Here are the best graphics cards for gaming, from high-end to budget solutions.

(Image credit: digitaltrends)

The best graphics cards power any gaming PC, and everything else follows. Even the fastest gaming CPUs can’t do much without a strong GPU pushing pixels. Since no graphics card is ideal for everyone, we’ve included options for every budget and mindset. Let us help you find the fastest, best, or most affordable graphics card.

Where our GPU benchmarks hierarchy evaluates cards solely by performance, our best graphics cards ranking considers the full package. GPU pricing, performance, features, efficiency, and availability are significant, but weighting is subjective. Considering all those factors, these are the greatest graphics cards available.A ‘new’ GPU hasn’t been released since February’s RX 7900 GRE, joining the RTX 4080 Super, 4070 Ti Super, 4070 Super, and RX 7600 XT from January. All of these are on our performance charts, and some replace discontinued entries in our overall picks.

Except for a surprise announcement in the next months, this should be the last new model until the Nvidia Blackwell and RTX 50-series, AMD RDNA 4, and Intel Battlemage GPUs arrive, possibly in fall 2024. Nvidia’s SFF-ready GPU specs could be considered new GPUs, however they may compromise performance to fit in a smaller volume.

Most GPUs’ lowest prices are MSRP or lower. That includes the $976 RTX 4080 Super, but the cheapest 4090 costs $1,730—better wait for the 5090. This guideline applies to most high-end cards: Waiting till October/November to upgrade makes sense.

(Image credit: techradar)

1. Nvidia GeForce RTX 4070 Super

The best overall GPU right now

SPECIFICATIONS

GPU: AD104

GPU Cores: 7168

Boost Clock: 2,475 MHz

Video RAM: 12GB GDDR6X 21 Gbps

TGP: 200 watts

TODAY’S BEST DEALS

Nvidia introduced Super models in 2024 to its 40-series roster. Of the three, the RTX 4070 Super will likely be most popular. It has all the latest Nvidia Ada Lovelace architecture features and the same $599 MSRP as the non-Super 4070 (now $549 to stay relevant). It’s barely better than a linear performance-to-price bump, which is good enough today.

Many RTX 4070 Super base-MSRP models are available at retail. The Founders Edition’s stealthy black design and silent operation are appealing, but third-party cards with better cooling are available at cheaper rates.

Over 20% more cores than the vanilla 4070, the 4070 Super performs well in our testing despite the memory subsystem’s lack of upgrades. Even with the same VRAM and bandwidth, it’s 16% faster at 1440p, because to the 33% increase in L2 cache size.

Even with less raw bandwidth, the 4070 Super outperforms the RTX 3080 Ti and the previous generation Ampere GPUs. It does all that while saving over 100W, which is astounding.

(Image credit: reddit)

2. AMD Radeon RX 7900 GRE

AMD’s best overall GPU right now

SPECIFICATIONS

GPU: RDNA 3 Navi 31

GPU Cores: 5120

Boost Clock: 2,245 MHz

Video RAM: 16GB GDDR6 18 Gbps

TBP: 260 watts

TODAY’S BEST DEALS

The Radeon RX 7900 GRE replaces the RX 7800 XT as our top AMD selection. It employs AMD RDNA 3 architecture but has 33% more compute units and the Navi 31 GCD instead of the 7800 XT’s Navi 32 GPU. After fixing the 7900 GRE overclocking ‘problem’, AMD balances that by limiting GPU and GDDR6 clocks. Overclocking may make up for those losses. The 7800 XT is a good alternative and a close second.

Although the 7900 GRE is 10% faster and 10% more expensive than the 7800 XT, both models offer acceptable performance and affordability. Team Red’s best card is the more expensive one due to linear performance scaling on high-end GPUs. This assumes you’re like us and want a superb 1440p gaming experience for less money. There are faster GPUs, but the 7900 XT costs 27% more and performs 20% better, so diminishing returns begin.

RDNA 3’s efficiency has been inconsistent, however the 7900 GRE is AMD’s most efficient GPU due to lower speeds. More processing clusters at lower clocks boost efficiency. It uses over 60W less power and is faster than the 6950 XT. AV1 encoding, DP2.1 video output, and increased computing and AI capabilities give it over three times the image throughput of the 6950 XT in Stable Diffusion.

AMD outperforms Nvidia in rasterization games but lags in ray tracing, especially in full path tracing games like Alan Wake 2. Want to try it? Though it improves graphics, it doesn’t impact gameplay. The best option for fully path traced games, including future RTX Remix mods, is Nvidia. Otherwise, AMD’s 7900 GRE rasterization matches the 4070 Super at 1440p for $50 cheaper.

We now want 16GB of memory for GPUs over $500. Nvidia fails there, although its 12GB cards look to match AMD’s 16GB cards. However, the RX 7900 GRE meets all the criteria for a high-end graphics card that most gamers can afford.

(Image credit: ign)

3. Nvidia GeForce RTX 4070

A finely-balanced high-end Nvidia GPU

SPECIFICATIONS

GPU: AD104

GPU Cores: 5888

Boost Clock: 2,475 MHz

Video RAM: 12GB GDDR6X 21 Gbps

TGP: 200 watts

TODAY’S BEST DEALS

Our review of Nvidia’s RTX 4070 found it to be comparable to the RTX 3080, but it has the latest Ada Lovelace architecture and features and costs $150 less. With the release of the RTX 4070 Super (see above), the lowest-priced cards now cost roughly $530.

We compared the RTX 4070 and RX 7900 GRE and chose the Nvidia card. This is due to Nvidia’s ecosystem, including DLSS, AI, and efficient architecture. If you solely care about performance, the RX 7900 GRE is tempting.

Nvidia rarely targets the actual value market, but the 40-series Super cards’ price changes are appropriate. The RTX 4070 supports DLSS, needs 200W of power (typically less), and outperforms AMD’s RX 7800 XT in raw performance—slightly slower in rasterization, faster in ray tracing.

Nvidia always touts the RTX 40-series’ faster performance with DLSS 3 Frame Generation. As mentioned previously, produced frames aren’t “real” frames and increase input latency. DLSS 3 isn’t horrible, but we prefer to compare non-enhanced performance, and we’d say it increases the feel by 10–20%, not 50–100% as Nvidia claims.

The option between the RTX 4070 and the 4070 Super depends on whether you’d rather save money or spend more for superior performance. Though the 4070 Super costs 13% more and performs 16% better, they offer similar value.

(Image credit: reviewed.usatoday)

4. Nvidia GeForce RTX 4090

The fastest GPU — great for creators, AI, and professionals

SPECIFICATIONS

GPU: Ada AD102

GPU Cores: 16384

Boost Clock: 2,520 MHz

Video RAM: 24GB GDDR6X 21 Gbps

TGP: 450 watts

TODAY’S BEST DEALS

Some consider the fastest graphics card the best, regardless of price. Nvidia’s GeForce RTX 4090 targets this group. It introduced Nvidia’s Ada Lovelace architecture and is the company’s most powerful card until later this year’s Blackwell GPUs.

Due to exorbitant price, many RTX 4090 cards now sell above $2,000. Due to China’s RTX 4090 export restrictions and the advent of AI and deep learning, many AI organizations are using RTX 4090 cards instead of paying three times more for the slower RTX 6000 Ada Generation. If you don’t have a 4090, skip it.

The RTX 4090 is further ahead of the next Nvidia GPU. It’s 35% faster than the RTX 4080 at 4K and 32% faster than the Super in our gaming benchmarks. It’s 47% quicker than AMD’s flagship RX 7900 XTX, but it costs nearly twice as much online.

Let’s be clear: the RTX 4090 works best on a high-refresh rate 4K panel. Its advantage over a 4080 Super drops to 22% at 1440p and 13% at 1080p, including demanding DXR titles. The 1080p lead over the RX 7900 XTX drops to 27%. To get the most out of the 4090, you need a high-resolution, high-refresh rate display and the quickest CPU.

It goes beyond gaming performance. The RTX 4090 outperforms the 4080 Super by 42% in Blender, Octane, and V-Ray. Blender is over three times quicker than the RX 7900 XTX. Don’t even mention AI tasks. In Stable Diffusion testing, the RTX 4090 outperforms the 7900 XTX for 512×512 and 768×768 pictures.

Many other AI workloads only operate on Nvidia GPUs. Nvidia understands content production apps. Its sole drawback is that drivers lock improved app performance (like SPECviewperf) to its actual professional cards, the RTX 6000 Ada Generation.

RTX 4090 is the current winner for raw performance, but AMD’s RDNA 3 reaction to Ada Lovelace may be better for rasterization games. To maximize the 4090, you may require a CPU and power supply update.

(Image credit: ign)

5. AMD Radeon RX 7900 XTX

AMD’s fastest GPU, great for rasterization

SPECIFICATIONS

GPU: Navi 31

GPU Cores: 6144

Boost Clock: 2500 MHz

Video RAM: 24GB GDDR6 20 Gbps

TBP: 355 watts

TODAY’S BEST DEALS

AMD’s fastest graphics card, the Radeon RX 7900 XTX, is near the top of the charts and has a generational price hike. Officially priced at $999, the cheapest units now cost roughly $900, and supply has caught up to demand. Demand is high since the 7900 XTX has AMD’s RDNA 3 architecture.

The 7900 XTX has 33% more memory, bandwidth, and compute than the previous version. The new GPU outperforms the RX 6950 XT by 44% at 4K, 34% at 1440p, and 27% at 1080p. It boosts performance without increasing power or graphics card size.

AMD is still a powerful answer for non-ray tracers, and when you see the significant performance hit for frequently rather moderate image fidelity enhancements, we can see why. RT games are growing, and most support Nvidia’s DLSS technology, which AMD hasn’t fully addressed despite FSR2/FSR3’s best efforts. For the finest DXR/RT experience right now, Nvidia wins.

AMD GPUs can be utilized for professional creative content creation, however results vary. AMD hardware works well for some SPECviewperf programs but not others. Nvidia cards are undoubtedly better for AI or deep learning research. AMD’s fastest GPU this generation is the RX 7900 XTX. As a cheaper alternative, the 7900 XT is worth investigating (see below).

How we test the best graphics cards

To determine pure graphics card performance, eliminate all other bottlenecks as much as feasible. Our 2024 graphics card testbed has a Core i9-13900K CPU, MSI Z790 MEG Ace DDR5 motherboard, 32GB G.Skill DDR5-6600 CL34 memory, Sabrent Rocket 4 Plus-G 4TB SSD, be silent! 80 Plus Titanium PSU, and cooling Master CPU cooling.

We test at 1080p and 4K with’medium’ and ‘extreme’ settings. We use’reference’ cards like Nvidia’s Founders Edition and AMD’s reference designs for all these tests. However, most midrange and lower GPUs lack reference models, and we sometimes test with factory overclocked cards. In such circumstances, we try to choose cards near to reference specs.

We test each graphics card the same way. We run one pass of each benchmark to “warm up” the GPU after opening the game, then at least two passes at each setting/resolution. If the two runs are similar (within 0.5%), we use the faster one. We repeat the test at least twice to determine “normal” performance if there’s a significant discrepancy.

We also analyze all the data for abnormalities, so RTX 3070 Ti, 3070, and 3060 Ti tend to perform within a small range—3070 Ti is roughly 5% faster than 3070, which is 5% faster than 3060 Ti. If games have evident outliers (i.e., performance is more than 10% higher for the cards indicated), we’ll retest the cards to determine the “correct” outcome.

Because GPUs take so long to test, updated drivers and game updates might affect performance. Retesting a few sample cards occasionally verifies our results, and if not, we retest the affected game(s) and GPU(s). Over the next year, we may add popular and testable games to our test suite. See our selection criteria for what constitutes a good game benchmark.

Best graphics cards performance results

We now have 19 games in our test suite, including Assassin’s Creed Mirage, Avatar: Frontiers of Pandora, Diablo IV, and The Last of Us, Part 1. Avatar and Diablo have DXR, and AC Mirage, Avatar, and The Last of Us are AMD-promoted. To get a new perspective, we enabled Quality mode upscaling in Avatar and Diablo (performance scaled identically on all GPUs, although FSR2 image quality is poorer than XeSS or DLSS).

The following charts show testing data from recent months. The fastest cards are tested at 1440p and 4K, however we try to test everything at 1080p medium and ultra. The first chart shows the geometric mean (equal weighting) for all 19 games for each resolution and setting. The second chart compares 11 rasterization games, and the third chart compares eight ray tracing games. For those that want all the info, there are 19 game charts.

Three games—Borderlands 3, Far Cry 6, and Total War: Warhammer 3—lack DLSS, FSR 2/3, or XeSS. AMD’s FSR 2.0 has been released for over a year, Nvidia’s DLSS 2 launched in mid-2019, and Intel’s XeSS launched in October 2023. 16 games in our test suite support DLSS 2, six support DLSS 3, ten support FSR 2, and five support XeSS.

Most of these benchmarks were conducted at native resolution, save for Avatar and Diablo, which we used Quality mode for. Our separate paper on FSR and DLSS concludes that DLSS and XeSS improve performance without compromising image quality, but FSR2/3 function on any GPU.

The charts below list all current Nvidia RTX 40-, AMD RX 7000-, and Intel Arc A-series graphics cards. Interested parties can find more GPU benchmark results in our hierarchy. AMD is red, Nvidia blue, and Intel gray to make the charts simpler to read.