with 7,680 cores and a 192-bit bus, sharing only a name with the real AD103 4080. They were different GPUs wearing the same badge. After public backlash NVIDIA "unlaunched" the 12 GB version in October 2022 and relaunched the identical hardware as the RTX 4070 Ti in January 2023 at $799.

It is worth remembering because it is the whole reason to read this generation by die, not by name. NVIDIA very nearly sold two different chips as one product, and the only thing that stopped it was that the split was too obvious to miss. The laptop lineup, as we will see, does something similar and got away with it.

The SUPER refresh

In January 2024 NVIDIA refreshed the middle of the stack with three SUPER cards. These were repositioning moves, mostly better value per core rather than new silicon:

• RTX 4080 SUPER at $999: a small core bump over the 4080 (9,728 to 10,240) on the same AD103, same 16 GB and 320 W, but $200 below the original 4080's launch price. The real change was the price.
• RTX 4070 Ti SUPER at $799: the most substantial of the three. It moved up a die, from AD104 to AD103, and gained a wider 256-bit bus and 16 GB of VRAM against the 4070 Ti's 192-bit and 12 GB. This is the card that fixed the 4070 Ti's biggest weakness.
• RTX 4070 SUPER at $599: a meaningful core bump (5,888 to 7,168) on the same AD104, same 12 GB, at the same price as the base 4070.

There were no SUPER variants at the 4090, 4060 Ti, or 4060 tiers. If you are shopping used, the SUPER cards are generally the stronger buys in their price bands, and the 4070 Ti SUPER specifically is the one that resolved a real VRAM and bandwidth complaint.

Reading the stack: VRAM, bus, and the L2 change

Two numbers do most of the work when comparing these cards: VRAM capacity and memory bus width. The capacities split cleanly, 24 GB on the 4090, 16 GB on the 4080-class and the 4070 Ti SUPER and 4060 Ti 16GB, 12 GB on the 4070 family, and 8 GB on the 4060 and 4060 Ti 8GB. For 1440p and 4K longevity, the 12 GB and 16 GB tiers hold up far better than 8 GB, which is the single most common regret on the lower cards as texture budgets grow.

The narrow buses lower down the stack look alarming and mostly are not. Ada Lovelace enlarged the on-die L2 cache dramatically, up to 96 MB physical on AD102 against roughly 6 MB on the previous generation's GA102. A large L2 services more memory requests on-chip and cuts traffic to DRAM, which let NVIDIA pair a 128-bit bus with the 4060-class cards without a catastrophic bandwidth penalty. The tradeoff shows up at 4K, where the cache hit rate falls and the narrow bus starts to bite, so the 128-bit cards are best understood as 1080p-to-1440p parts. If the mechanism behind that is unfamiliar, our explainer on how to read NVIDIA GPU specs works through the bus, bandwidth, and cache relationship in detail.

The laptop trap

This is the part of the guide most likely to save someone from a bad purchase. A mobile RTX 40 GPU is not the desktop card of the same name at lower clocks; it is a different, smaller die at a fraction of the power.

Two things follow. First, the laptop "RTX 4090" is built on AD103, the same die family as the desktop 4080, and desktop-4090 AD102 silicon never shipped in a laptop at all. Read the mobile name as a performance-tier label, not as the desktop card miniaturized. Second, and just as important, laptop performance is set by the configurable TGP the manufacturer chooses within that range. Two laptops with an identically named GPU can differ by tens of watts and perform very differently; a 4070 Laptop at 35 W is a different machine than one at 115 W. Always check the TGP a specific laptop ships at, because the name does not tell you.

The generational features

Three Ada features define the series. DLSS 3 Frame Generation is the headline, an Ada-exclusive that generates entirely new interpolated frames to raise frame rate beyond what the shaders render; it stays gated to the RTX 40 line even though the later DLSS 3.5 Ray Reconstruction was extended to older cards. Dual AV1 NVENC encoders arrived with Ada, a real gain for streaming and video export over the previous generation's single encoder. And the move to a custom TSMC 4N process delivered a large perf-per-watt improvement over Ampere's Samsung 8nm, which is why even the smaller Ada cards run cool and efficient.

Which one, for whom

• RTX 4090: the halo card and the pragmatic choice for creators and local-AI users who need 24 GB. Covered in depth in its own guide.
• RTX 4080 SUPER: high-end 4K gaming with 16 GB, at a saner price than the original 4080. Check its Market Watch page for current cost.
• RTX 4070 Ti SUPER and 4070 SUPER: the value sweet spots for 1440p and entry 4K, with the 4070 Ti SUPER preferred where the extra VRAM and bus matter and the 4070 SUPER where budget leads.
• RTX 4060 and 4060 Ti: capable 1080p and light 1440p cards, but weigh the 4060 and the 8 GB versus 16 GB 4060 Ti carefully, because 8 GB is the tier most likely to feel tight within the card's service life.

Bottom line

The RTX 40 series rewards buyers who read it by die and by VRAM rather than by name. The SUPER refresh improved the value cards, the unlaunched 4080 12GB is a permanent reminder to check the silicon, and the laptop lineup demands that you verify both the die and the shipped TGP before trusting a mobile GPU's badge. Prices across the stack are still distorted by the current memory shortage, so anchor any purchase to what these cards actually cost today on Market Watch rather than to their launch MSRPs.

Sources:

• NVIDIA: GeForce RTX 40 series
• NVIDIA: RTX 4080 SUPER, 4070 Ti SUPER, and 4070 SUPER announcement
• The Register: NVIDIA unlaunches the 12GB RTX 4080
• NVIDIA: GeForce RTX 40 series laptops