The GeForce RTX 4070 Ti that we review today started life as the GeForce RTX 4080 12 GB, and was announced at the same time as the RTX 4080 16 GB. Immediately, tech enthusiasts all over the world cried out because these two RTX "4080" cards were based on completely different hardware configurations. The RTX 4080 16 GB was built around the AD103 GPU with a 256-bit wide memory interface, and the RTX 4080 12 GB used AD104 with 192-bit memory instead. After weeks of public drama, NVIDIA finally gave in and "unlaunched" the RTX 4080 12 GB and is releasing it now as "RTX 4070 Ti." Besides the name there is no difference, you're still getting AD103, with 7680 GPU cores, 12 GB GDDR6X memory on a 192-bit bus and 60 RT cores, just with a name that's not as confusing to less informed buyers.
NVIDIA isn't producing a Founders Edition of the RTX 4070 Ti, so we have only factory-overclocked custom designs in our launch day reviews (
ASUS TUF OC,
Gigabyte Gaming OC,
MSI Gaming X,
MSI Suprim X,
Palit Gaming Pro OC,
PNY OC). While I usually start here by reporting the performance differences vs other cards, I'd like to begin with performance scaling at different resolutions. You've probably noticed it while looking at the charts: RTX 4070 Ti loses quite some performance as the resolution is increased—more so than other cards in the same performance tier. For example, at 1080p, the 4070 Ti is 3% faster than the RTX 3090 Ti, at 1440p both cards are evenly matched, and at 4K the 3090 Ti is 10% faster. That's a surprisingly big range; things are no different when compared to the RTX 7900 XT: -4% at 1080p, -9% at 1440p, and -10% at 4K—same 13-14% delta. It's definitely not an architectural problem, because we're seeing the same trend against the RTX 4080, which is based on Ada, too: -13% at 1080p, -18% at 1440p and -25% at 4K. To me it seems that the underlying reason for this behavior is that the AD104 GPU has a ton of shading power, but becomes limited by cache size and memory interface at higher, more memory-intensive resolutions.
This puts the RTX 4070 Ti in an interesting position. RTX 4070 Ti is an amazing choice for gaming at 1440p, at maxed out details. With RT disabled you'll be able to drive high-refresh-rate monitors easily with 120 FPS and more. If you turn on ray tracing you're still getting 60+ frames per second, even without DLSS. Turn on DLSS, and optionally DLSS 3 frame generation, and you'll hit your 120+ FPS targets with ease. At 4K the card works well enough to give you a solid gaming experience, but you'll have to dial down the details in some of the most demanding games to hit 60 FPS. While Radeon RX 7900 XT and RTX 3090 Ti are roughly the same perf as 4070 Ti at 1440p, they do scale considerably better at 4K, so they will get you those extra 10 FPS that make it easier to hit 4K60 everywhere.
When comparing the RTX 4070 Ti against the RX 7900 XT, at 1440p, with ray tracing turned off, the AMD card offers a little bit higher performance, with a 7% advantage. The RTX 4070 Ti roughly matches the RTX 3090 Ti's performance—an important achievement, and the gen-over-gen uplift (vs RTX 3070 Ti) is an impressive 45%. The RTX 4080 is 18% faster, and the Radeon RX 7900 XTX is 20% ahead. NVIDIA's newest card is also considerably faster than RTX 3080 and RX 6800 XT, which are among the most popular cards in recent years.
Ray tracing promises to enhance your games with additional fidelity by using physically accurate rendering of light, shadows and reflections. RT is executed as an additional pass besides classic rasterization and comes with a serious performance hit. NVIDIA cushions that by including their third generation hardware ray tracing units on the RTX 4070 Ti, whereas AMD offloads most of that work to their GPU's shading units, even on the latest RDNA 3 architecture. The end result is that NVIDIA is clearly the better choice if you're betting on ray tracing, but the differences aren't exactly huge for the RTX 4070 Ti. It seems that due to the cache/memory configuration, the card sees a bigger performance hit from enabling ray tracing than other GeForce 40 cards, especially at 4K. That's not to say that RTX 4070 Ti is bad at ray tracing—it still is one of the fastest cards in that scenario, but the performance scaling is something you should be aware of.
MSI's graphics card have offered excellent fan settings for years, and the Gaming X is no exception. With only 27.9 dBA under full load, the card is whisper-quiet. Temperatures are good, too, reaching 68°C. While other cards might achieve even lower temperatures, they usually are louder at the same time. If you prefer lower temperatures, then you can activate the "Gaming" BIOS, which increases the fan speeds just enough to get you a few degrees better temperatures, without ending up being too loud and suffering from diminishing returns. Our apples-to-apples heatsink comparison test reveals that the Gaming X cooler is one of the weaker ones tested today, but the differences are small. What's very important, too, is how the fan profile is configured, and here MSI did a great job, making best use of the available cooling capacity. As expected from all modern graphics cards, all GeForce RTX 4070 Ti models come with the idle-fan-stop capability that shuts off the fans when not gaming.
The highest priority for GPU engineers today is improving energy efficiency, as that's the limiting factor for performance improvements. On the other GeForce 40 Series cards we saw impressive advancements here, the RTX 4070 Ti is no different. It is considerably more efficient than all previous GPUs from both AMD and NVIDIA. It's even slightly more energy-efficient than AMD's new Radeon RX 7900 Series cards, only the GeForce RTX 4080 and RTX 4090 operate with higher efficiency. The result of these improvements is not only lower power draw, i.e. lower power bill and PSU requirements, the lower heat output also makes it easier for the thermal solution to keep your card cool and the fans can run at lower speed, emitting less noise. With slightly less than 300 W, the RTX 4070 Ti consumes roughly the same amount of power as the RTX 3070 Ti, while providing considerably better performance at the same time—300 W isn't a lot by today's PSU standards—a 650 W PSU should be able to power the system easily, maybe 750 W if you use a power-hungry CPU. While we've seen shockingly high multi-monitor and video playback power consumption numbers on AMD's new cards, this works perfectly fine with the GeForce RX 4070 Ti—just 22 W.
There is no change in how overclocking works, just push the GPU and memory clocks up until the card is unstable, and you're getting good gains. On the new AMD cards, things are more complicated, you must use undervolting to achieve meaningful OCs, and bumping the power limit is almost a requirement, too. What I find surprising is that none of the cards tested today go above the NVIDIA default power limit of 285 W, probably to not lose energy efficiency. Overclockers will always want to push their cards higher and are willing to increase the power limits manually. Despite the solid VRM design, MSI chose to limit the manual adjustment power limit to only 305 W, or +20 W. Maybe this was done to segment the card from the Suprim X, which goes up to 365 W.
The GeForce RTX 4070 Ti will be available in stores tomorrow starting at $800, which is a pretty competitive price point when considering the current landscape of the market. At this price, the card is considerably more affordable than the AMD RX 7900 XT ($900), offering slightly better Performance per Dollar, even with RT disabled, and much better with RT enabled. This price is also "shots fired" at heavily discounted last-generation cards like the Radeon RX 6950 XT ($830), RX 6900 XT ($700), and even RX 6800 XT ($650). All those cards are pretty much obsolete now, because the 4070 Ti offers better performance at a price point that's close enough to tempt potential buyers. Of course, the MSRP of 3070 Ti was $600, now 4070 Ti costs $800, so $200 higher, or 33%, but you're getting almost 50% in extra performance. It's still a lot of money, especially for a "x70" card, which psychologically people expect to sit much closer to $500. But right now, with RX 7900 XT not taking off as much as expected, there's no reason why NVIDIA, or any other corporation, would want to hand out freebies. For this round it doesn't look like AMD will come to our rescue, they seem happy setting their price to slightly below what NVIDIA decides. It also looks like NVIDIA prepared their board partners for a price war much better than AMD. Check out the RTX 4070 Ti board designs and then compare them to Radeon RX 7900 XT. The latter are much more complex with more components that overall increase production cost substantially.
MSI's Gaming X is expected to sell for $840, which is a $40 increase over the NVIDIA MSRP, or 5%. While there's no way that the factory overclock will get you that much performance, I still feel like the improved cooler, with its low noise profile is worth it, especially if low noise matters to you. Strong competition comes from the PNY RTX 4070 Ti OC, which sells for only $830, yet offers virtually the same noise levels with slightly better temperatures. After seeing the 4K results for the 4070 Ti, previous generation flagships such as the 3090 / 3090 Ti / RX 6950 XT can still be compelling at that resolution, assuming the price is right. What could change things though is DLSS, which renders the game at lower resolution, meaning cache and memory won't be hit as hard, and that will lead to better scaling for RTX 4070 Ti, especially at 4K. DLSS 3 could be a game changer here, too, because GeForce 30 and Radeon 7000 Series don't support that technology, but not all games support DLSS 2 or DLSS 3. For 1440p I see no convincing alternative to RTX 4070 Ti, unless you're willing to accept lower performance (RX 6800 non-XT, used GeForce 20). AMD's Radeon RX 7900 XT suddenly feels overpriced at $900, and I doubt that RTX 4080 can stay at $1200 for long—there's now a huge $400 gap between two NVIDIA offerings with no space left in-between for a new model, so the coming months will be interesting to see how GPU prices continue to develop.
![Editor's Choice](/images/editorschoice.gif)