Best graphics cards in 2025: I've tested pretty much every AMD and Nvidia GPU of the past 20 years and these are today's top cards

Nvidia RTX 5090 and XFX RX 9070 graphics cards
(Image credit: Nvidia | XFX)

I don't remember a time when picking the best graphics card was such a tough decision. In previous years we've maybe made it easier on ourselves by going on essentially a sliding performance scale, with a nod to price banding, but this year I want to be a bit more definitive about the actual GPUs we recommend rather than just giving you a big ol' list of graphics cards to pick from.

But I've tested and gamed on pretty much every new graphics card that's been released in this generation, from the Intel Battlemage cards that launched late last year, through the initial Nvidia RTX Blackwell launch, and on to AMD's RDNA 4 cards. I believe that makes me well placed to talk you through what the best graphics card is of this generation, and it's the AMD Radeon RX 9070.

For most PC gamers this is the graphics card which offers the most performance for the money, delivering decent 1440p performance, and much improved ray tracing over previous AMD generations of GPU. It's also rocking 16 GB of VRAM to be able to deliver on high graphics settings at that resolution now and into the future, too. It was priced appallingly at launch, but now it's settling down and delivering performance close to the RX 9070 XT and therefore the RTX 5070 Ti.

The absolute top, from a pure performance standpoint, however, is the Nvidia RTX 5090. That's the best high-end graphics card with no other contenders getting near its mix of style, features, and raw computational grunt. It's also a card where no-one gets close in terms of price, either, especially with prices the way they are right now.

High. That's what prices are right now, with quoted MSRPs being pure fiction while supply and demand seem so unevenly balanced, and then there's the considerable flux of tariff related price hikes, too. Essentially it's tough to be after a new graphics card in 2025.

Quick list

Curated by...
Dave James
Curated by...
Dave James

Dave first started as a PC hardware journalist, after years of freelance games journalism, way back in 2005 when he started with PC Format magazine. In that role, and many others since, he's tested pretty much every single graphics card generation and architecture released in the intervening years. He's the person best placed around PC Gamer to tell you which graphics card is the best right now, because he's used them all.

Recent updates

July 5, 2025: I've added in a full rundown of exactly how we test each and every graphics card that crosses our desks. And there's a lot of work that goes into benchmarking and reviewing a new GPU, and we're now presenting all the data we collect here, as well as in all the reviews that we do.

July 4, 2025:
Complete rewrite of the best graphics card, with an entirely new list and categorisation. I have also included the specs of all the latest generations of AMD and Nvidia graphics cards, as well as a GPU heirarchy listing for the latest graphics cards, too.

January 30, 2025:
Added in the new RTX Blackwell cards now that they have been released... and immediately disappeared from stock. Though surely restocks will be coming along soon for the RTX 5090 and RTX 5080 GPUs.

1. Best overall graphics card: AMD Radeon RX 9070

The best graphics card of 2025

Specifications

Shaders: 3584
Boost clock: 2520 MHz
TFLOPs: 36.12
Memory: 16 GB GDDR6
Memory clock: 20 GT/s
Memory bandwidth: 640 GB/s
TGP: 220 W

Reasons to buy

+
Not far off the performance of the XT version
+
Comfortably beats RTX 5070
+
Especially when overclocked
+
Healthy VRAM capacity

Reasons to avoid

-
The $100 price bump is an issue
-
The XT version will become available again at MSRP
Buy if...

You want a good 1440p GPU, and a great 1080p one: With the ray tracing updates baked into RDNA 4, the RX 9070 is now a fantastic all-round graphics card. It can nail 1440p and 1080p resolutions at top settings, and can call on FSR and AMD's frame generation feature to give you a boost if you need it.

You're happy to undervolt/overclock: With some easy tweaks in the AMD drivers you can boost the performance of the GPU to within a couple of percentage points of the RX 9070 and RTX 5070 Ti. And the best part is that it doesn't demand a ton more power or cooling to get there, so shouldn't stress you or your fancy new card.

Don't buy if...

You can find an RX 9070 XT for anything close to the same price: While you can overclock the non-XT card to get within 2% of the performance, prices being equal the XT card is a no-brainer.

The bottom line

🪛 The Radeon RX 9070 is an excellent graphics card if you're looking for high 1440p performance without the almost $1,000 price points of the competing Nvidia cards. With the market as it is, the AMD card is the best GPU for most PC gamers.

The RX 9070 is a rather agonising pick as the best graphics card around right now. There are inevitable caveats around my decision, because quite obviously this second-tier RDNA 4 GPU is not the most powerful GPU you can buy. The AMD Radeon RX 9070, however, is the graphics card I think makes the most sense to the most PC gamers.

And, realistically, it's going to be the one that I would suggest to my PC gaming friends if they come looking for a recommendation. Luckily, I don't actually have friends.

The RX 9070 offers close to the performance of AMD's most powerful RDNA 4 GPU, still packs in 16 GB of GDDR6 video memory, can be easily overclocked without demanding much more cooling or power, and handily outperforms Nvidia's similarly priced, lower-spec RTX 5070.

I was relatively cool on the card in my initial review of the GPU, mainly because our first taste of it was a card that XFX hiked the price of out the gate, which made it more than a reference RX 9070 XT. As things have evened out, there are reliably cheaper RX 9070 cards around and when you look at the graphics card market as a whole, the Radeon stands out as a great pick.

Yet I still haven't come to that conclusion lightly. My first thought was going to be recommending the $370 RX 9060 XT, because it is great value, performs well, and pretty much offers performance parity with the more expensive Nvidia option. But it doesn't feel like a great recommendation for the best graphics card, given that it is a third-tier Radeon that isn't particularly exciting.

The Radeon RX 9070 XT, though, is a fantastic card and were it not for the ludicrous over-pricing of AMD's finest it would absolutely be my top pick. With an MSRP of $599 it would be an easy choice for its impressive performance out of the box, stunning under-volting frame rate bumps, and the excellent filip of the new FSR4 upscaler. But the cheapest you can get it for these days is still near $750, with the RX 9070 generally around $100 cheaper with very similar gaming performance.

Though the RX 9070 is still $100 more than AMD's MSRP for the GPU, which does absolutely sting. These are the graphics card hell times, however, so that is almost to be expected. In the UK it is more reasonable, with a £540 price tag trending down towards its MSRP, and that is welcome news for PC gamers and hopefully gives a glimpse as to what price normalisation might start to look like for folk in the US down the line.

Though it is absolutely worth noting that Nvidia's competing, and competent, RTX 5070 is generally cheaper and closer to its own MSRP.

That card is generally behind the curve when it comes to gaming performance; we've found it on average around 7% slower in straight raster frame rates. It too is a great overclocking card, but when both the RTX 5070 and RX 9070 are pushed to their limits, the AMD card actually stretches its lead to around 10% faster than the GeForce GPU.

In pure performance terms, then, it's the AMD RX 9070 all the way. But there is one trick the Radeon cannot match, and that's the RTX 5070's Multi Frame Generation (MFG) feature. Being able to drop in up to three extra AI generated frames in between each actually rendered one can give the Nvidia card far more impressive performance in supported games, though it can also introduce graphical artifacts where the initial gaming performance is too low.

Basically, the lower you go down the GPU stack, the less effective MFG is. Though, I will say that with AMD's new machine learning-based FSR 4 upscaler being in so few of the games that you will be playing, Nvidia's DLSS 4 and Frame Gen feature package remains a very compelling one.

In the end though, the extra raster performance of the Radeon RX 9070—and how close it can get to the more expensive RTX 5070 Ti and RX 9070 XT GPUs—makes it the best graphics card to buy right now for most PC gamers.

Read our full AMD Radeon RX 9070 review.

2. Best value graphics card: AMD Radeon RX 9060 XT 16 GB

The best value graphics card of this generation

Specifications

Shaders: 2048
Boost clock: 3130 MHz
TFLOPs: 25.64
Memory: 16 GB GDDR6
Memory clock: 20 GT/s
Memory bandwidth: 320 GB/s
TGP: 160 W

Reasons to buy

+
Relatively cheap
+
Gives the RTX 5060 a thorough pasting
+
Capable of trading blows with the RTX 5060 Ti
+
Runs cool and power efficient

Reasons to avoid

-
RTX 5060 Ti slightly faster overall
-
Limited overclocking potential
-
All dependent on price
Buy if...

You want a lot of VRAM without spending a ton of cash: The 16 GB version of the RX 9060 XT has an MSRP of $349, and even in these price-inflated times it's regularly available right now for well under $400.

You want an affordable upgrade for modern 1080p or 1440p gaming: Smooth 4K gaming is beyond even the fastest budget cards, but at lower resolutions the RX 9060 XT delivers great performance for the cash, particularly compared to previous generations.

You want bang for your buck: The RX 9060 XT might not be the fastest card on the market, but nothing touches it at this price.

Don't buy if...

You can really stretch that budget: For under $400 there is nothing that touches the RX 9060 XT, but if you can push on to ~$450 that brings you into RTX 5060 Ti 16 GB territory. That is an overall faster GPU with more overclocking potential.

You want productivity performance: This is a gaming card, through and through, so if you're looking for productivity chops then Nvidia is the way to go.

The bottom line

🪛 The AMD Radeon RX 9060 XT is not a sexy graphics card, it's not the fastest graphics card, but it's an honest GPU that will deliver good gaming performance at the most popular resolutions with top settings. And all for under $400.

To be able to pick up a genuine 16 GB graphics card, with a seriously performant GPU core at its heart, for under $400 is pretty great. It would be even better if the card in question was actually retailing for its original MSRP, but even at the current slightly over MSRP pricing, the Radeon RX 9060 XT is still our pick for the best value graphics card today.

Okay, the latest AMD RDNA 4 GPU is not quite living up to its launch claims of beating the RTX 5060 Ti, certainly not across our benchmarking suite, but at 1080p the Nvidia card is less than a percentage point ahead on average, and it's by only 3% at 1440p. Effectively, I think we can kinda call that parity.

The GeForce card is more consistently ahead at 4K—even though they both carry 16 GB on the top versions of the card—but neither are capable of actually playable frame rates at that resolution. And even the boon of Multi Frame Generation can't help in reality.

But the parity at 1440p and 1080p itself is impressive for the AMD card, especially when you take into account that we're testing games which have pretty hefty ray tracing workloads in them as part of our benchmarking gauntlet. Historically that's where AMD cards have fallen down against Nvidia, but the new RDNA 4 architecture has changed around how the Radeon cards deal with ray tracing, giving them more dedicated silicon to do the work, and that has made all the difference.

Another area where there has been historic Nvidia dominance is in the feature set, and honestly, the green team still has the edge on that count. The twin pillars of DLSS 4 upscaling and the Multi Frame Generation (MFG) feature do give the GeForce card some real merit. But the quality experience of MFG is very dependent on there being a relatively high, consistent frame rate before the interpolation of up to three extra frames kicks in. That means its effectiveness does diminish lower down the GPU stack.

AMD has introduced its own new upscaler in this generation, however, with FSR 4. That's a new take on the GPU feature, more closely resembling Nvidia's DLSS as it now uses machine learning to improve the visual fidelity of the upscaled image. It's only in a few of the games we're playing at the moment, but hopefully it will start to become as ubiquitous as standard FSR and DLSS going forward. It's certainly an impressive piece of tech, and when combined with AMD's own frame gen tech it does give the RDNA 4 architecture some effective weapons in its arsenal, closing that gap with Nvidia a whole bunch in this generation.

But don't expect much in the way of overclocking with this card. A surprising feature of all the latest GPUs from Nvidia and AMD has been their proclivity for overclocking or undervolting. Not so here. Our testing has shown that AMD's engineers have obviously pushed the Navi 44 silicon as hard as it can in order for it to get as close to the RTX 5060 Ti as possible. That means there's no real headroom there for the end user to play with.

The RTX 5060 Ti, however, does have a little extra in the tank, so if you are happy to overclock, we've found you can consistently create a performance delta of around 10% between the two cards in the Nvidia GPU's favour.

It's worth noting here that it's the 16 GB version of this card that we're recommending here (though if you simply cannot go north of $300, the 8 GB version wouldn't be a bad shout). Both AMD and Nvidia have released twin versions of their cards with 16 and 8 GB VRAM allocations and, given the memory-intensive direction detail-heavy modern games are going, that extra memory headroom will be valuable even if you're gaming at 1440p.

So, our recommendation would be to aim for that 16 GB version if you have the patience to save up a little longer, as it will help in the long run. Then you'll be rocking a real solid, pretty low power GPU, that can stand toe-to-toe with the more expensive Nvidia cards.

Read our full Radeon RX 9060 XT review.

3. Best budget graphics card: Intel Arc B570

3. Intel Arc B570

The best budget graphics card

Specifications

Shaders: 2304
Boost clock: 2500 MHz
TFLOPs: 11.52
Memory: 10 GB GDDR6
Memory clock: 19 GT/s
Memory bandwidth: 380 GB/s
TGP: 150 W

Reasons to buy

+
Super cheap at its MSRP
+
More reliable than Intel Alchemist

Reasons to avoid

-
Hard to find at MSRP in the US
-
There will be new games where it has bugs
Buy if...

You are on a restrictive budget: Around the Arc B570's MSRP there is nothing that can really compete right now. If you can spend a bit more you will get a lot more performance out of a $300 RTX 5060 8 GB, but that's only an option if you the spare cash.

You're not afraid of finding a fix: As a relatively new player in discrete GPUs, Intel's drivers will turn up issues occasionally in newly released games, but if you can wait for a driver fix, or are willing to find a workaround they can be ironed out.

Don't buy if...

You can find an RTX 5050 at a similar price: The cheapest RTX Blackwell GPU is currently suffering stock issues, but has a launch price that is close to the Arc B570 and should beat it by around 10% on average in games.

You don't ever want to deal with driver issues: You will likely have some issue with the B570 at some point, in some new game, and if that's going to be a deal-breaker for you then maybe steer clear. Though it is worth noting that Nvidia's drivers have shown a proclivity for flakiness in recent times... Intet is not alone in that.

The bottom line

🪛 The Intel Arc B570 is a 'right now' recommendation for the best budget graphics card. If Nvidia's RTX 5050 hits its ~$250 price point then that becomes a far more compelling option than Intel's budget card unless that can drop below the $200 mark. But right now it is the best option for those on a limited budget looking for good 1080p gaming performance and a decent level of VRAM.

Difficult GPU recommendations is kinda the theme for this graphics card tier list, and they don't come much tougher than the budget pick. In a world where GPU pricing has been trending ever higher since the pandemic, and even prior to that with the ethereum mining booms, it's hard going trying to find a really cheap graphics card that's still worth the money in 2025.

Now, bear with me, because you might be reasonably surprised at my recommendation of the Intel Arc B570 as the best of anything in the GPU market. But Intel's drivers have improved measurably since the first Arc Alchemist cards arrived, and while you will find some new games that don't play nice at launch, and there remain some compatibility issues with AMD's X3D chips, when it works properly it's the best sub-$300 card around.

At least until we're sure how the new RTX 5050 performs, but Nvidia has been holding back reviews of its 8 GB cards in this generation. For reasons. But, according to Nvidia's own numbers it's essentially an RTX 4060 level card with MFG, and if you can find that for $250 MSRP that's not a bad shout for best budget GPU as that would put it around 16% quicker than the Arc card on average; sometimes much quicker. It's just launched, though, so the nominally $250 card is out of stock everywhere right now anyways.

Which leaves the B570 as the only game in town. In the US it too is a ~$250 card—a bit higher than its $220 MSRP—but in the UK it's available for below £200. That does put it in proper budget GPU territory and if you take inflation into account it's around the same sort of level as the classic GTX 1050 Ti.

For the most part, that's going to get you around 60 fps at the very highest 1080p graphics settings, and that's native performance before you start to bring in either upscaling or frame generation. When you do that then the Arc B570 is actually capable of delivering that 60+ fps performance at 1440p, too, and obviously even higher at 1080p.

Part of that is because this isn't just another 8 GB graphics card, Intel has stuck 10 GB of GDDR6 on this budget card, and given it more memory bandwidth than the RX 9060 XT, too.

If pricing really is an issue, and you want a card that will deliver decent frame rates without breaking the bank, then Intel's Arc B570 is worth a look. The caveat, however, is that while drivers are better you will find some kinks still to be worked out in Intel's drivers (my B570 specifically won't work in Cyberpunk 2077 at 1080p with the Ryzen 7 9800X3D, but is fine at other resolutions and with other games) which means you have to be willing to accept that.

If you just want a budget card that will just work, then the RTX 5050 might well be a better card for you. Though, I will say that Nvidia's own drivers, which have previously been rock-solid, have been frustratingly unreliable with the RTX Blackwell generation of GPUs. When it comes to budget graphics cards, there is always going to be some compromise, I'm afraid.

4. Best mid-range graphics card: Nvidia RTX 5070 Ti

The best mid-range graphics card

Specifications

Shaders: 8960
Boost clock: 2452 MHz
TFLOPs: 44.35
Memory: 16 GB GDDR7
Memory clock: 28 GT/s
Memory bandwidth: 896 GB/s
TGP: 300 W

Reasons to buy

+
Multi Frame Generation still slaps
+
Overclocks like a beast
+
Close to RTX 5080 performance
+
Theoretically cheaper than RTX 4070 Ti/Super

Reasons to avoid

-
In reality the 5070 Ti is likely to cost the same as the RTX 5080 MSRP
-
You can knock 20% off the score if this one ends up a $900+ GPU
Buy if...

You can find it for close to MSRP: I get it, prices have been buck-wild recently, but they're coming down across most of the globe, and at its more reasonable price the RTX 5070 Ti is a fantastic graphics card.

You are happy to overclock: Such is the level of GPU headroom in the GB203 that overclocking the RTX 5070 Ti is really a no-brainer. Even more so because it doesn't unduly stress the graphics silicon, is easy to do, and can genuinely get you a bunch more frames per second for little effort.

You're not ideologically opposed to Multi Frame Generation: The RTX 5070 Ti's native performance comes in at a level where MFG really comes into its own, delivering super high, super smooth frame rates where it's supported. Which is a lot of games, right now. But if you're against it on principal no matter how good it looks then the RX 9070 XT is probably the GPU for you.

Don't buy if...

The price delta between it and the RX 9070 XT is over $100: The RTX 5070 Ti and RX 9070 XT cards go head-to-head in the performance stakes, and Nvidia's cards just about has the edge when it comes to feature set and overclocking, but that's only if prices are relatively close. When there's a big gap between them then the cheaper AMD card makes for the best mid-range graphics card.

The bottom line

🪛 The Nvidia RTX 5070 Ti is the best mid-range graphics card you can buy if you can buy it for a reasonable price. It's an outstanding 1440p card, and with all the gaming assists on, such as upscaling and MFG, it's an impressive 4K GPU, too. But prices have been far too close to $900 in recent times, and that makes it a harder recommendation. Closer to its $750 MSRP, however, and it's a great graphics card.

If you were wondering where all the Nvidia cards were in this list you're about to hit the motherlode, because at the higher-end of the market the GeForce GPUs are the ones you really ought to covet. The RTX Blackwell architecture may not be a huge silicon improvement over its forebear, but there is enough in the mix of DLSS 4, Multi Frame Generation (MFG), and the slight GPU redesign to give you a serious boost in performance.

MFG does its best work when there is a decently high frame rate going in, which means you can smooth out performance, delivering very high frame rates without massively spiking the PC latency. That's the best case scenario for MFG, and with the RTX 5070 Ti as the best mid-range graphics card you're getting the Nvidia feature set working to its fullest potential.

There is still a case to be made for the cheaper AMD Radeon RX 9070 XT, because with subsequent driver updates its performance has improved to the point where at native resolutions it will sometimes outpace the Nvidia competition. Now prices are starting to settle down on the AMD side it's getting far more affordable, while in the US especially, the RTX 5070 Ti still commands a high price tag.

Things are more even in the UK and the rest of the world, which does make it harder to pick the AMD over the Nvidia GPU in terms of performance and gaming experience. Where prices are closer together I'm going to side with the GeForce card thanks to its more effective feature set, the fact that MFG does genuinely work at this level, and that it is a very capable overclocker.

To be fair, both AMD and Nvidia cards are at this level, but the RTX Blackwell has so much easily accessible GPU headroom that overclocking is not only simple, but bears little to no actual risk in running your card at a higher level permanently. It's not really any hotter, or more demanding of power, so why wouldn't you? I've personally tested four different RTX 5070 Ti cards and every single one of them could be pushed to around the same level of 3+ GHz clock speeds without even touching 70 °C.

With all the assists on, the RTX 5070 Ti will consistently deliver triple-figure frame rates in all the latest games at 1440p, and given its regular 60+ fps natively at that resolution, you'll also find it a more than capable 4K graphics card with upscaling and MFG thrown in for good measure.

And when prices normalise across the board as time marches inexorably along, then we'll get back towards the MSRP of the RTX 5070 Ti and that should cement its place as the best mid-range graphics card in every territory.

Read our full Nvidia RTX 5070 Ti review.

5. Best high-end graphics card: Nvidia GeForce RTX 5090

The best graphics card

Specifications

Shaders: 21760
Boost clock: 2,410 MHz
TFLOPs: 104.8
Memory: 32 GB GDDR7
Memory clock: 28 GT/s
Memory bandwidth: 1,792 GB/s
TGP: 575 W

Reasons to buy

+
Stunning AI augmented performance
+
Decent gen-on-gen frame rates
+
FE card looks great
+
Huge potential for the future

Reasons to avoid

-
Lots of coil whine with one PSU
-
$400 price jump on RTX 4090
-
Transformer model feels very v1.0
Buy if...

You want the best: If you want to nail triple figure frame rates in the latest 4K games, then you're going to need the might and magic of Multi Frame Gen, and that's only available with the RTX 50-series cards. And yes, I do like alliteration.

You to get in on the ground floor of neural rendering: The RTX Blackwell GPUs are the first chips to come with a full set of shaders that will have direct access to the Tensor Cores of the card. That will enable a new world of AI-powered gaming features... when devs get around to using them in released games.

You're after a hyper-powerful SFF rig: The Founders Edition is deliciously slimline, and while it generates a lot of heat it will fit in some of the smallest small form factor PC chassis around.

Don't buy if...

You need to ask the price: With a $400 price hike over the RTX 4090, the new RTX 5090 is a whole lot of cash at its $1,999 MSRP. The kicker, however, is that you'll be lucky to find one at that price given the third-party cards are looking like $2,500+ right now.

The bottom line

🪛 The RTX 5090 is the most powerful consumer graphics card on the planet right now, and delivers gaming performance far beyond what you could manage on other GPUs, especially if you're playing something which supports Multi Frame Generation.

The RTX Blackwell generation of new GPUs has kicked off with a bang, and means that right now, the best graphics card is undoubtedly the Nvidia RTX 5090. And is likely to remain that way for the next two years at least. Given the fact that AMD isn't going to release a competitor card for the top GeForce GPU in the RDNA 4 generation, you can be confident if you pick one of these up today (or when they come back in stock) you will still be gaming on the best GPU probably until the next Nvidia generation is released.

While that might make for miserable reading for AMD fans, it should be a little more comforting for anyone hellbent on spending $2,000+ on a new graphics card. It will, at least, last the course at the top of benchmarking tree.

Sure, you are only getting some 30% extra gen-on-gen performance over the RTX 4090 at 4K, and that is the smallest performance bump over a previous generation's top card since Turing came along. But there's a magic trick up the sleeve of the RTX Blackwell cards, and that is Multi Frame Generation.

Right now, it's limited to the RTX 50-series—else it would likely cannibalise sales to a huge extent down the stack if RTX 40-series GPUs were allowed into the MFG party—and it adds in up to three more frames in between each actually rendered pair. Using a feature called Flip Metering, which utilises an enhanced bit of silicon in the RTX 50-series Display Engine, it offloads all the burden of frame pacing from the CPU, puts it all on the GPU, and allows the RTX 5090 to queue up all these extra AI-generated frames perfectly for the display.

Along with a new AI-based frame generation model, Multi Frame Generation is able to hugely increase the potential performance of the RTX 5090 in any game which supports it, and the results are frankly incredible. There are some small artifacts—though nothing that would stop me using the feature—but the really impressive thing is that it adds practically no extra latency on top of the standard 2x Frame Generation experience.

And that itself has been lowered thanks to that new FG AI model which makes it 40% quicker and 30% less VRAM hungry at the same time. The good news for RTX 40-series patrons is that model at least is coming to standard Frame Gen in the Ada generation, too.

There are also a ton of games at launch with immediate compatibility with MFG, either with it natively implemented in the game, à la Cyberpunk 2077, or via the new DLSS Override feature in the Nvidia App, as in Dragon Age: The Veilguard.

It's also the most powerful consumer GPU when it comes to creator tasks, too, thanks to its hefty 32 GB of GDDR7 and its massive bandwidth, but also because it's so damned good with an AI noodling if GenAI is your thing.

In short, the RTX 5090 is the best graphics card for anyone who wants the absolute finest silicon and feature set of any consumer GPU going. You're just going to have to pay through the nose for it until stock settles down. If it ever does to an extent that MSRP versions of the card become readily available.

Read our full Nvidia GeForce RTX 5090 review.

How we test graphics cards

Various graphics cards set atop a white plinth on a pink gradient background

(Image credit: Future)

I have been benchmarking graphics cards since the 2000s, and have used many different games, applications, and methodologies over the intervening years. And, for the most part, testing GPUs is a largely objective process; you test the silicon in different ways, come out with a set of benchmark figures which you then compare against another set of numbers from a different card, or cards, to be able to objectively tell which is empirically the best.

Our current GPU test suite consists of Black Myth Wukong, Cyberpunk 2077, F1 24, Homeworld 3, Metro Exodus Enhanced Edition, The Talos Principle 2, andTotal War: Warhammer III for our gaming tests. These tests are carried out using the same game and system settings across all the graphics cards we test, and are run at 1080p, 1440p, and 4K resolutions.

We measure using both the average frame rate and the 1% Low FPS figure. This gives us a general measure of in-game performance as well as allowing us to see just how consistent that frame rate level is. The 1% Low FPS measure shows the average of the highest 1% of frame times in any given benchmarking run. Translating that into frames per second (1000/x ms) gives us the data to show whether there are regularly large drops in performance or whether it's relatively stable.

We capture this data using the Nvidia Frameview app running over the top of whatever game we are benchmarking, whether the game will give its own data output or not.

We also use the UL suite of benchmark software to get some synthetic testing done against high-end rastererisation performance with 3DMark Time Spy Extreme, and ray tracing performance with 3DMark Port Royal.

Nvidia RTX 5080 Founders Edition graphics card from different angles

(Image credit: Future)

With the new RTX Blackwell cards, we've also been using a selection of games to help us get a bead on the impact of Multi Frame Generation on both frame rate and PC latency, too. As well as retesting different levels of MFG (at either 1440p for lower end cards or 4K for high-end GPUs) with Cyberpunk 2077, we also use the ultra-demanding Alan Wake 2 and Dragon Age: Veilguard for its implementation via the Nvidia App.

Actual gaming performance isn't the whole story with graphics cards, however, as they are used for different uses outside of gaming. While we are PC Gamer, we know that some people want to be able to use their PC for 3D rendering, editing, or generative AI uses, and so we run the Blender benchmark, PugetBench for DaVinci Resolve, and Procyon's image generation benchmark using StableDiffusion.

We also capture a ton of system data, too, using the Nvidia PCAT tool (a riser board which sits between the GPU and PCIe slot) to measure actual graphics card power draw. This means we can track both peak and average power use when gaming, and a given GPU's performance per Watt metrics, too.

For this we use three back-to-back runs of the Metro Exodus Enhanced Edition benchmark, at 1080p and 4K, to give us the power numbers, as well as peak and average temperatures and average GPU clock speed, too.

On top of this we will also test the overclocking capabilities of graphics cards, by pushing them as far as we can using standard overclocking methodologies; ie. the same as you would be able to easily do at home. No LN2 sniffing going on here.

But there are also subjective measures which come into play when actually picking the best graphics card. And that all comes down to a consistent driver experience when using the GPU, how loud the cooling fans can get, whether there is discomforting coil whine or other electrical noise, and just how much damned money manufacturers are charging for these cards.

In short, there's a lot that goes into our testing.

In the PC Gamer office—and sometimes my own satellite office up the hill if I'm testing into the wee hours of the morning—we have a dedicated test rig that we use for testing graphics cards. This is our AMD Ryzen 7 9800X3D-based system:

CPU: AMD Ryzen 7 9800X3D | Motherboard: Gigabyte X870E Aorus Master | RAM: G.Skill 32 GB DDR5-6000 CAS 30 | Cooler: Corsair H170i Elite Capellix | SSD: 2 TB Crucial T700 | PSU: Seasonic Prime TX 1600W | Case: DimasTech Mini V2

We also have two other systems, kindly provided by MSI and CyberpowerPC, which we use if another member of the team, such as Jacob, Nick, or Andy, need to benchmark a card. These travel around the UK, and both house the same set of components so we can maintain multiple testing PCs which deliver data we can use to accurate reference other cards.

Where are the best graphics card deals?



GPU hierarchy

Below we have listed multiple generations of graphics card based on a simple 3DMark Time Spy Extreme GPU index score. This is only a rough approximation of relative gaming performance between the differnt graphics cards, as there is more variance than just with a single synthetic benchmark, but it is still a good snap shot of where the cards stack up against each other.

Nvidia GPU specs

Swipe to scroll horizontally
Nvidia RTX 50-series (RTX Blackwell)

Model

MSRP (US$)

Launch

GPU

Lithography

CUDA cores

Memory size (GB)

Die size (mm²)

Transistors (B)

SM count

TMUs

ROPs

Tensor cores

RT cores

L2 cache (MB)

Boost clock (MHz)

Memory type

Memory bus width (bits)

Memory bandwidth (GB/s)

TDP (W)

RTX 5050

$249

Jul 16, 2025

GB207-300

TSMC 4N

2560

8

121

15.1

20

80

32

80

20

32

2572

GDDR6

128

320

130

RTX 5060

$299

May, 19, 2025

GB206-250

TSMC 4N

3840

8

181

21.9

30

120

48

120

30

32

2497

GDDR7

128

448

145

RTX 5060 Ti

$379 / $429

Apr, 16, 2025

GB205-300

TSMC 4N

4608

8 / 16

181

21.9

36

144

48

144

36

32

2572

GDDR7

128

448

180

RTX 5070

$549

Mar 5, 2025

GB205-300

TSMC 4N

6144

12

263

31.1

48

192

80

192

48

48

2512

GDDR7

192

672

250

RTX 5070 Ti

$749

Feb 20, 2025

GB203-300

TSMC 4N

8960

16

378

45.6

70

280

96

280

70

64

2452

GDDR7

256

896

300

RTX 5080

$999

Feb 20, 2025

GB203-400

TSMC 4N

10752

16

378

45.6

84

336

112

336

84

64

2617

GDDR7

256

960

360

RTX 5090

$1999

Jan 30, 2025

GB202-300

TSMC 4N

21760

32

750

92.2

170

680

176

680

170

96

2407

GDDR7

512

1792

575

Swipe to scroll horizontally
Nvidia RTX 40-series (Ada Lovelace)

Model

MSRP (US$)

Launch

GPU

Lithography

CUDA cores

Memory size (GB)

Die size (mm²)

Transistors (B)

SM count

TMUs

ROPs

Tensor cores

RT cores

L2 cache (MB)

Boost clock (MHz)

Memory type

Memory bus width (bits)

Memory bandwidth (GB/s)

TDP (W)

RTX 4060

$299

Jun 29, 2023

AD107-400

TSMC 4N

3072

8

159

18.9

24

96

32

96

24

24

2460

GDDR6

128

272

115

RTX 4060 Ti

$399

May 24, 2023

AD106-350

TSMC 4N

4352

8

188

22.9

34

136

48

136

34

32

2535

GDDR6

128

288

160

RTX 4060 Ti 16GB

$499

Jul 18, 2023

AD106-351

TSMC 4N

4352

16

188

22.9

34

136

48

136

34

32

2535

GDDR6

128

288

165

RTX 4070

$599

Apr 13, 2023

AD104-250

TSMC 4N

5888

12

294.5

35.8

46

184

64

184

46

36

2475

GDDR6X

192

504

200

RTX 4070 Super

$599

Jan 17, 2024

AD104-350

TSMC 4N

7168

12

294.5

35.8

56

224

80

224

56

48

2475

GDDR6X

192

504

220

RTX 4070 Ti

$799

Jan 5, 2023

AD104-400

TSMC 4N

7680

12

294.5

35.8

60

240

80

240

60

48

2610

GDDR6X

192

504

285

RTX 4070 Ti Super

$799

Jan 24, 2024

AD103-275

TSMC 4N

8448

16

379

45.9

66

264

96

264

66

48

2610

GDDR6X

256

672

285

RTX 4080

$1199

Nov 16, 2022

AD103-300

TSMC 4N

9728

16

379

45.9

76

304

112

304

76

64

2505

GDDR6X

256

716.8

320

RTX 4080 Super

$999

Jan 31, 2024

AD103-400

TSMC 4N

10240

16

379

45.9

80

320

112

320

80

64

2550

GDDR6X

256

736

320

RTX 4090

$1599

Oct 12, 2022

AD102-300

TSMC 4N

16384

24

608.5

76.3

128

512

176

512

128

72

2520

GDDR6X

384

1008

450

Swipe to scroll horizontally
Nvidia RTX 30-series (Ampere)

Model

MSRP (US$)

Launch

GPU

Lithography

CUDA cores

Memory size (GB)

Die size (mm²)

Transistors (B)

SM count

TMUs

ROPs

Tensor cores

RT cores

L2 cache (MB)

Boost clock (MHz)

Memory type

Memory bus width (bits)

Memory bandwidth (GB/s)

TDP (W)

RTX 3050

$169 / $249

Jan 27, 2022

GA106-150

Samsung 8nm

2560

6 / 8

276

12

20

80

32

80

20

2

1777

GDDR6

128

168 / 224

130

RTX 3060

$329

Feb 25, 2021

GA106-300

Samsung 8nm

3584

8 / 12

276

12

28

112

48

112

28

3

1777

GDDR6

192

240 / 360

170

RTX 3060 Ti

$399

Dec 1, 2020

GA104-200

Samsung 8nm

4864

8

392

17.4

38

152

80

152

38

4

1665

GDDR6/X

256

448 / 608

200

RTX 3070

$499

Oct 29, 2020

GA104-300

Samsung 8nm

5888

8

392

17.4

46

184

96

184

46

4

1725

GDDR6

256

448

220

RTX 3070 Ti

$599

Jun 10, 2021

GA104-400

Samsung 8nm

6144

8

392

17.4

48

192

96

192

48

4

1770

GDDR6X

256

608.3

290

RTX 3080

$699

Sep 17, 2020

GA102-200

Samsung 8nm

8704

10

628

28.3

68

272

96

272

68

5

1710

GDDR6X

320

760.3

320

RTX 3080 12 GB

$799

Jan 11, 2022

GA102-220

Samsung 8nm

8960

12

628

28.3

70

280

96

280

70

5

1710

GDDR6X

384

912.4

350

RTX 3080 Ti

$1199

Jun 3, 2021

GA102-225

Samsung 8nm

10240

12

628

28.3

80

320

112

320

80

6

1665

GDDR6X

384

912.4

350

RTX 3090

$1499

Sep 24, 2020

GA102-300

Samsung 8nm

10496

24

628

28.3

82

328

112

328

82

6

1695

GDDR6X

384

936.2

350

RTX 3090 Ti

$1999

Mar 29, 2022

GA102-350

Samsung 8nm

10752

24

628

28.3

84

336

112

336

84

6

1860

GDDR6X

384

1008

450

Row 10 - Cell 0 Row 10 - Cell 1 Row 10 - Cell 2 Row 10 - Cell 3 Row 10 - Cell 4 Row 10 - Cell 5 Row 10 - Cell 6 Row 10 - Cell 7 Row 10 - Cell 8 Row 10 - Cell 9 Row 10 - Cell 10 Row 10 - Cell 11 Row 10 - Cell 12 Row 10 - Cell 13 Row 10 - Cell 14 Row 10 - Cell 15 Row 10 - Cell 16 Row 10 - Cell 17 Row 10 - Cell 18 Row 10 - Cell 19
Swipe to scroll horizontally
Nvidia RTX 20-series (Turing)

Model

MSRP (US$)

Launch

GPU

Lithography

CUDA cores

Memory size (GB)

Die size (mm²)

Transistors (B)

SM count

TMUs

ROPs

Tensor cores

RT cores

L2 cache (MB)

Boost clock (MHz)

Memory type

Memory bus width (bits)

Memory bandwidth (GB/s)

TDP (W)

RTX 2060

$299 / $349

Jan 15, 2019

TU106-200 / TU106-300 / TU104-150

TSMC 12 nm

1920 / 2176

6 / 12

445 / 545

10.8 / 13.6

30 / 34

120 / 136

48 / 64

240 / 272

30 / 34

3

1650 / 1680

GDDR6

192

336

160 / 185

RTX 2060 Super

$399

Jul 9, 2019

TU106-410

TSMC 12 nm

2176

8

445

10.8

34

136

64

272

34

4

1650

GDDR6

256

448

175

RTX 2070

$499

Oct 17, 2019

TU106-400

TSMC 12 nm

2304

8

445

10.8

36

144

64

288

36

4

1620

GDDR6

256

448

175

RTX 2070 Super

$499

Jul 9, 2019

TU104-410

TSMC 12 nm

2560

8

545

13.6

40

160

64

320

40

4

1770

GDDR6

256

448

215

RTX 2080

$699

Sep 20, 2018

TU104-400

TSMC 12 nm

2944

8

545

13.6

46

184

64

368

46

4

1710

GDDR6

256

448

215

RTX 2080 Super

$699

Jul 23, 20190

TU104-450

TSMC 12 nm

3072

8

545

13.6

48

192

64

384

48

4

1815

GDDR6

256

499.2

250

RTX 2080 Ti

$999

Sep 20, 2018

TU102-300

TSMC 12 nm

4352

11

754

18.6

68

272

88

544

68

5.5

1545

GDDR6

352

616

250

Titan RTX

$2499

Dec 18, 2018

TU102-400

TSMC 12 nm

4608

24

754

18.6

72

288

96

576

72

6

1770

GDDR6

384

672

280

AMD GPU specs

Swipe to scroll horizontally
AMD Radeon RX 7000-series (RDNA 4)

Model

MSRP (US$)

Launch

GPU

Compute Units (CUs)

Memory Size (GB)

Fab

Transistors (B)

Die Size (mm²)

Shaders

TMUs

ROPs

Ray Accelerators

AI Accelerators

Memory Type

Memory bus

TBP

RX 9060 XT

$269

May 25, 2023

Navi 44

32

8 / 16

TSMC N4P

29.7

199

2048

128

64

32

64

GDDR6

128

150 / 160

RX 9070 GRE

$449

Sep 6, 2023

Navi 48

48

12

TSMC N4P

53.9

356.5

3072

192

96

48

96

GDDR6

192

220

RX 9070

$499

Sep 6, 2023

Navi 48

56

16

TSMC N4P

53.9

356.5

3584

224

128

56

112

GDDR6

256

220

RX 9070 XT

$899

Dec 13, 2022

Navi 48

64

16

TSMC N4P

53.9

356.5

4096

256

128

64

128

GDDR6

256

304

Swipe to scroll horizontally
AMD Radeon RX 7000-series (RDNA 3)

Model

MSRP (US$)

Launch

Code Name

Compute Units (CUs)

Fab

Transistors (B)

Die Size (mm²)

Shaders

TMUs

ROPs

Ray Accelerators

AI Accelerators

Memory Size (GB)

Memory Type

Memory Bus

TBP

RX 7600

$269

May 25, 2023

Navi 33

32

6nm

13.3

204

2048

128

64

32

64

8

GDDR6

128

165

RX 7700 XT

$449

Sep 6, 2023

Navi 32

54

5nm

28.1

346

3456

216

96

54

108

12

GDDR6

192

245

RX 7800 XT

$499

Sep 6, 2023

Navi 32

60

5nm

28.1

346

3840

240

96

60

120

16

GDDR6

256

263

RX 7900 XT

$899

Dec 13, 2022

Navi 31

84

5nm

57.7

529

5376

336

192

84

168

20

GDDR6

320

315

RX 7900 XTX

$999

Dec 13, 2022

Navi 31

96

5nm

57.7

529

6144

384

192

96

192

24

GDDR6

384

355

Swipe to scroll horizontally
AMD Radeon RX 6000-series (RDNA 2)

Model

Launch

Code Name

Fab

Transistors (B)

Die Size (mm²)

Compute Units (CUs)

Shaders

TMUs

ROPs

Memory Size (GB)

Memory Type

MSRP (US$)

RX 6600

Oct 13, 2021

Navi 23

7nm

11.06

237

28

1792

112

64

8

GDDR6

$329

RX 6600 XT

Aug 11, 2021

Navi 23

7nm

11.06

237

32

2048

128

64

8

GDDR6

$379

RX 6700 XT

Mar 18, 2021

Navi 22

7nm

17.2

335

40

2560

160

64

12

GDDR6

$479

RX 6800

Nov 18, 2020

Navi 21

7nm

26.8

520

60

3840

240

96

16

GDDR6

$579

RX 6800 XT

Nov 18, 2020

Navi 21

7nm

26.8

520

72

4608

288

128

16

GDDR6

$649

RX 6900 XT

Dec 8, 2020

Navi 21

7nm

26.8

520

80

5120

320

128

16

GDDR6

$999

RX 6950 XT

May 10, 2022

Navi 21

7nm

26.8

520

80

5120

320

128

16

GDDR6

$1099

Swipe to scroll horizontally
AMD Radeon RX 5000-series (RDNA)

Model

Launch

Code Name

Fab

Transistors (B)

Die Size (mm²)

Compute Units (CUs)

Shaders

TMUs

ROPs

Memory Size (GB)

Memory Type

MSRP (US$)

RX 5500 XT

Dec 12, 2019

Navi 14

7nm

6.4

158

22

1408

88

32

4/8

GDDR6

$169

RX 5600 XT

Jan 21, 2020

Navi 10

7nm

10.3

251

36

2304

144

64

6

GDDR6

$279

RX 5700

Jul 7, 2019

Navi 10

7nm

10.3

251

36

2304

144

64

8

GDDR6

$349

RX 5700 XT

Jul 7, 2019

Navi 10

7nm

10.3

251

40

2560

160

64

8

GDDR6

$399

PC Gamer graphics card reviews

Nvidia RTX 5080 | January 2025

Nvidia RTX 5080 | January 2025
"The RTX 5080 Founders Edition uses the same lovely shroud as the top RTX Blackwell card, and brings the same DLSS/MFG feature set to the table. But that's all that is really setting the second-tier card apart from the RTX 4080 Super as the gen-on-gen performance difference is marginal at best. It might not be an exciting GPU, but at least the veneer of Multi Frame Generation will make it feel like a generational leap to most gamers."

PC Gamer score: 76%

AMD RX 9070 XT | March 2025

AMD RX 9070 XT | March 2025
"The Radeon RX 9070 XT, ably demonstrated by this Asus Prime version, is a great price/performance card that takes Nvidia to task on pricing, and shows AMD has taken great strides forward with both RT and AI processing. It could be a hugely consequential GPU, if only the AIBs can keep their worst pricing excesses in check."

PC Gamer score: 87%

Nvidia RTX 5070 | March 2025

Nvidia RTX 5070 | March 2025
"At this price, and with this competition looming large, I just don't know how I can recommend the RTX 5070 as a genuine purchase for any PC gamer without major caveats. The GPU at its heart feels like it should be the basis for an RTX 5060 and now the rest of market might just force that upon Nvidia."

PC Gamer score 61%

Nvidia RTX 5060 Ti | April 2025

Nvidia RTX 5060 Ti | April 2025
"A solid pick for an entry-level graphics card, especially at its lower-than-last-gen price. The RTX 5060 Ti would make a mean upgrade for someone still using an ageing GPU, provided you can get one close to MSRP."

PC Gamer score: 81%

Intel Arc B580 | December 2024

Intel Arc B580 | December 2024
"The one thing the Arc B580 had to do was perform consistently, had it done so it would be the best budget gaming GPU out there. For games where it does perform it actually is right now. But inconsistency not only in straight performance, but also whether a game would run or not, have made it all but impossible to recommend this card at launch."

PC Gamer score: 65%

Nvidia GeForce RTX 4090 | October 2022

Nvidia GeForce RTX 4090 | October 2022
"The RTX 4090 may not be subtle but the finesse of DLSS 3 and Frame Generation, and the raw graphical grunt of a 2.7GHz GPU combine to make one hell of a gaming card."

PC Gamer score: 83%

Nvidia RTX 4080 Super

Nvidia RTX 4080 Super
"The RTX 4080 Super is more a relaunch of the much maligned original RTX 4080 than an exciting new card in its own right, but it is still a GPU with serious gaming chops. But with no tangible performance difference it's the $200 relative price cut that does the work of rehabilitating the erstwhile RTX 4080. AMD's similarly priced, similarly performing RX 7900 XTX is still a thorn in the second-tier Ada card's side, however."

PC Gamer score: 81%

AMD Radeon RX 7900 XTX

AMD Radeon RX 7900 XTX
"The RX 7900 XTX is not a direct competitor to the RTX 4080 in every way, but it's notably cheaper and delivers a whole lot more than the top RX 6000-series card."

PC Gamer score: 81%

TOPICS
Dave James
Editor-in-Chief, Hardware

Dave has been gaming since the days of Zaxxon and Lady Bug on the Colecovision, and code books for the Commodore Vic 20 (Death Race 2000!). He built his first gaming PC at the tender age of 16, and finally finished bug-fixing the Cyrix-based system around a year later. When he dropped it out of the window. He first started writing for Official PlayStation Magazine and Xbox World many decades ago, then moved onto PC Format full-time, then PC Gamer, TechRadar, and T3 among others. Now he's back, writing about the nightmarish graphics card market, CPUs with more cores than sense, gaming laptops hotter than the sun, and SSDs more capacious than a Cybertruck.

With contributions from