Enlarge / It's not actual GeForce RTX 4080 cards slotted into GeForce Now's "Superpods," but Nvidia says the hardware is pretty close. (credit: Nvidia)

Cloud-based gaming service GeForce Now's new Ultimate tier is rolling out today, promising a series of adjectives about game streaming that might have seemed impossible just a few years ago: high-resolution, ray-traced, AI-upscaled, low-latency, high-refresh-rate, and even competition-ready.

I tested out the Ultimate tier, powered by Nvidia's RTX 4080 "SuperPODs " on a server set up for reviewer early access, for a week. If I hadn't been hyper-conscious of frame numbers and hiccups, I could have been tricked into thinking the remote 4080 rig was local. Except when I was playing ray-traced games AAA games on systems that had no business playing them, like laptops with discrete GPUs, iPads, or my TV with no gaming console attached—that always felt weird, in a fun way.

Ars had previously described our GeForce Now 3080 experience as "dreamy" and called the performance "a white-hot stunner that rivals the computing power you can muster" with the same RTX 3080 card in your PC. It's easy to lay at least the same kind of praise on the new Ultimate tier. It replaces the previous RTX 3080 option with the next generation's chipset for the same price ($20 per month, $99 for six months). That might be a steep price tag for a service that mostly makes you buy your games, but given the 4080's $1,200 price tag, the rent-versus-buy question is worth considering at this level.

Enlarge / Nvidia's GeForce RTX 4080. (credit: Andrew Cunningham)

Very little about Nvidia's GeForce RTX 4080 is surprising—especially now that the confusing, scrapped 12GB version is being renamed . In pretty much all of our performance tests, it slots in right where you'd expect it to, comfortably ahead of the RTX 3080 Ti but trailing the $1,500 RTX 4090 by enough to justify the $300 price gap. It's usually capable of hitting or exceeding 60 fps at 4K, and games with DLSS support (or some other kind of upscaling tech) can buy you a solid frame rate increase. And its power requirements aren't as stratospheric as the 4090's, either, so most people with an existing xx70 or xx80-class gaming PC shouldn't need to switch out their power supply.

The major downsides, as of this writing? As a group, the cards are often just as huge and cumbersome as the RTX 4090 (the Founders Edition is identical, and partner cards largely follow suit). The $1,200 starting price is historically high—the 3080 Ti launched with a GPU-shortage-inflated MSRP of $1,119 GPU, but the 2080 and 3080 were both a mere $699 at launch. And even if you are willing to pay that price—surprise, surprise—it's basically impossible to find in any form anywhere close to MSRP.

Which means, hooray: another GPU review that exists mostly as a theoretical exercise! If you could buy this GPU for the amount of money it's supposed to cost, and if its competitors were also available for what they were supposed to cost, then here's how it would stack up. That world doesn't exist right now, but if you're reading this in a few months, circumstances may have changed. In the meantime, imagine with us, won't you?

Enlarge (credit: Nvidia)

Last month, Nvidia took the unusual step of "unlaunching" a previously announced product . The 12GB version of the GeForce RTX 4080 graphics card was, by the company's admission, "not named right" and was delayed and rebranded to avoid confusion with the 16GB version of the RTX 4080 that launched. Besides having less RAM, the 12GB version of the RTX 4080 also offered less memory bandwidth and fewer GPU cores than the 16GB version.

Nvidia didn't announce exactly what branding it would use for the revived RTX 4080, but regulatory filings submitted by Gigabyte (as reported by VideoCardz ) suggest that the company has settled on calling it the "4070 Ti."

This isn't guaranteed to be the actual name—regulatory filings like this can be placeholders rather than actual products—but this branding would be more consistent with how Nvidia has named past GPU generations. The xx80 cards usually use the same physical GPU die as the flagship xx90 cards but run at lower clock speeds and with parts of the die switched off; this allows Nvidia to use GPU dies with defects rather than tossing them out. The xx70 cards generally use a smaller, less-performant GPU die based on the same architecture.

Enlarge (credit: Getty Images)

Late last Friday, Nvidia decided that it was " unlaunching " the lower-end 12GB version of its upcoming GeForce RTX 4080 graphics card so that it could be renamed and released at a later date. This was good news, for the people who care about this kind of thing—the $899 12GB RTX 4080 and $1,199 16GB RTX 4080 were substantially different cards with much different performance levels. Giving them both the same name could have created unnecessary disappointment and confusion for buyers of the cheaper card.

The problem for GPU makers is that Nvidia planned to launch those cards in mid-November, and partners had already started manufacturing and packaging them so they could be shipped out to retailers. Gamers Nexus has spoken with sources at two of Nvidia's board partners about some of these logistical hurdles, reporting that existing boxes for 12GB RTX 4080 cards were being "collected and destroyed" and that Nvidia "is at least subsidizing the boxes, or part of them, to be replaced." The relabeled GPUs will supposedly be reintroduced or relaunched (or un-unlaunched?) around CES in January 2023.

There will be costs for other board partners, too, both for GPUs that have already been produced and those that will be manufactured after Nvidia has settled on a name (Gamers Nexus says this hasn't happened, but that "4070" or "4070 Ti" seems most likely). GPU coolers usually have the card's name and model number printed on it somewhere, occasionally in a prominent place with programmable LEDs underneath. These coolers will either need to be rebadged, reprinted, or replaced to switch out the old RTX 4080 branding with the new branding.

Enlarge

Nvidia announced today that it would be " unlaunching " the 12GB version of its upcoming GeForce RTX 4080-series GPU on the basis that "having two GPUs with the 4080 designation is confusing." The card will not be launching in the originally planned November window, but it will return at some point following a rebranding.

"The RTX 4080 12GB is a fantastic graphics card, but it's not named right," reads a brief, anonymous Nvidia blog post . The 16GB version of the 4080 will still launch on November 16.

The 16GB version of the 4080 was in keeping with past xx80-series Nvidia cards—it used the same GPU die as the high-end RTX 4090 , but with some of the CUDA cores switched off and the clock speeds reduced. But the difference between the 12GB and 16GB versions of the 4080 was much larger than their names would suggest; the 12GB version came with many fewer CUDA cores and a narrower 192-bit memory bus, and rumors indicated it used a totally different GPU die. Benchmarks from Nvidia suggested the 12GB version could be as much as 30 percent slower than the 16GB version of the 4080.