If you’re like most people, you’re probably using a graphics card that’s several years old. And with, video editing, animation and other graphics-intensive activities, that few years is forever in performance. A lot has changed in the past few years, so chances are you’re no longer using the best graphics card available for taking advantage of modern technologies like smart resolution upscaling and ray-tracing acceleration. Meanwhile, games and software used for applications like 3D tools and video editors have only become more demanding.
Even if you just need the basics for streaming video or surfing the web, a good graphics card can make your system feel snappier by improving the acceleration of video decoding or redrawing your screens faster, especially if you had previously used a budget GPU. With a-equipped laptop or iMac, you can even upgrade the graphics using an external graphics processing unit (an eGPU with its own power supply) or a dedicated graphics card.
But let’s be frank: This is a horrible time to shop for a new video card. They’reor . The now-current generation of flagship cards from Nvidia and AMD launched in October 2020, with new models rolling out regularly since then, but they all continue to be in the LOL-try-to-get-one stage.
And you do want one: Nvidia’s GeForce entire available stock, creating shortages and driving up prices., , and , and AMD’s , , perform noticeably better than the previous generations. We expect the same from Nvidia’s workstation versions of the 3000 series, the A4000, A5000 and A6000. But you can’t get one — or anything, really — because cryptocurrency mining and bots have bogarted the
Nvidia started throttling its cards for crypto miningas a deterrent to having them snapped up by miners (and announced an alternate GPU line specifically for that purpose, the CMP). It’s a driver-based solution — the driver handshakes with the card’s firmware to detect and throttle mining-specific operations.
AMD’s latest Radeon RX 6600 XT hits (in theory) at a $379 price, making it the current entry to the RX 6000 series. But given its price and performance niche between the RTX 3060 and RTX 3060 Ti and intent of fast, high-quality 1080p gameplay, it’s unlikely to remain the lowest-end card in that line.
AMD chooses to take a more traditional approach to managing availability at launch — bringing as many cards as possible to market on day one and attempting to limit sales to one per customer. But as with the rest of GPUs, the bots ‘n’ middlemen who broker the cards to sell them at least twice the list price jumped into action, so you certainly can’t find either at their nominal prices.
This list is updated somewhat regularly. For the most recent updates, I’ve left the price categories in place for reference so you can see where they were before the market went nuts. Since you can’t find them to buy, the real prices are kind of moot, anyway, and stock trackers are a better friend for you.
The RTX 3000 series follow on the 20-series Super equivalents, and in the case of the 3090, the Titan RTX. The cards use the latest Ampere architecture, which has improved algorithms and more processing power dedicated to ray tracing (a second-gen Turing core), AI (for more efficient upscaling via DLSS) and programmable shaders. They deliver some big jumps in performance over the 2000 series.
AMD’s latest GPUs are based on its RDNA 2-gen architecture, used in the Infinity Cache design (all have 128MB) and enhanced design of the compute units (including a new Ray Accelerator core for each compute unit). They combine to improve the memory subsystem by reducing the latency of moving data around, increase bandwidth by up to 2.2x with a narrower path (256 bits) and deliver better energy efficiency. That also allows the processors to hit higher clock frequencies without a substantial increase in power requirements., and consoles, and for the first time target 4K gamers (the company previously concentrated on 1080p and 1440p gaming). Hardware performance improvements stem partly from the higher-density on-die
The AMD GPUs have been optimized to achieve peak performance when used in conjunction with the company’s latest DirectStorage programming interface, which accelerates SSD access by circumventing the CPU to improve storage-intensive game tasks like load times in games developed with it in mind.series of desktop CPUs (and subsequently AMD added support for the Ryzen 3000 series), though it doesn’t sound like they get much of a boost from it. If every frame counts, though, it’s something to keep in mind. They also support Microsoft’s
The new architectures for ray-tracing acceleration are accompanied by a larger set of technologies that tend to be lumped in with them because they also improve or accelerate rendering in general. These include upscaling algorithms, for example, which render for a higher resolution screen using native-resolution textures (while maintaining frame rates); in other words, using textures for 1080p to render for 1440p. Nvidia’s Deep Learning Super Sampling and AMD’s Radeon Contrast Adaptive Sharpening do this.
Ready to throw down some cash for a new graphics card for your gaming rig or laptop? Don’t spend a single cent on a graphics card for gaming until you read this buying guide, wherein we consider everything from video memory, refresh rate and frame rate to power consumption, memory clock and gaming performance. Plus, our general GPU shopping tips at the end will help you make your choice.
Sure, it’s a reasonable price. But if you’re planning to spend around $100 on a budget graphics card, don’t expect to game with the GeForce GT at 1080p — 720p at best unless a game is very lightweight, though Fortnite, CS:GO, League of Legends and other multiplayer competitive games generally fall under the “can play on a potato” umbrella. Many games may simply go from unplayable to a little less unplayable. This Nvidia graphics card does for a gaming PC what Nvidia’s MX chips do for laptops. In other words, plenty of the latest games will run on it, but many users won’t benefit. Cards can come with the chip overclocked, which gives it a little extra oomph as well.
If you’ve got an old desktop with integrated graphics that don’t support the current versions of graphics programming interfaces such as DirectX 12 or Vulkan, have a game that won’t run unless it detects dedicated graphics memory or if you just want to make your Windows experience feel a little more snappy or smooth, a GT 1030-based card can help. The GT line is designed with lower power requirements than the more popular GeForce GTX models, so it can fit in systems with lesser power supplies and compact designs. Unlike most gaming graphics cards, 1030-based cards can be low-profile and take up just a single slot for connectivity, and are quieter because they only require a single fan.
You may see a random higher-end card drop down below $100, and that’s a good choice if you’re looking for something with a little extra gaming oomph over the 1030 or support for two monitors. But they take a lot more space and power than the simple GT half-height replacement cards.
You can actually find these otherwise sub-$100 cards in stock for $130 and up.
Even before prices spike there used to be more options in the $100-$150 range. But between $150 and $200 you could theoretically find the Nvidia GeForce GTX 1650 Super-based cards and the AMD Radeon RX 5500 XT cards, both of which deliver very similar, solid entry-level 1080p gaming at low or medium settings for all but the most GPU-intensive games. Now you can at least find the 1650S in stock for $350-$400; more than is worth paying for the performance, but still in the sanity ballpark.
And since much basic photo editing still isn’t very GPU-intensive, a fast, high-core-count CPU still gives you more performance value for the money than a higher-power graphics card.
One distinction between the two that may affect your decision is power draw: the RX 5500 XT takes about 30 watts more than the 1650S. Since they’re both under 150 watts, though, your power supply probably isn’t a problem.
At the moment, cards based on the RTX 3060 are the go-to in this price class; performance is close enough to the more expensive AMD competitor, the RX 6600 XT, that it’s not really worth the extra bucks. At least, they would be if you could find them at the target base price, which Nvidia had hoped would be $329. This actually looks like it’s in stock as of this update, if you’re willing to pay upwards of $800. That’s too much, but it’s down from the $1,000-plus it was in April.
Read our EVGA GeForce RTX 3060 XC Black hands-on.
With reasonably comparable performance at “lower” prices, Nvidia’s new RTX 3060 Ti cards have a price edge over their RX 6800-based competitors, though the latter is a good graphics card to stick in an external GPU for a Mac. This card is really hard to find, even at high prices.
Read our RTX 3060 Ti hands-on.
The RTX 3070 Ti is generally a better GPU than the 3070, but not worlds better, making the cheaper RTX 3070 my choice here. One exception is for 4K video editing or workstation-level graphics, where the 3070 Ti’s higher memory bandwidth has a more noticeable impact.
You can find the RTX 3070 in stock for over $1,500; I don’t recommend spending that much unless you’re desperate, though.
Read our RTX 3070 hands-on.
As with the step-down price segment, the RX 6800 XT generally outperforms the more-expensive RTX 3080 especially at higher resolutions and in professional graphics applications, thanks to the better memory bandwidth and more video memory. But likewise, that doesn’t always hold true, especially with software that takes advantage of Nvidia CUDA.
Out of stock for the most part, unless you feel like paying $1,400 or so for it.
Read our RX 6800 XT hands-on.
I haven’t yet had a chance to test either the RTX 3090 or its competitor, the RX 6900 XT, so this is a tentative recommendation. But yet again, you can’t find it in stock anywhere, anyway. And even out of stock these cards seem to run more than $3,000. It may be a bit easier to find the RTX 3080 Ti if you want to save some money.
Things to keep in mind when shopping for the best graphics card:
- Once you’ve narrowed down your choice to a few options, searching for people’s complaints about a product is critical to discovering important information — like how many slots a card really requires as opposed to the manufacturer’s claims. It may take two slots, for example, but be just thick enough to make it impossible to put another card in a slot next to it, or just a little too long to handle a motherboard because of obstructions.
- Power consumption: Always check the power capabilities of a card against your power supply’s output. Don’t forget to take the other cards and devices in your system into account concerning power usage and the possible effect on battery life.
- Most of the negative reviews of graphics describe artifacts and failures that are usually the symptoms of overheating. If this worries you, then don’t buy an overclocked card (usually indicated by “OC” in the name). When buying cards, make sure that you have sufficient cooling and that your case’s airflow and the positions of your other cards will allow for optimal heat dissipation. That may mean, for example, moving another PCI card into a different slot.
- GTX models may be a little smaller than the RTX models and may generate less heat, and the RTX 3000 series has higher power requirements than the 2000 series.
- The most powerful GPU on the planet won’t make a difference if your CPU is the bottleneck (and vice versa) — think overkill.
- You’ll see a lot of price variation across cards using the same GPU. That’s for features such as overclocking, better cooling systems or flashy (literally) designs.
- All Nvidia GTX and RTX cards support the various flavors of G-Sync and all AMD Radeon cards RX 400 or later support AMD’s line of FreeSync adaptive refresh technologies ( ). These sync with your monitor to reduce artifacts caused by a mismatch between screen refresh rate and frame rate — so if you’re keeping your monitor, you may want to get a card that supports the right tech.
- Performance generalizations are just that — generalizations. If you’re looking to boost performance in a particular game, run a search on, say, “Fortnite benchmarks” and “best cards for Fortnite.”
- Don’t assume that replacing an old card will automatically give you noticeably better or smoother performance.
- Don’t assume that the newer Nvidia RTX 30-series cards will be consistently faster than the 20-series cards they replace.
- Dual cards are usually more of a pain than they’re worth. Video editing is usually the exception, depending upon application support.
- If you want a card for content creation, game benchmarks aren’t usually representative. To research those, start by running a search on “workstation GPUs” or, for example, “best GPU for Premiere.” It’s important to match the GPU to the application, because, for instance, Nvidia RTX A-series GPUs (the workstation GPUs formerly known as Quadro) are generally more powerful than their AMD Radeon Pro or WX series equivalents, but application developers who are tight with Apple — which doesn’t support Nvidia GPUs — optimize their applications for AMD GPUs. The biggest example of this is Blackmagic Design’s DaVinci Resolve video editor.
- For photo editing, it may no longer suffice to use a low-end or middling graphics card, though it depends on your software. With the latest generation of Photoshop and Lightroom, Adobe has begun to expand its use of AI-related technologies in meaningful ways. For instance, Photoshop’s new Replace Sky and Neural Filters can take advantage of GPU hardware designed to accelerate AI to speed them up, such as the Tensor cores in Nvidia’s RTX cards. But if you don’t have at least 32GB memory, graphics applications may get a bigger boost from upgrading that before the GPU, unless the graphics card is really old.
- For video editing, the amount of memory on the card can have a big impact on real-time performance as you work with higher-resolution video (4K and up).
Relative performance of recent GPUs
|Maingear Turbo (RTX 2080 Ti)||Microsoft Windows 10 Home (2004); 3.8GHz Ryzen 9 3900XT; 32GB DDR4 SDRAM 3,600; 11GB Nvidia GeForce RTX 2080 Ti; 1TB SSD + 4TB HDD|
|MSI Aegis RS (RTX 3060 Ti)||Microsoft Windows 10 Home (2004); 3.8GHz Intel Core i7-10700K; 16GB DDR4 SDRAM 3,000; 8GB Nvidia GeForce RTX 3060 Ti; 1TB SSD|
|MSI Aegis RS (RTX 3060)||Microsoft Windows 10 Home (2H20); 3.8GHz Intel Core i7-10700K; 16GB DDR4 SDRAM 3,000; 12GB EVGA GeForce RTX 3060 XC Black Gaming; 1TB SSD|
|MSI Aegis RS (RTX 3070 FE)||Microsoft Windows 10 Home (1909); 3.8GHz Intel Core i7-10700K; 16GB DDR4 SDRAM 3,000; 8GB Nvidia GeForce RTX 3070 Founders Edition; 1TB SSD|
|MSI Aegis RS (RTX 3070 Ti)||Microsoft Windows 10 Home (21H1); 3.8GHz Intel Core i7-10700K; 32GB DDR4 SDRAM 3,200; 8GB Nvidia GeForce RTX 3070 Ti ; 1TB SSD|
|MSI Aegis RS (RTX 3080 Ti)||Microsoft Windows 10 Home (21H1); 3.8GHz Intel Core i7-10700K; 32GB DDR4 SDRAM 3,200; 12GB Nvidia GeForce RTX 3080 Ti ; 1TB SSD|
|MSI Aegis RS (RX 6600 XT)||Microsoft Windows 10 Home (21H1); 3.8GHz Intel Core i7-10700K; 32GB DDR4 SDRAM 3,200; 8GB Asus ROG Strix Radeon RX 6600 XT OC; 1TB SSD|
|MSI Aegis RS (RX 6700 XT)||Microsoft Windows 10 Home (2H20); 3.8GHz Intel Core i7-10700K; 32GB DDR4 SDRAM 3,200; 12GB AMD Radeon RX 6700 XT; 1TB SSD|
|MSI Aegis RS (RX 6800 XT)||Microsoft Windows 10 Home (1909); 3.8GHz Intel Core i7-10700K; 16GB DDR4 SDRAM 3,000; 16GB AMD Radeon RX 6800 XT; 1TB SSD|
|MSI Aegis RS (RX 6800)||Microsoft Windows 10 Home (1909); 3.8GHz Intel Core i7-10700K; 16GB DDR4 SDRAM 3,000; 16GB AMD Radeon RX 6800; 1TB SSD|
|MSI Trident X (RTX 2070 Super)||Microsoft Windows 10 Home (1909); (oc) 3.8GHz Intel Core i7-10700K; 32GB DDR4 SDRAM 2,932; 8GB Nvidia GeForce RTX 2070 Super; 1TB SSD|
|Origin PC Chronos (RTX 3080)||Microsoft Windows 10 Home (2004); Intel Core i9-10900K; 16GB DDR4 SDRAM 3,200; 10GB Nvidia GeForce RTX 3080 (EVGA); 1TB SSD + 500GB SSD|