ASUS Prime RTX 5070 Ti review — price, power, and performance, all perfectly balanced

You probably saw our analysis and review of the new NVIDIA flagship, the GeForce RTX 5090, and the equally as impressive GeForce RTX 5080. In both of those reviews, we pointed out that the device delivers a whopping 30% increase in raw power for the 5090, and roughly 15% for the 5080 over its equivalent card. When combined with DLSS 4, those numbers can shoot to the stratosphere, offering upwards of 300 to 400% improvement over raw capture. Both of those cards have extraordinary power, but they also carry a commensurate price. If you’ve got the funds, and you demand the very best, both are easy choices. But what if you are more budget conscious and looking for that mid-tier card that hits the center of the price, power, and performance triangle? NVIDIA is purporting that their RTX 5070 Ti does precisely that. I threw it into my machine and ran it through the same tests I used for the 5080 and 5090 (and a few new ones you might like) – let’s see if the RTX 5070 Ti lives up to NVIDIA’s aspirations.

As always, our first stop on the tour is to head under the hood to see where the 5070 Ti lands in terms of cores, speeds, and feeds. Before we do that, let’s just preface this very, very clearly – you cannot directly compare across generations with just the amount of cores or even speeds. The generational differences between cards (e.g. third-gen RT Cores and fourth-gen Tensor Cores in the 4000 series, vs. fourth-gen RT cores and fifth-gen Tensor Cores in the 5000 series) is always less about the amount of cores, but more about what’s being done with them. Let’s take a look:

The RTX 5070 Ti’s closest numerical equivalent is the RTX 4070 Ti Super, though it has a 3-5% lead over even that card. Utilizing the same GB203 processor as the RTX 5080, the RTX 5070 Ti carries the same technologies as its bigger brothers (hardware raytracing, a wide array of upscaling options, etc.) and does so with 8,960 shading unit cores, 280 texture mapping units, and 128 render output units. That’s joined by 280 tensor cores which are directly responsible for the AI upsampling that DLSS utilizes to ramp framerates into the stratosphere (more on that later, of course). The card has 70 raytracing cores that are responsible for the realtime lighting and shadow technologies. Just like the RTX 5080, the RTX 5070 Ti is equipped with 16 GB of GDDR7 memory which flows through a 256-bit memory interface. The GPU runs at an operating frequency of 2300 MHz with a boost clock of up to 2452, with memory clocking in at 1750 MHz for 28 Gbps effective throughput. Just for basis of comparison, the RTX 4090 could only muster 21 Gbps throughput.

An LED on the card notifies you when your connection is secure

From a power perspective, the card utilizes the same 1x16-pin power connector for a single cable solution, or you can use the included three-cable splitter for power. The max power draw is rated to 300 W maximum, which is more than sufficient for an enthusiast-level card. (The models from Asus have a set of LEDs next to the connector to indicate that the plug is indeed fully seated. Something I hope to see on all cards in the next card refresh.) It has four outputs to choose from – 1x HDMI 2.1b, and 3x DisplayPort 2.1b. The RTX 5090 and 5080 were both two slot cards, but Asus and several other manufacturers are happy enough to take up three – this Asus Prime RTX 5070 Ti fattens it out a bit. There are no Founder’s Edition cards for this unit, so it’ll vary based on whatever editions manufacturers like Asus, Gigabyte, MSI, or PNY make. In this case, there is a slight difference between the cooling on the Founder’s Edition cards and the Asus variant. Instead of the fans being located on opposite corners of the card, top and bottom, Asus goes with a more traditional two fan solution on the bottom, directing air from the bottom of the case, through the fin block, and upwards to be carried away from the card.

Unboxing the ASUS Prime GeForce RTX 5070 Ti GPU

The RTX 5090 came in at $1999, with the RTX 5080 hitting for $1000. The RTX 5070 Ti will run you $749, with some manufacturers adjusting upwards to justify whatever improvements or overclocking they might have added to the mix. That could be a more palatable cost for some, depending on how our benchmarking turns out.

Outside of gaming, one of the nice surprises for the GeForce RTX 5000 series is a new rendering format. AV1 has enabled blisteringly fast render speeds in DaVinci Resolve, and now we have AV1 Ultra High Quality. This new format cleans up visual noise on large open spaces like skies and walls. It also adds another notch on the speed belt – something I’ve seen in action as I render the videos for this review! I cued them up on my 4090 and saw videos that said they’d render in roughly 90 seconds suddenly render in less than 30 on the 5070 Ti. Given that YouTube didn’t choke on them at all, it’s a win across the board.

The ASUS PRIME 5070 Ti has a dual BIOS - performance and a quiet

Before we get into the numbers, I want to talk about DLSS 4 and “Fake frames”. If you are unfamiliar, GeForce cards since the 2000 series have a technology called DLSS – deep learning super sampling. Newer cards also picked up additional technologies like Deep Learning Anti-Aliasing, Ray Reconstruction, and Super Resolution. We’ve deep dived all of them in the past, but let’s focus on DLSS 4.

DLSS 4 utilizes the advanced power of the combined Blackwell architecture to interpolate up to three additional AI-generated (no, not the thieving kind of AI-generated) frames into the stream, driving framerates into the stratosphere. Games like Cyberpunk 2077 are notoriously brutal on videocards, with the flagship of the 4000 or 5000 series barely able to pop above 30 fps when running without synthetic frames. As you’ll see when you read our benchmarks, DLSS 4 makes a world of difference. You’re welcome to complain about “fake frames” but I’ll be over here playing Cyberpunk 2077 on Psycho settings in 4K at nearly 400 fps (at least on Ultra Performance anyway).

For those of you concerned about the visual impacts of DLSS 4, I’m certain people will spend exorbitant amount of time and energy looking at still images zoomed in 4000%, but here’s Cyberpunk 2077’s benchmark on every single possible DLSS setting, as well as rasterized performance. These were pulled from the RTX 5080 but the point remains the same – the improvements to AI generated images is substantial. Much of the blurry look that permeated the first iteration of DLSS is gone, as is much of the misinterpreted “mouse tails” issues we saw in DLSS 2. DLSS 3 was a big improvement over 2, fixing repeated or jagged textures, and we see a similar uplift with DLSS 4. Now that I’ve had a few weeks to spend with it, it’s readily apparent that the new transform mode, combined with whatever other secret upsampling sauce is going on in DLSS 4 has resulted in an extraordinarily clean image quality, while maintaining a huge bump to framerate. Here’s Cyberpunk 2077 running on a 5080 with a wide variety of settings, and as requested, it’s timestamped, so go nuts:

For basis of comparison, and to demonstrate just SOME of the permutations available for DLSS.

When you DO zoom in, you’ll likely see the VASTLY improved Super Resolution image reconstruction. Standard Super Resolution had a soft and often washed out look, whereas DLSS 4 is near-indistinguishable from a native frame. The new Tensor cores are providing the power for this reconstruction. It's most evident on mid-distance objects where you might otherwise see shimmering or jagged edges, now replaced with clean edges that interact with objects and the environment properly. Things that would otherwise be obfuscated or blurred badly by motion, such as a plane propeller in Microsoft Flight Simulator 2024, are now clean and sharp as well due to the Transformer Model.

Let’s get into the topic of Multi-Frame Generation. The new Ray Reconstruction Transformer engine pulls a frame directly from the engine. It’s then passed to the Blackwell “Flip Metering” tech to generate up to three additional frames using AI. NVIDIA is stating that it can provide upwards of 8X performance boost over native rendering – something we can test in our benchmarks.

75 games had DLSS 4 support when the RTX 5090 launched. Alan Wake II, Dragon Age: The Veilguard, Stalker 2, Diablo IV, God of War: Ragnarök, Cyberpunk 2077, Marvel Rivals, Hogwarts: Legacy, Deep Rock Galactic, Still Wakes the Deep, Remnant II, Indiana Jones and the Great Circle, Senua's Saga: Hellblade II, Icarus, Hitman: World of Assassination, The Thaumaturge, Deliver Us Mars, The Finals, Star Wars: Outlaws, and much more. Put simply – it's the most bleeding edge games we're all playing, not something from the back catalog. New games like Doom: The Dark Ages, Dune Awakening, Vampire the Masquerade 2: Bloodlines, and more are coming every day.

Previous implementations of DLSS (Convolutional Neural Networks) and Super Resolution occasionally had the odd jagged edge when you look at trees or plants. The demonstration that NVIDIA liked to show was light cutting through a vent with a fan in it. It caused streaking and distorted the individual louvers in the hood. I’m happy to report that this seems to be eradicated by the new Transformer model in the 5000 series of cards. DLSS 4 uses a sophisticated frame interpolation / extrapolation / asynchronous projection AI-driven system to build three frames for every one and smoothing it in a way that looks natural to the eye. I’ll do a whole article on this at some point, but just know that it’s complex, and it’s why so much effort has been put into separating computational pipelines and using AI to assist. The end result is one constructed from various sampling points in a frame, against where it anticipates where the player might move, then recombines them in a new frame, multiple times, before it crack checks the work on the 4th frame. Broken down in a rudimentary way, think of all the empty space in a frame, and then only do calculations on the things that move and you’ll see how much computational power you just saved - that’s the new Transformer model. This, combined with over 6 years of intensive AI analysis and adjustment data (Do you remember the Battlefield V reveal trailer – yes, it’s been that long), the cards are capable of not only doubling but potentially quadrupling the visible frames (framerate might not be the right term for this) for the player. What’s amazing is just how clean the result ends up being. It’s not perfect, but I’m going to guess by the next iteration it’ll be there or damned close to it.

One of the things I really appreciate is that NVIDIA is also giving some attention to the 4000 series cards, and even some RTX cards before them. While the 4000 series doesn’t have the hardware to do multi-frame generation, they are getting an update for DLSS 3.5 for single frame gen. It’s a great gift to folks who picked up the 4000 series, and it’s honestly the biggest barrier for most folks who will now ask themselves if they need to upgrade to the 5000 series. Similarly, the 3000 and 2000 series are getting some level of upgrade, up to whatever maxes out that generation of cores and processing units in the card. For me, the improvements to things outside of games makes it worth it. If you’re a professional rendering video or AI-driven content, the 5000 series of cards deliver enough gaming and non-gaming improvements to make it an easy choice.

It’s important to note that you enable DLSS 4’s new Transformer Model on a per-game basis currently. It's not something you'll find in a game menu for every title. Instead, you'll need to head into the NVIDIA app, select "Program Settings", then select the game in question and scroll down to the heading entitled Driver Settings. Under this you'll find DLSS Override - Frame Generation. Within that option you'll see "Use the 3D application Setting" and you can select 3X or 4X as a Global option. You'll find similar options for Super Resolution, RTX, and more – just be mindful that you might want to just let the NVIDIA App tune it as you'll likely get a more optimized result for the combined hardware you have.

Reflex was introduced back in 2020, providing improvements to mouse latency for cards all the way down to the GeForce GTX 900. The way it works is by optimizing the rendering pipeline between your mouse to the CPU, to the render cue, to the GPU, then to the Display and back reducing overall latency. Reflex 2 aims to do this even more, dropping up to another 50% of the latency in this chain. Frankly, this is just magic as it can turn an otherwise jerky experience into one that is as smooth as silk. If you aren’t excited about DLSS that’s fine, but there’s plenty to be excited about here. Honestly, I can’t do an explanation justice for how frames get shifted in motion, so check out this video from the team:

Testing the effect on latency is something best left to the experts, so I can’t comment on frame warp other than to say that it feels very smooth in my experience. Anecdotally, in shooters like The Finals I saw my latency go from roughly 50ms down to 20ms in Reflex 1 and Reflex 2 down to around 4ms. That’s not a typo – the latency bounced between 2 and 5, but never higher than 5. If this quadruple reduction sounds in any way familiar, you’ve guessed it – Reflex 2 uses a similar interpolation / extrapolation / reconstruction method to anticipate where you’ll be and make sure your position is relatively close to there. In practice it’s a noticeable improvement that approaches or even exceeds what you’d get out of a “native” polling rate, if there is such a thing.

Onto the benchmarks, but first a note – we are testing pre-launch drivers, so the numbers we’ll be showing will evolve over time. Developers will continue to patch, and NVIDIA will squeeze these cards harder and harder, resulting in better performance overall. As such, please look at the benchmarks as a moment in time. They’re also rather complicated. As you can see in the hour long Cyberpunk 2077 benchmarks above, there are a LOT of options with framerate results ranging from around 30fps at 4K with everything cranked to the maximum all the way up to over 350. As such, our benchmarks will show a sprinkling of those options, but not all of them – that’s an unreadable mess. Not every game will have as extensive an option set as Cyberpunk, but suffice it to say we’ll look at raw power raster output, and then a smattering of options for DLSS 4. I’ve also been using the “quality” DLSS setting for a long time now, due to the aforementioned occasional visual issue from DLSS 3 or earlier. From what I’ve seen of the new Transformer model, when combined with Reflex 2, that might be a thing of the past. I’m struggling to find a difference between native and Performance at this point. Time will tell, but I might have to come back and update this review with a fresh set of charts when I feel comfortable saying that I’ve tested and can recommend moving from Quality to Performance. That’s a wild sentence to type out – pretty cool stuff. One last thing to mention before we get into that, though, is that you need to match the hardware all the way down the chain.

Some of our benchmarks for the ASUS Prime GeForce RTX 5070 Ti

Finding your sweet spot that matches the capabilities of your monitor or TV is key as going further than that is rather pointless. That said, there are plenty of 4K / 240Hz monitors out there, and if you’re willing to spend $750 on this GPU, what’s another $1000 for a monitor to match it? Anything above 240 is going to be in an elite tier that is so forward-looking that you likely can’t use all that power with today’s hardware – another complication. While there are competition monitors with framerates approaching insane (the upcoming TCL CSOT has a 1,000Hz refresh rate, but isn’t out yet and doesn’t have a price. It’s not gonna be cheap), you’ll have to find the balance of image quality, capability, and desired framerate yourself.

Remembering that our benchmarks are run at Quality instead of Ultra Performance, there’s a big difference between my numbers and the ones presented at NVIDIA’s presentation. Note this – I’m able to fully replicate all of the numbers they showed, but I wanted to show what sort of hit moving from Performance to Quality can cost. Again, I do want to note that this might not be a necessity any longer as the native to Performance visual difference is so minimal, especially in motion, as to be imperceptible. More important than that is that you’ll see that it’s roughly a 15-20% drop from the 5080 (linear to the price, thankfully) but a sharp generational improvement over the 4070 Ti Super – a staggering 50-60% in nearly any game tested. This card is punching far above its weight class, delivering a ton of power, a fantastic 16 GB of RAM (compared to the 12 on the 4070 Ti Super), and at a very palatable price. To help you try to find one, here's a quick list to hit at launch – may the odds be in your favor!

Putting together a review for a video card is a lot of capture and encoding. To that end, I’d be remiss if I didn’t mention the performance on the new encoders. The jump from the 3090 to the 4090 was significant, cutting it more than in half. Videos that once took 30 minutes were coming in at 13 minutes. Now that same video can be done in about 7. Obviously that isn’t going to likely matter much until you start rendering 4K or 8K video, or a particularly long one. It also seems to be smoother when running in DaVinci Resolve, but that is purely anecdotal. It’s a nice bonus, and I’m glad to see NVIDIA still investing in this area.

We always end our hardware reviews talking about price and warranty. We don’t make assertions on value as that’s different for everyone. That said, this is a $750 card, which surprisingly is the same cost as the 4070 Ti Super at launch. It’s $250 off of the 5080 price but at only a 15-20% power drop. The Asus warranty runs for three years from the date of purchase.

While it uncorks some absolutely eye-wateringly high 4K framerates, DLSS 4 will also enable the eventual lower-tier cards to deliver performance well beyond their base capabilities. Throw in the improvements on AI, the new render and encoding engine, Reflex 2, and more and it’s very easy to see the value proposition of the 5070 Ti. Obviously these cards will suffer from some artificial scarcity at launch, but be patient – don’t give into scalpers. The wait will be worth it.

Review Guidelines

ASUS Prime GeForce RTX 5070 Ti

Excellent

The ASUS Prime GeForce RTX 5070 Ti is an excellent card, delivering an awesome amount of power at a very reasonable price. DLSS 4 technology continues to mature, and with image quality to match. Combined this gives you a card that can deliver with or without upsampling, and a price that doesn’t break the bank.