Positioning & Architecture
The GeForce RTX 5060 Ti is NVIDIA's fifth graphics card launch this year, and the Blackwell gaming stack is almost complete. Last week we saw a huge number of RTX 5060 Ti 16 GB reviews and not a single 8 GB model, even though both were launched at the same time. I asked the various board partners for samples of the 8 GB card, but nobody was able to provide anything. That's why I bought a card on launch day, like a normal consumer, no special press treatment, no bot hitting F5. I just went to the usual big online stores here in Germany, checked what 8 GB models they had in stock and purchased one. Due to the Easter holidays, shipping took a bit longer, but here we are, 8 GB RTX 5060 Ti reviewed. Later in May, probably around Computex time, the RTX 5060 non-Ti will be launched, which has an 8 GB model only.

With the RTX 5060 Series, NVIDIA is introducing the GB206 graphics processor to their lineup. It comes with 4,608 cores, which are all enabled in the RTX 5060 Ti. Compared to the RTX 4060 Ti, that's an increase of 6%. You also get 48 ROPs, 36 RT cores and a 128-bit GDDR6 memory interface. As mentioned before, the RTX 5060 Ti is available in both 8 GB and 16 GB variants, except for the memory size, all base specs are identical.

The Blackwell architecture introduces several architectural improvements under the hood, like giving all shaders the ability to run FP32 or INT32 instructions, on Ada only half the cores had that ability. The Tensor Cores are now accessible from the shaders through a new Microsoft DirectX API, and they now support FP4 and INT4 instructions which run at lower precision, but much faster with less memory usage. There's numerous additional architecture improvements, we talked about all of them on the first pages of this review.

From a fabrication perspective nothing has changed though—Blackwell is built on the same 5 nanometer "NVIDIA 4N" TSMC node as last generation's Ada. NVIDIA claims this is a "4 nanometer process," but during Ada it was confirmed that NVIDIA 4N is actually not TSMC N4 (note the order of N and 4), but 5 nanometer. At the end of the day the actual number doesn't matter much, what's important is that NVIDIA is using the same process node.

Performance
For this launch we've updated our test setup again and retested all comparison cards with the newest drivers. We also updated the BIOS on our 9800X3D and added several new games, like our first RT exclusive title Indiana Jones, and Path Tracing is now an additional section in all reviews. At 1080p, the RTX 5060 Ti 8 GB is able to match the performance of the 16 GB model exactly. If we take a look at individual game tests, we can see that in most games the 8 GB model is like 2-3% faster than the 16 GB model—how can that be? It seems that manufacturing variances between GPUs, even from the same series, can result in a different boosting behavior. Typically, when not constrained by VRAM, the 8 GB model boosts around 50 MHz higher than the 16 GB model. The 8 GB card is a normal card bought in retail, the 16 GB is a review sample, so it's not a case of golden sample. Both cards have the same specs: 2572 MHz Boost, 2407 MHz base, same 180 W power limit on both.

At 1080p, we can find several titles that run at significantly lower FPS, because they're running out of VRAM: Spider-Man 2, The Last of Us and Monster Hunter Wilds. Remember, we're testing at max settings, no RT, native, no upscalers. I'll go into more game-specific details a bit later in this conclusion. Once we bumped the settings to 1440p, the 8 GB model ran out of steam even more, but 5% slower than the 16 GB model on average is definitely not "falling apart." At 4K, the differences are pretty drastic though, with the 16 GB card offering 26% more frames per second. Now the games list shows 13 out of 24 games as running out of VRAM.

Ray Tracing & Neural Rendering
Ray tracing is the future and Blackwell comes with several improvements here. The problem is that with 8 GB VRAM, ray tracing will push VRAM usage well over 8 GB. Even at 1080p (but without upscaling), we're seeing five out of nine titles impacted, Indiana Jones will even crash during level loading (due to the highest texture setting). Does that mean RTX 5060 Ti 8 GB is unfit for ray tracing? I don't think so. If you play with the settings carefully, mix and match upscalers and frame generation, you'll still be able to achieve a decent gaming experience in virtually all games. Can you set details to ultra, max out RT, and expect a great gaming experience, even at 1440p and 4K—no way, there will be compromises.

I also noticed that some games with RT do run a bit better on the RTX 4060 Ti 8 GB than on the RTX 5060 Ti 8 GB, which is unexpected, because the latter has equal or more of everything. Since this happens only when out of VRAM, it could be due to how Blackwell manages its VRAM internally, which of course isn't publicly documented anywhere.

With Blackwell, NVIDIA is introducing several new technologies. The most interesting one is Neural Rendering, which is exposed through a Microsoft DirectX API (Cooperative Vectors). Technically, this feature will be able to lower the VRAM requirements of games considerably, because textures can be compressed much more efficiently, with no or minimal loss in image quality. It's also much easier to define materials as math formula instead of pixels, which takes up much less memory as well. Game support for Cooperative Vectors is basically non-existent yet, so I don't think this will see widespread use in a lot of games any time soon enough to make a difference for the RTX 5060 Ti 8 GB.

Individual Gaming Experiences
For my reviews I test all games at the same settings. This standardized approach allows an objective performance comparison, but of course it has its drawbacks. Many gamers don't play at Ultra settings, and lowering settings often yields free FPS with no visual loss in image quality. There's also DLSS upscaling and Frame Generation. That's why I spent a few hours on popular "problematic" titles to see if I could get them into an "enjoyable" state, and I mean more than just "playable," where the game simply starts and can be played. I'm looking for a good gaming experience that's fun and doesn't constantly make me think, "I should have bought a console or a faster GPU." To achieve this you must stop using the "low," "medium," "high" profiles, and change settings individually. No doubt, this introduces subjectiveness, and sorry for making the conclusion so much longer.

Alan Wake 2: With RT, the game simply runs too slow. I found that disabling RT at 1440p with max settings runs very well, as long as you enable DLSS Quality. Thanks to the Transformer DLSS model, it still looks fantastic. To me, DLSS Frame Generation x2 provided the best gaming experience, FG off was too stuttery, and DLSS x3 and x4 had a bit too much latency. Of course, you can improve latency with a higher DLSS upscaling factor like Balanced or Performance, but I rather have the fine details of DLSS Quality than the extra FPS of FG x3. If you really must have ray tracing, RT Low is doable, especially with DLSS Balanced—not worth it in my opinion.

Assassin's Creed Shadows: The game would always crash after a fresh install, before I had the chance to adjust any settings. I also tried manually editing the auto-generated config files—didn't help. Ready to declare this game a total failure on RTX 5060 Ti 8 GB, I googled and found discussions that suggest some game bug in a DLC, and that deleting the DLC10 folder in the game fixes it—indeed it did. Good settings for me were Max settings, RT only in home area, DLSS Quality, no Frame Generation. If you really want DLSS Frame Generation, then you must at least enable DLSS Performance, dial down some secondary settings, to have a base FPS that's high enough so that the latency from FG isn't noticeable.

Cyberpunk 2077: 1440p, with RT off, DLSS Quality and Frame Generation x2 ran extremely well.

Star Wars Outlaws: You must absolutely use DLSS Quality at least, that lets you run 1440p just fine at max settings. Frame generation helps for a smoother experience, I had no issues with FG latency.

Dragon Age: Veilguard: Activating Frame Generation increases VRAM usage in all games, which eats into your VRAM budget. In this title specifically, I felt that FG off gave me some "extra" memory for textures, which resulted in a much nicer look overall. You definitely have to use DLSS Performance, but with the Transformer model this looks good enough to not be distracting during gameplay. Also disable ray tracing, so you don't have to dial down the textures and the RT performance hit just isn't worth the minimal improvement in image quality.

Indiana Jones: This game is Vulcan based, which means that it will crash when running out of VRAM. Unlike DirectX-based games, there is no standard mechanism that is able to move resources from the GPU into main memory, roughly similar to how the pagefile works for main memory. Of course this is MUCH slower, but games with DirectX won't just crash when running out of VRAM (usually). In Indiana Jones this means that you should target 7 GB VRAM usage instead of 8 GB, to have a cushion for busy scenes that might just use a bit more VRAM—you really don't want to just crash in those. I found DLSS Quality, Max everything, textures on low an excellent starting point and the game looks very good overall. Textures on high with everything else on low is an option, too, but doesn't look nearly as good, because of increased pop-in and worse geometry. Frame generation can be hit-and-miss in this game, it often resulted in FPS cut in half, instead of doubled. It appears to me that the frame generation part is able to end up in main memory without crashing, which will result in a huge performance penalty of course. Enabling it can still be viable, but that means lowering a lot of other settings to free up enough VRAM—it wasn't worth it to me, especially since the performance of Indiana Jones is pretty good already.

Monster Hunter Wilds: When running out of VRAM, MHW does have a tendency to crash, but it will happily stutter, too, while running out of VRAM and using system memory, I also noticed that it does end up with worse 1% lows more quickly than other titles. You should definitely turn off ray tracing, and enable at least DLSS Quality. Texture memory must be dialed down, I found textures at "medium" a good compromise, but that only works when using DLSS Quality or higher. DLSS Upscaling will render your games at a lower resolution, which lowers the VRAM usage, so that's a good dial to bring memory consumption down.

This short list confirms that even problematic titles can run well, but it also confirms that it's easy to find a scenario where 8 GB VRAM will fail, and 16 GB will not—especially at 4K. While RTX 5060 Ti 16 GB can run many games quite well at adjusted settings, do not expect this to be the case with the 8 GB model. In many titles at 4K, no matter what you do, the experience will suck. Is that unreasonable for a x60 class card? I don't think so. After reading this list you'll probably think "but there are so many settings to change"—yes, that's right. Unless you are willing to play with settings—beyond just selecting a "low" or "medium" profile, you will not get the optimal gaming experience. If you're not willing to work the settings, do not buy an 8 GB card. I think the optimal scenario for a RTX 5060 Ti 8 GB is running at 1080p, not 1440p. While it can handle 1440p in most games, I am not confident enough to say that this will be the case for the years to come, and you will definitely have to dial down settings. No doubt, these compromises are not what you want on a $400 GPU, on the other hand, it's not unexpected that an x60-class card can't run everything at ultra?

VRAM
Ideally, there would be a 12 GB VRAM model of the RTX 5060 Ti, which would have been a good middle choice, as that's good enough for virtually all titles at 1080p and 1440p, but more economical. NVIDIA designed their GPU with a 128-bit memory bus, which means four memory chips, which each connect using a 32-bit interface. On the RTX 5060 Ti 16 GB, NVIDIA doubled up the memories to eight chips, so that two each share a 32-bit interface (they do not use larger capacity chips). A 192-bit bus would certainly be possible, like on RTX 5070, but that would have required a wider bus design in the GPU, with support for the extra bus width, and more pins in the design, and a more complex PCB, etc. All these changes would make the card more expensive, too.

DLSS 4 Upscaling & Frame Generation
For the RTX 5060 Ti 8 GB, DLSS is one of the most important capabilities to have, and it's the only way to achieve a good gaming experience in AAA titles without having to go to low settings. First of all, DLSS 4 Multi-Frame-Generation. While DLSS 3 doubled the framerates by generating a single new frame, DLSS 4 can now triple or quadruple the frame count. While this sounds fantastic in theory, for the RTX 5060 Ti it's not a fire-and-forget solution. When using FG, gaming latency does NOT scale linearly with FPS, you need a base FPS of like 40 or 50, or you will notice the added input latency, even though the movements on the screen are super smooth.

Specifically for the 8 GB model, it's important, too, to be aware of the fact that enabling Frame Generation will increase the VRAM usage, by 1 GB or so—memory that's now not available for the game's textures and models. That's why it's sometimes better to leave DLSS FG disabled and use upscaling for better performance, especially in games that tend to use a lot of VRAM, but run very well when it comes to the amount of GPU compute they do per frame. One such example is Indiana Jones.

It also depends on the gameplay, in slower, single-player, games it's easy to tolerate a bit more input latency than in fast-paced multiplayer shooters where every millisecond counts.

In the past there were a lot of debates whether DLSS upscaling image quality is good enough, some people even claimed "better than native"—I strongly disagree with that—I'm one of the people who are allergic to DLSS 3 upscaling, even at "quality." With Blackwell, NVIDIA introduced a "Transformer" upscaling model for DLSS, which is a major improvement over the previous "CNN" model. I tested Transformer and I'm in love. The image quality is so good, "Quality" looks like native, sometimes better. There is no more flickering or low-res smeared out textures on the horizon. Thin wires are crystal clear, even at sub-4K resolution! You really have to see it for yourself to appreciate it, it's almost like magic. The best thing? DLSS Transformer is available not only on GeForce 50, but uses Tensor Cores on all GeForce RTX cards! While it comes with a roughly 10% performance hit compared to CNN, I would never go back to CNN.

Both these technologies are very important, and they are the reason why I think the RTX 5060 Ti 8 GB is a better buy than the Arc B580 12 GB, for example. While the Arc card is more affordable, and has more VRAM, it lacks DLSS, which is what's making the difference to turn stuttery gameplay at native into a fluid experience that looks good, without a lot of upscaling artifacts, and that is supported in pretty much all games. Sure, if you look hard you can spot something that's not perfect, and if you stop playing the game and start reviewing the latency you might be able to feel something, but if you really just plan on enjoying your games it's a great alternative to spending a few hundred dollars extra on a faster GPU.

Physical Design, Heat & Noise
The Gainward RTX 5060 Ti Ghost is a cost-optimized "base" implementation of the RTX 5060 Ti. You get a plastic cooler and backplate, but you do get an RGB lighting element, for a little bit of extra bling. Still, the card definitely feels more plasticky than other RTX 5060 Tis we've reviewed—but it's more affordable, too, of course. The cooler is the weakest out of the cards we've tested, but it handles the heat output from the GPU just fine and is "almost quiet" when it comes to noise levels, which is better than what I have expected, thanks to well-balanced fan settings.

PCI-Express 5.0
NVIDIA's GeForce Blackwell graphics cards are the first high-end consumer models to support PCI-Express 5.0. This increases the available PCIe bandwidth to the GPU, yielding a small performance benefit. Of course PCIe Gen 5 is backwards compatible with older versions, so you'll be able to run the cards in an older computer, too. The RTX 5060 Series specifically operates at PCIe 5.0 x8, not x16 like the other RTX 50 Blackwell cards. This cuts the available bandwidth in half, but the switch from Gen 4 to Gen 5 doubled it.

We tested PCI-Express scaling of the RTX 5060 Ti 16 GB here. I will rerun the same tests with the 8 GB model, to see what effect limited bandwidth has when running out of VRAM, and system memory is used as overflow.

Power Consumption
Power consumption of the RTX 5060 Ti is good. While some other Blackwell cards had quite high power consumption in idle, multi-monitor and media playback, this isn't a problem at all here. As mentioned before, the 8 GB models do use a little bit less power for the VRAM, because they have just four chips instead of eight, which helps with the power limit, because it makes some extra power headroom available for the GPU. Gainward's Ghost card does not allow any manual power limit increases, 180 W is the fixed power limit.

Overclocking
We've seen good overclocking potential from some previous Blackwell cards, but the RTX 5060 Ti is remarkable. With manual OC we were able to increase the clocks by 400 to 500 MHz, which turned into a +13% real-life improvement. That's real FPS gains, not some theoretical increase, and that's the missing performance that would have turned RTX 5060 Ti's gen-over-gen rating from "meh" to "nice." What puzzles me is that NVIDIA must know about this, they spend a significant amount of time on planning the final frequencies, hundreds of people are busy benchmarking, comparing, theory-crafting, and then they launch an essentially underclocked product. Power is not the problem here, it increases only marginally. Maybe to tempt people to buy RTX 5070? Maybe to leave room for RTX 5060 Ti Super? Maybe so they can maximize harvesting and use chips that don't clock so high due to manufacturing variances? I don't know, it's a lost opportunity, but excellent potential for people who are willing to do some manual overclocking. Memory overclocking is equally impressive and topped out at NVIDIA's artificial cap of +375 MHz / +3000 MT/s—the chips could certainly take more.

Pricing & Alternatives
NVIDIA's MSRP for the RTX 5060 Ti 8 GB is set at $380, whereas the 16 GB model comes at an MSRP of $430. Finding the 16 GB model at $430 seems very hard, while finding the 8 GB card at MSRP is much easier, at least here in Europe. In the States, it seems that the GPU market is in a completely broken state. I bought this review sample for €400, including 20% VAT, so €333 without tax, which converts to $382—pretty much matching the MSRP. Considering all the compromises that are needed for the 8 GB card to run well, it's a no-brainer to spend an extra $50 for the 16 GB model, if that can be found for $430. It does look like the 16 GB card's real MSRP is actually $500 and that there is a limited-time rebate for the first shipments to hit the $430 price point. At a $120 price difference, I'd definitely start considering the 8 GB model, especially when money is tight, and the primary screen resolution is 1080p with no focus on ray tracing. As mentioned before, you MUST be willing to adjust the settings. If you don't want to go that route—buy the 16 GB model, or some of the alternatives I'll be discussing next.

AMD's Radeon RX 7700 XT comes with 12 GB VRAM, and offers better performance, especially if you're targeting 4K with no upscaling. With upscaling, the DLSS Transformer model will give you a much better image quality, and 7700 XT will not support the new FSR 4 AI-powered model. Ray tracing without upscaling runs better on the 7700 XT than on the 5060 Ti 8 GB—pretty surprising, considering how weak AMD's RT cores are.

Buying a last-generation RTX 4060 Ti 8 GB will not solve your VRAM problem, but if the pricing is good, it could be an option, DLSS multi-frame-generation is not the most important thing to have in this performance bracket. You will still have x2 frame generation, which is supported on the RTX 40-Series cards, and they also support the new Transformer upscaling model, which is the most important feature in this segment in my opinion.

The RTX 4060 Ti 16 GB is a very solid option (assuming a good price of course), the extra VRAM will give you peace of mind. My favorite is the RTX 4070 non-Ti, which is 12 GB and has a lot of extra GPU core horsepower, resulting in a much better gaming experience in virtually all scenarios. Here, too, pricing matters, things get interesting around $500.

If you want ray tracing, then RTX 5060 Ti 16 GB is your best option, and of course anything upcoming from AMD in that price bracket—we've been hearing rumors about an RX 9060 Series after Computex, an RX 9070 GRE is also likely, but both are unknowns in terms of performance and pricing. We do know that they use the RDNA 4 architecture, which means support for FSR 4 and stronger RT cores. Maybe Intel has an ace up their sleeves, their bigger Arc Battlemage cards could add more competition, but the release date is completely unknown, some rumors say that those cards have been canceled.