The RTX 5090 is the most powerful gaming GPU ever released, and comes with 21,760 CUDA cores on a GB202 Blackwell chip, which is just over double the amount that comes in the RTX 5080. October leaks are correct The 24GB RTX 5080 becomes the best GeForce card gamers can buy, and Nvidia’s flagship gaming offering essentially loses half its computing power overnight. Jeff Fisher, senior vice president of all things GeForce at Nvidia, He stated this to Taiwanese media The 5090 isn’t going anywhere, but that doesn’t mean much when no one can actually afford one.
The 24GB RTX 5080 is still half the RTX 5090
The rumored launch of the 24GB RTX 5080 variant is expected to carry the ‘Super’ branding that has been dormant since the 40-series GPUs, but this new variant doesn’t change anything in terms of the flagship silicon. The RTX 5080 already uses the full GB203 die, so Nvidia has nothing left to open without moving to a truncated GB202 die, which is already in short supply. The leaks also describe a power bump of up to 415W TDP, an increase of 55W over the 360W on the standard card.
For those who game at 4K with graphical fidelity, the memory and power upgrades make sense. 16GB already causes problems in the most demanding games. Indiana Jones and the Grand Circlefor example, collapses at 4K with track tracking on a 16GB card. However, the on-card memory upgrade launched in January 2025 isn’t groundbreaking, and it just means that Nvidia will be topping its GeForce lineup with silicon it’s been selling for nearly two years already.
Realistically, the market had already made that choice, because the 5090 was out of reach for the vast majority of gamers, even those who could afford the $1,999 MSRP. The cheapest RTX 5090 is currently priced around $8,000 at the time of writing, while the cheapest RTX 5080 is around $1,700. The 5080 has been Nvidia’s best true gaming card in a while.
The Class of 80 has drifted off the top for three generations
The gap has become larger
Two generations ago, the gap looked very different: the RTX 3080 had 8,704 CUDA cores versus the RTX 3090’s 10,496, or about 83%. Both cards came from the same GA102 die, so the 3090 was essentially a “fuller” version of the 3080, although they were much closer than the cards that came after them.
The RTX 4080 moved to the smaller AD103 die while the 4090 stayed on the AD102, and the 4080’s 9,728 cores came out to about 59% of the 4090’s 16,384 cores, and that continues with the Blackwell. Over three generations, the 80-class cards went from being major cards to being about half of one.
As someone who owns both an RTX 3080 and an RTX 5080, I’ve never felt like either was inadequate for high-resolution gaming for titles released around the time of the card’s release. The 3080 was particularly excellent value in retrospect, considering that it can still handle many of the newer games at 4K while upping the quality. The 5080 is certainly a meaningful upgrade, offering a much better 1% improvement over the 3080 when the resolution isn’t high, but with the larger VRAM buffer and bump in compute helps it be an everyday 4K gaming card without an upgrade.
Compared to what the claimed RTX 5080 Super will be like, it’s hard to tell without having one on hand, but I’ve never personally encountered a game where 16GB was a limiting factor in performance, even with texture quality and filtering maxed out.
GB202 has more profitable places
Turns out the 5090 is a much better computing card than a gaming card
The GB202 powers the RTX 5090 and RTX Pro 6000 Blackwell workstation cards. Every GB202 that becomes a GeForce card is a card that a Pro card can’t get much more expensive, and it’s no secret that even the RTX 5090 is in high demand for its prowess as a workstation card over a gaming card.
Fisher’s denial addressed end-of-life, not how to customize the GB202 chips, and the popularity of silicon for workstation purpose is much more profitable meaning the price of the 5090 won’t drop anytime soon. That doesn’t mean the leak is correct, and Nvidia may simply be struggling to get enough GDDR7 for a 32GB consumer card. But buyers don’t care; If there is no stock they can buy, the result is the same either way.
We will find out more about the alleged super update of the 50th Series soon
CES 2027 should be the next checkpoint for the GeForce brand as a whole. If Nvidia announces a 24GB RTX 5080 there, how you position it matters a lot: if it’s the same GB203 with more memory, and presented as the best GeForce card you can actually buy, then the GB202 is basically dead for gamers. there Rumors of an “SE” version of the 5090which includes a cut GB202, but these rumors are unfounded at the time of writing.
