I Review Graphics Cards for a Living, Let Me Help You Pick an Nvidia GPU

Whether you’ve been playing PC games for years oregon you’re gathering your archetypal gaming PC, picking the close graphics paper is astir apt 1 of the astir consequential choices. It doesn’t assistance that present there’s 3 companies making graphics cards, which means determination are much options retired determination than ever before.

But determination are a batch of reasons to prime an Nvidia graphics paper implicit 1 made by AMD oregon Intel. For one, whether you emotion it oregon hatred it, Nvidia has plentifulness of bundle and features that are exclusive to its graphics cards, whereas things similar FSR (FidelityFX Super Resolution) from AMD volition enactment nary substance what GPU is successful your system. Features similar DLSS, oregon Deep Learning Super Sampling, marque an Nvidia graphics paper an charismatic choice.

Nvidia besides has the astir almighty graphics cards connected the marketplace close now, particularly astatine the high-end. There simply is not different graphics paper that tin lucifer the RTX 5090 successful axenic performance. But adjacent erstwhile it comes to much affordable graphics cards similar the RTX 4060 Ti, the AI upscaling offered by DLSS tin truly assistance agelong the performance, and tin adjacent propulsion into higher resolutions than you would different beryllium capable to.

It's inactive important to fig retired what solution you privation to play games at, due to the fact that that’s going to alteration what benignant of graphics paper you privation to spell for. If you privation to play each the newest PC games astatine 4K with each the settings maxed, you’re going to request a overmuch much almighty paper than idiosyncratic that conscionable wants to play World of Warcraft astatine 1080p.

TL;DR: These Are Best GPUs

Graphics Cards Basics

While graphics cards are highly analyzable devices, buying for 1 doesn’t request to beryllium that overmuch of a hassle. As agelong arsenic you support an oculus retired for immoderate elemental specs, you should person a beauteous bully thought of what you’re getting into.

The astir evident happening to look for is whether oregon not your graphics paper is really portion of the existent generation. You don’t privation to miss retired connected show oregon features, particularly if you’re spending hundreds of dollars connected conscionable one part of your computer. With Nvidia graphics cards, this is really highly easy, conscionable look astatine the 2 numbers pursuing the ‘RTX’ oregon ‘GTX’ successful the graphics card’s name. Nvidia conscionable launched its newest procreation of graphics cards, which are labeled with '50.' If it’s ‘40,’ that’s the erstwhile procreation from 2022; ‘30’ and ‘20’ are the 2 generations earlier that.

Don’t get maine wrong, getting a 30-series oregon 20-series paper volition inactive supply an fantabulous gaming experience, with the RTX 3080 inactive capable to powerfulness astir games astatine 4K without breaking a sweat.

But what astir the 2nd 2 numbers? Well, instrumentality a look astatine the RTX 5090, the top-end graphics paper connected the market. Then, instrumentality a look astatine the RTX 4050, which is lone a laptop GPU, but is way weaker. The standard typically goes from 50-90, with the higher fig meaning a much almighty graphics card. Nvidia does sometimes adhd other letters oregon words astatine the extremity of its graphics cards, with ‘Ti’ and ‘Super’ being communal variants. This typically means a somewhat much almighty card. For instance, the Nvidia GeForce RTX 5070 Ti is going to beryllium somewhat much almighty than the RTX 5070. The basal regularisation of thumb is that, aft the number, immoderate other letters oregon words typically bespeak a faster graphics card.

When it comes to specs, graphics cards person a batch of numbers and lingo to swallow, which tin decidedly beryllium overwhelming. But if you conscionable privation to slot a GPU successful your rig and hide astir it, you don’t request to wage attraction to each small detail.

The magnitude of VRAM you request is going to mostly beryllium connected what solution you privation to play your games at. If you privation to play games astatine 4K, you’re going to privation arsenic overmuch video representation arsenic you tin afford. There are games that volition devour up upwards of 20GB of VRAM astatine that solution if you fto them, truthful the entity truly is the limit. For little resolutions similar 1080p, however, you tin get by with 8GB of RAM, though I would urge going with a 12GB oregon 16GB paper if it’s successful your budget. An 8GB paper volition get you done astir games, but determination is simply a increasing fig of AAA PC games that request much memory.

There are immoderate different specs you tin look astatine if you truly privation to, similar timepiece speed, CUDA halfway number and Streaming Multiprocessors – what Nvidia calls its compute units – but those mostly conscionable get amended arsenic you get a much high-end graphics paper either way. Compute Units are fundamentally the aforesaid arsenic CPU cores connected a processor, and each 1 contains 128 CUDA cores. So, the RTX 5080 with its 84 Compute Units, contains a full of 10,752 CUDA cores. Just support successful caput that straight comparing 2 graphics cards of antithetic generations by the halfway counts unsocial won’t archer you however overmuch of a show leap it’ll be.

Once you’ve picked the graphics paper you want, you request to marque definite you person a powerfulness proviso that tin grip it. You tin usually cheque the container your graphics paper comes successful to get an thought of however overmuch powerfulness the paper requires, and if not, Nvidia has it listed retired connected its website. My proposal would beryllium to cheque the required powerfulness for your graphics cards and past get a powerfulness proviso that tin supply conscionable a spot much wattage. For instance, if your graphics paper recommends a 450W powerfulness supply, I’d get a 550 oregon a 650W portion conscionable to beryllium connected the harmless side.

1. If You Just Want the Best: RTX 5090

2. Best for 4K (for Most People): RTX 5080

3. Best for 1440p: RTX 4070 Super

4. Best for 1080p: RTX 4060 Ti

5. Best connected a Budget: GeForce GTX 1660 Super

Upcoming Nvidia Graphics Cards

The archetypal question of Nvidia Blackwell graphics cards are each here, and it seems similar that's each we're going to get for now. Usually, Team Green would travel up its archetypal question of graphics cards with refreshed "Super" variants. However, thing successful the satellite of PC gaming hardware is mean close now, and it looks similar the Super lineup has been canceled successful the look of precocious RAM prices and AI datacenter demand. Right now, it looks similar we'll beryllium waiting for the RTX 6090 whenever that comes retired – apt sometime successful the 2nd fractional of 2027.

What Is DLSS?

Deep Learning Super Sampling, oregon DLSS is an AI upscaling method that uses Nvidia’s Tensor Cores to amended representation quality. The Tensor Core volition instrumentality ocular information from the framework being rendered, on with question vector data, successful bid to accurately upscale the representation to a higher resolution. This process results successful an representation that looks beauteous adjacent to the autochthonal resolution, but with a overmuch higher framework rate.

In its aboriginal days, DLSS needed to bid the AI exemplary connected each crippled it would support, with developers needing to upload information to Nvidia. However, DLSS has gone done respective iterations, and present it does not request to beryllium trained connected idiosyncratic games. Instead, if a developer wants to see DLSS successful their game, they tin conscionable inject Nvidia’s API into the game. This means much games tin enactment the technology, without having to hold for Nvidia’s grooming connected each game.

With DLSS 3.0, Nvidia added Frame Generation to the equation. This exertion takes the ocular information from 2 frames, on with question vector information from the crippled motor and question accusation from its ain hardware, and creates an wholly caller framework that’s sandwiched betwixt the 2 archetypal frames. This would present a batch of latency, but Nvidia requires its Reflex exertion to beryllium enabled earlier Frame Generation tin happen. Reflex fundamentally syncs the graphics paper and the processor, eliminating the request for the CPU to queue up frames for the GPU to render later.

Nvidia updated to DLSS 4.0, past aboriginal 4.5, which introduced multi-frame procreation and a caller upscaling model. The caller framework procreation lets RTX 5000 graphics cards usage AI to make up to 3 frames per rendered frame, greatly improving framerates. The caller DLSS besides moves the upscaling algorithm to a transformer model, which allows for overmuch amended representation quality, though with a flimsy deed to performance. But, hey – that's what framework procreation is for, I guess.

Together, Frame Generation and Reflex greatly amended your framework rate, but there’s a catch. Because truthful overmuch of it relies connected question data, you already request to person a decent framework complaint for Frame Generation to enactment smoothly. So, this exertion is champion utilized for folks that tin already get 60fps oregon much and conscionable privation to propulsion to a higher framework rate.

What Is Ray Tracing?

Ray tracing is conscionable a mode to render airy realistically. It does this by taking a airy source, past simulating each ray of airy arsenic it bounces astir the scene. It’s a elemental conception to beryllium sure, but it ends up requiring a ton of compute powerfulness to propulsion off. Any airy being formed perchance has thousands of rays of light, each of which volition bounce astir hundreds of times, multiply that by needing a caller framework 60 times a second, and you tin ideate however overmuch powerfulness you request to propulsion it off.

That’s wherefore ray tracing successful video games needs specialized hardware to propulsion this lighting method disconnected without grinding your framerate to a standstill. Luckily, Nvidia has been moving this hardware into its graphics cards since the RTX 2080 successful 2018.

Fast guardant to today, and adjacent with that dedicated ray tracing hardware being built into each mainstream graphics paper – not conscionable from Nvidia – we inactive request to bounds ray tracing successful bid to maximize performance. There are lone a fewer games that enactment afloat ray tracing, oregon ‘path tracing,’ with astir different games limiting the ray tracing elements to a definite portion of the scene, similar shadows oregon reflections, and besides limiting the magnitude of bounces calculated for each ray of light.

It’s an costly mode to make lighting, but it looks incredible, particularly successful games that trust connected lighting for atmosphere. In Metro Exodus, for instance, the close lighting often leads to darker environments, amplifying the aggravated ambiance the crippled already has.

Jackie Thomas is the Hardware and Buying Guides Editor astatine IGN and the PC components queen. You tin travel her @Jackiecobra