Nvidia introduces $399 RTX 4060 Ti and $299 4060 without introducing a price hike.

201

Nvidia is beginning to roll out its new Ada Lovelace GPU architecture to sub-$500 graphics cards, the models that the vast majority of PC gamers buy and use. There's good news and bad news.

The good news is that the GeForce RTX 4060 Ti will launch on May 24 for $399, the same price that the RTX 3060 Ti launched for in December 2020—it's the first GPU launch in ages not to come with a price hike built in. Like other RTX 4000-series GPUs, it supports DLSS 3's AI-assisted frame rate-boosting technologies and AV1 video encoding support, and it's much more power-efficient than its predecessors. The bad news is that the 4060 Ti is an unusually mild upgrade over the 3060 Ti, with the same 8GB bank of RAM and (according to Nvidia) just 15 percent faster performance than the 3060 Ti in games without the DLSS Frame Generation feature enabled.

Nvidia will sell a Founders Edition version of the 4060 Ti that looks about the same size as its Founders Edition RTX 4070; it also uses the new 12VHPWR connector, though cards from Nvidia's partners will often opt to use a traditional 8-pin power connector instead.

Nvidia will fill out the 4060 lineup later in July; a $499 4060 Ti with 16GB of memory will sit in between the 8GB model and the $599 RTX 4070. A lower-end non-Ti RTX 4060 will also launch for $299, a small drop from the 3060's $329 introductory price for what Nvidia says should be 20 percent better performance; not a bad deal for the first generation-over-generation price drop we've seen from Nvidia in years. The 8GB 4060 Ti and 4060 are both pitched as 1080p cards, partly because 8GB of graphics memory is becoming a bottleneck in newer games at higher resolutions.

Meet AD106

	RTX 4090	RTX 4080	RTX 4070 Ti	RTX 4070	RTX 4060 Ti	RTX 4060	RTX 3060 Ti	RTX 3060
CUDA Cores	16,384	9,728	7,680	5,888	4,352	3,840?	4,864	3,584
Boost Clock	2,520 MHz	2,505 MHz	2,610 MHz	2,475 MHz	2,535 MHz	?	1,665 MHz	1,777 MHz
Memory Bus Width	384-bit	256-bit	192-bit	192-bit	128-bit	128-bit	256-bit	192-bit
Memory Clock	1,313 MHz	1,400 MHz	1,313 MHz	1,313 MHz	2,250 MHz	?	1,750 MHz	1,875 MHz
Memory size	24GB GDDR6X	16GB GDDR6X	12GB GDDR6X	12GB GDDR6X	8GB or 16GB GDDR6	8GB GDDR6	8GB GDDR6	12GB GDDR6
TGP	450 W	320 W	285 W	200 W	160 W	115 W	200 W	170 W

The 4060 Ti series is based on the AD106 GPU die, the fourth and smallest (so far) of the chips Nvidia has shipped in the RTX 4000 series after the AD102 (4090), AD103 (4080), and AD104 (4070 series). The AD106 die is also used in the laptop versions of the RTX 4070. The 4060 will use the AD107, an even smaller chip.

A fully enabled AD106 die includes 4,608 of Nvidia's CUDA cores on a 128-bit memory interface, but the 4060 Ti GPUs use a partially disabled version (see the table above for details). As rumored, the 4060 Ti's specs are a bit underwhelming when compared directly to the 3060 Ti, which had more CUDA cores and a 256-bit bus, which contributes to its smaller-than-typical improvement over the last-gen card. The other RTX 4000-series GPUs have typically included as many or more CUDA cores than their RTX 3000-series counterparts, though narrower memory bus widths have been the norm for every GPU other than the 4090.

To close that gap, the 4060 Ti benefits from Ada's architectural improvements, plus significantly higher clock speeds. Nvidia has also added extra L2 cache to save on the number of times the GPU needs to go out to main memory, reducing the impact of the narrower memory bus (the 4060 Ti has 32MB of L2 cache, and the 4060 has 24MB, compared to 4MB and 3MB for the 3060 Ti and 3060).

Nvidia's press briefing and slide deck avoided comparing most of these specs directly, opting instead to do generation-over-generation spec comparisons using TFLOP numbers. Clearly, the intent is to drive home that the 4060 Ti is a faster card than the 3060 Ti despite these specs, because people who follow GPU news online are incapable of being chill when they perceive that a product has been named wrong or positioned wrong or priced wrong (expect a fair amount of outrage about this to be expressed via YouTube thumbnails).

Nvidia's press briefing used TFLOP numbers instead of the number of cores or memory bus width here, since the numbers don't look very flattering out of context. Credit: Nvidia

One of the things that impressed us most about the RTX 4070 was its power efficiency, and both RTX 4060 GPUs look pretty good on that front, too. The 4060 Ti comes with a TGP of 160 W, compared to 200 W for the 3060 Ti, while the 4060 only needs 115 W instead of 170 W. That's awfully close to the 75 W of power that a PCI Express slot can provide without an external power connector—a hypothetical RTX 4050 could hit that number, which could bode well for people trying to put a GPU into an underpowered budget desktop tower from Dell or HP with no 8-pin power connectors to spare.

A note on DLSS 3

Nvidia leans heavily on DLSS Frame Generation throughout its announcement to make the 4060 Ti look even faster than the 3060 Ti, which is all well and good. For games that support it, DLSS FG can take frames rendered by your GPU and generate an interpolated frame to show in between them, hypothetically boosting frame rates without asking the GPU to render more frames.

But there are three problems with DLSS FG, especially on mid-range GPUs. First, not all games support it, though many newer and more-demanding ones do (and pending Unreal Engine 5 support should make it even easier for developers to add support in new games). Second, it adds latency, which can make controls feel laggier in fast-paced shooters or other games where precision is important.

And third, it works best in situations where your GPU is already pushing a decent frame rate. More rendered frames means that Nvidia's AI algorithm needs to "guess" less when generating its interpolated frames, and imperfections in those AI-generated frames are harder to spot at 120 fps than they are at 40 fps. We encountered some of this in our RTX 4070 review, where more visual artifacts were more noticeable in games like Cyberpunk 2077 where the GPU was rendering at closer to 30 fps than 60.

Obviously, buying a midrange-to-low-end graphics card has always been about making compromises and balancing smooth frame rates with resolution and graphical fidelity. Despite its tradeoffs, DLSS FG is a powerful lever that people can pull to try to find a balance that works for them. It just isn't a substitute for better rasterized game performance, which benefits all games in all scenarios rather than just a subset.

What about the other 4060 GPUs?

We don't know a ton about the 16GB version of the 4060 Ti, aside from its $499 price tag. The $100 price increase seems a little steep if all you're getting is more memory, and Nvidia could always decide to sell it with a fully enabled AD106 die or boosted clock speeds to help close the gap between it and the $599 RTX 4070. But as far as we can tell, it's the exact same card with 16GB of RAM instead of 8GB, which might help in some games but won't suddenly make this a 4K-capable card.

As for the 4060, Nvidia's numbers say that it is around 20 percent faster than the RTX 3060, making it look a bit more appealing compared to its last-gen counterpart. And the $299 price tag will probably make it the most popular 4000-series card all by itself, like loads of xx60 cards before it. But stepping down from the RTX 3060's 12GB of RAM to 8GB is a bummer, as newer games come with ever-increasing video RAM requirements if you're trying to turn the resolution and all the settings up.

Listing image: Nvidia

Andrew Cunningham Senior Technology Reporter

Andrew is a Senior Technology Reporter at Ars Technica, with a focus on consumer tech including computer hardware and in-depth reviews of operating systems like Windows and macOS. Andrew lives in Philadelphia and co-hosts a weekly book podcast called Overdue.

201

View Comments