Look, we understand; we’re fully aware that the best graphics card isn’t easy to come by right now. The greatest graphics cards are becoming increasingly hard to find as demand grows and supply shrinks.
But keep in mind that it’s not an impossible task. You might be able to find a powerful graphics card that’s worth your money in 2022 if you know exactly what you’re looking for in terms of price/performance.
1. Nvidia GeForce RTX 3080
When the Nvidia GeForce RTX 3080 was first announced, it had to outperform the company’s previous generation’s top graphics cards. Thankfully, it’s been up to the task, just like the rest of the 3000 series, delivering 4K gaming to the public.
The RTX 3080’s gains over the cards it replaces appear to be the most significant generational increase in power we’ve seen in a long time. It outperforms the RTX 2080 Ti by 20-30% and, even more impressively, the RTX 2080 by 50-80 percent.
What makes this GPU even more appealing is that the performance boost comes at a significantly lower retail price. It costs nearly half as much as the previous generation GPU.
The Nvidia GeForce RTX 3080 makes high-end gaming more accessible to the ordinary gamer, running them to play the top PC games at greater resolutions and refresh rates for less money.
And, if that wasn’t enough, it’s been revealed that a 12GB RTX 3080 version is on the way, with performance and speed exceeding expectations.
The Nvidia GeForce RTX 3080 Founders Edition will be available on September 17 for $699 (£649, around AU$950). As with any major graphics card introduction, dozens of aftermarket graphics cards from companies like MSI, Asus, Zotac, and others will be available.
Just keep in mind that, depending on factors like exotic cooling systems and factory-tuned overclocks, some of these aftermarket card designs may see significant price rises over the Founders Edition.
However, any RTX 3080 should perform similarly to the Nvidia-branded performance. The Nvidia GeForce RTX 3080 is based on Nvidia’s new Ampere graphics architecture, which improves raw performance and power efficiency dramatically.
Because Nvidia boosted the power budget so much over the RTX 2080 while increasing power efficiency, the entire performance profile is much beyond what any Nvidia Turing graphics card could achieve.
The RT and Tensor cores have improved – we’re on the second and third generation, respectively – but the rasterization engine has likely seen the most progress.
Nvidia was able to increase the amount of CUDA cores on each Streaming Multiprocessor (SM) by allowing both data channels on each SM to perform Floating Point 32 (FP32) workloads, which is a significant advance over Turing, which only had one data path dedicated to integer operations.
This practically doubles raw FP32 throughput core for core, however, this won’t necessarily translate to a doubled frame rate in your favorite PC games – at least not for many of them.
This means that, while having only 46 percent more SMs (68) than the RTX 2080, the Nvidia GeForce RTX 3080 has more than doubled the CUDA core count (from 2,944 to 8,704). This equates to a roughly threefold increase in potential FP32 throughput, from around 10 to 29.7 TFLOPs – a tremendous generational leap.
When you combine the increase in CUDA cores with massive increases in Cache, Texture Units, and Memory Bandwidth – all thanks to the switch to faster GDDR6X memory on a 320-bit bus – gaming performance sees one of the biggest generational leaps in years, even if it falls short of the ‘2x performance’ target that we’re sure some people were hoping for. But we’ll get to that later.
Nvidia RT cores are also back – after all, that’s why Nvidia has the RTX moniker – and they’ve gotten a lot better. Second-generation RT cores are included in Nvidia Ampere graphics cards, including the RTX 3080, and will perform comparably to first-generation RT cores while being twice as efficient.
When using ray tracing, the SM will cast a light ray in a rendered scene, and the RT core will take over from there, performing all of the calculations required to find where that light ray bounces and reporting that information back to the SM.
This means the SM is left to render the rest of the scene on its own. However, we’re still not at the point where enabling ray tracing does not affect performance. Perhaps one day.
Tensor cores are also twice as powerful this time around, prompting Nvidia to incorporate only four in each SM instead of the eight seen in a Turing SM.
DLSS performance improves dramatically as a fact of the coupled number of SMEs in general. Nvidia RTX I/O, which will operate in collaboration with Microsoft’s DirectStorage API to transmit data directly from your SSD to your graphics card, is one of the features we’re most interested to see implementing.
This should not only drastically cut loading times in next-generation games, but also reflect the breakthrough I/O performance teased with next-generation consoles like the PS5 and Xbox Series X. This technology is far more crucial to future gaming technology than fast frame rates or beautiful graphics.
Unfortunately, this is a technology that game creators must integrate into their games, and we weren’t able to observe what type of effect it will make in the real world. Though that’s something we’ll be actively testing until the technology is widely adopted – and, given the consoles will be using identical technology, we expect it to be faster than ray tracing.
Nvidia chose an entirely new cooler design for the Founders Edition graphics card, which is far more practical than anything it’s ever done with a reference design previously. To make the back end of the card all heatsink, the business employed a shorter, multi-layered PCB.
Nvidia was able to put a fan on the back of the graphics card, sucking cool air through the heatsink and expelling it up and out of the chassis as a result of this.
We were concerned when we initially saw this fan design that it would influence CPU and RAM temperatures because it blows hot air directly over these components, but we didn’t see any difference in performance in our system, which has a Noctua NH-12UA air cooler on an AMD Ryzen 9 3900X.
We suppose it helps that most PC games don’t truly push both the GPU and the CPU to their limits — at least not yet.
The Founders Edition also includes three DisplayPort and one HDMI 2.1 output for screens, which is a nice feature. However, we are disappointed that Nvidia has removed the USB-C output, as producers will undoubtedly want to use this tremendously powerful card, and many professional monitors are USB-C monitors.
Despite our minor gripes with the Founders Edition – and despite our initial impressions that it was unattractive – it’s a beautiful piece of hardware in person. The RTX 3080 looks like a professional-grade piece of hardware, with its all-black design and silver accents.
We expected the RTX 3080 to be a fast graphics card based on Nvidia’s (overblown) marketing, but calling it “fast” is a bit of an understatement. It’s been in our machine since the day we got the box, running everything from Final Fantasy XIV to Control, and just coming out to be hooked into our test bench for genuine benchmarks.
2. AMD Radeon RX 6800 XT
AMD’s huge comeback to the high-end GPU market is the AMD Radeon RX 6800 XT. This powerful graphics card can even equal the performance of Nvidia’s GeForce RTX 3080, performing better in some games – such as Resident Evil Village at 1440p without ray tracing – while performing worse in others, all while having a slightly lower price tag. Finally, AMD has a viable and more enticing alternative to Nvidia’s heavy hitter on its hands.
For $649 (about £649, AU$960), gamers can receive 4K gaming power and advanced features like ray tracing with the new AMD Radeon RX 6800 XT.
For those looking to save a few dollars, this means a more cheap 4K gaming alternative. It also means that selecting the finest graphics card for 4K gaming has never been as difficult as it is now.
We don’t expect the AMD Radeon RX 6800 XT to dethrone Nvidia, but we do expect a lot of gamers, particularly those on a budget, to prefer it. The AMD Radeon RX 6800 XT is based on Team Red’s RDNA 2 graphics architecture, which is still based using a 7nm (nanometer) technology.
Rather than shrinking the die, AMD has added new features and increased efficiency, resulting in graphics cards that are significantly more powerful than their predecessors.
The new Infinity Cache is the most important of these, especially as the Radeon RX 6800 XT uses 16GB of GDDR6 memory rather than the GDDR6X memory found in the Nvidia GeForce RTX 3080 and RTX 3090.
The Infinity Cache is a 128MB global cache that, according to AMD, is 3.25 times quicker than GDDR6 VRAM on a 256-bit interface.
This allows data from the bigger 16GB VRAM pool to be sent to the GPU, and the combination of the fast cache and GDDR6 memory provides 2.4x more bandwidth per watt than the VRAM alone. This helps AMD close the gap it would otherwise have when competing against Nvidia’s GDDR6X memory in flagship-class cards.
The DirectX 12 Ultimate-compatible Ray Accelerators are the other key enhancement to the design. Each of the 72 Compute Units has one of these, which is what allows the AMD Radeon RX 6800 XT to handle real-time ray tracing at a playable framerate.
We’ll go into how effective these are later, but we expect performance to improve as AMD’s drivers mature, similar to how Nvidia’s RTX cards improved throughout the Turing cards’ existence.
Each Compute Unit (CU) is laid out similarly to the original RDNA cards, except the new Ray Accelerators. On the Radeon RX 6800 XT, each CU has 64 Stream Processors (SPs), for a total of 4,608 SPs. Of course, because the technology hasn’t shrunk, this means a significantly larger die, with 26.8 billion transistors spread across a 519mm2 area.
This larger die is coupled with greater clock speeds, too. The Game Clock of the Radeon RX 6800 XT is 2,015MHz, with a Boost Clock of up to 2,250MHz.
The Game Clock is what you’ll notice the most while playing games, whilst the Turbo Clock is essentially how fast it can boost up to for highly bursty workloads rather than steady loads like the finest PC games.
With more die space and faster clock speeds, power consumption increases dramatically. Although the AMD Radeon RX 6800 XT is rated for 300W, we never saw ASIC power (the amount of power drawn by the entire card) exceed 294W. In games, our MSI Afterburner overlay only reported 230-250W most of the time during our testing. That means this graphics card has a lot more room to increase its performance.
Hopefully, a new VBIOS (VGA BIOS) will be published that will show the 6800 XT to stretch even farther, allowing it to truly outperform the RTX 3080.
However, it falls short of expectations right out of the box. Fortunately, this means that temperatures will remain low. We measured a maximum temperature of 78°C during our testing, which is only slightly higher than the cut-down Radeon RX 6800.
The hotspot temperature, which only measures the temperature of the hottest region of the die, reached a stunning 114°C, slightly more than the 110°C AMD claimed was within functioning specs. It didn’t cause any thermal throttling, but if you’re keeping track of temps, you might notice some very high numbers.
What’s more, the Radeon RX 6800 XT is essentially a beefed-up version of the GPU found inside the Xbox Series X and PlayStation 5. This could suggest that games for the next generation of consoles will be easier to optimize for this graphics card.
The benefits of this can be seen in games like Assassins Creed Odyssey and Far Cry 5, which aren’t next-generation games but are mostly tuned for AMD hardware and demonstrate just how far performance can be improved through optimization.
AMD introduces two optional features with the Radeon RX 6800 series, which can assist boost performance. Smart Access Memory (SAM) and Rage Mode can both help you boost your performance, but keep your expectations in check.
Rage Mode is an AMD-approved overclock in the sense that it does not void your warranty. When you go into the Radeon program to enable it, it will worry you that you’re meddling with arcane settings at your peril – but don’t worry, if you’re using one of AMD’s presets, your warranty won’t be affected.
3. Nvidia GeForce RTX 3060 Ti
You don’t have to break the bank to get access to Ampere architecture with the Nvidia GeForce RTX 3060 Ti. While the RTX 3060 may be had for even cheaper, the 3060 Ti is a better deal because it offers higher performance for the same price.
The 3060 Ti, in particular, enhances the 3060’s strong 1080p performance by allowing you to play the greatest PC games at 1440p on high settings. You might even be able to get some 4K gaming out of it if you utilize DLSS. It also outperforms the RTX 2080 Super, a GPU that cost nearly twice as much when it was released.
The Nvidia GeForce RTX 3060 Ti isn’t the fastest or most durable GPU on the market, but it delivers good performance for a low price. It’s also an excellent GPU for those who find flagship cards like the Nvidia GeForce RTX 3080 or the more expensive RTX 3090 prohibitively expensive. For most gamers, it is likely the greatest graphics card.
The Nvidia GeForce RTX 3060 Ti is based on the same Ampere architecture as the RTX 3080, and it shows us how the Ampere portfolio will appear in the mid-range and budget GPU market. This isn’t going to be the graphics card that can push Watch Dogs Legion to 4K with ray tracing, but it can certainly handle 1080p.
This graphics card comes with 8GB of GDDR6 memory, the same as the RTX 3070, as well as 38 Ampere Streaming Multiprocessors (SM).
However, due to enhancements made by Nvidia to its SM since Turing, each of them now has 128 CUDA cores, which is double the number of CUDA cores found in each Turing SM. This means that the RTX 3060 Ti now has 4,864 CUDA cores, up from 2,176 in the RTX 2060 Super.
The RTX 2060 Super has a higher power consumption than the RTX 2060, although it’s not as high as previous Nvidia Ampere cards. The RTX 3060 Ti’s total graphics power (TGP) is 200W, up from 175W for the RTX 2060 Super.
It’s more, but not so much that you’ll have to worry about updating your power supply to accommodate it – provided you already have one that’s adequate to the task.
Each SM in Turing has two datapaths: one for Floating-Point (FP32) workloads and the other for Integer tasks. Nvidia was able to construct the SM with Ampere, and thus the RTX 3060 Ti, to have one of the datapaths practically do double duty, which is why the number of CUDA cores per SM has effectively doubled with Ampere.
Each SM additionally features Tensor Cores for AI applications like Deep Learning Super Sampling (DLSS) and dedicated ray-tracing cores in addition to CUDA cores. One RT core and four Tensor Cores are included in each SM.
The observant among you may have noted that each SM has half the number of Tensor Cores as Turing did, but there’s a reason for that: they’re more than twice as fast this time.
The ray tracing that everyone is talking about is handled by the RT core. The goal is to offload the tremendous computational load that any type of ray tracing implies so that playable framerates can be achieved.
In a gaming engine, when a light ray is cast, the SM sends that information to the RT core, which calculates where the ray bounces and reports that information back to the SM so it can render the image.
However, as you can see from Metro Exodus’ significant performance drop when ray tracing is enabled, this does not make ray tracing computationally free.
The performance hit without these dedicated RT cores, though, would convert the game into a slideshow.
Fortunately, the Tensor Core, another dedicated sort of core, is the second part of the equation that enables ray tracing reasonable to activate when playing the greatest PC games. DLSS, which is essentially an AI-enabled upscaling technology, can be integrated into games using the Tensor Core.
This improves speed by having the SM produce a scene at a lesser resolution, then having the Tensor Core intelligently scale that scene up to full resolution using data from Nvidia’s Supercomputers.
You can make up a lot of the speed difference that ray-tracing brings to the table with DLSS, so you can enjoy the amazing image quality benefits of ray tracing without sacrificing too much performance. Ray tracing would be unprofitable without DLSS.
The Nvidia GeForce RTX 3060 Ti also comes with a long list of functions. Nvidia Reflex, which attempts to reduce system latency, RTX IO, which interacts with Microsoft DirectStorage to load data directly from your system storage to your VRAM, and Nvidia Broadcast, which filters out background noise and visuals while broadcasting or video chatting, will appeal to esports players.
That final one has been a big help to us during the lockdown and the tens of thousands of video calls we’ve had to make from our bedroom in 2021. While many pieces of software, such as Zoom, Google Meet, and Microsoft Teams, have comparable capability, having all of the settings in one place saves a lot of time.
The Founders Edition of the Nvidia GeForce RTX 3060 Ti is nearly identical to the Founders Edition of the Nvidia GeForce RTX 3070. The two cards are identical in size, have the same number of fans, and have the same exposed fins on the rear of the card, which aid in the removal of hot air from the graphics card and out of the system.
The only significant difference between it and the RTX 3070 is that it is a lighter gray hue. Nvidia has created some of the best-looking graphics cards we’ve ever seen, just like the rest of the Founders Editions in the Ampere family.
What’s even better? The fashionable cooler is also functional. Temperatures in our complete testing suite peaked at 72°C, which is similar to the 72°C seen by the RTX 2060 Super, but at a 10% higher power consumption.
But it isn’t all sunshine and roses. While the cooler itself is fantastic, the fact that the GPU board has to be shrunk to accommodate it has the same issue as the rest of the range. The RTX 3060 Ti, in particular, has the same 12-pin power connector as the RTX 3060.
Although Nvidia includes an adaptor in the box that allows you to connect one 8-pin PCIe cable to the 12-pin connector, it complicates cable management. Fortunately, PSU manufacturers have already begun developing dedicated 12-pin PCIe cables, thus this issue should be resolved over time.
4. AMD Radeon RX 6900 XT
The AMD Radeon RX 6900 XT is the company’s new halo product, outperforming the Radeon RX 6800 XT and RX 6800 by including a fully powered Navi 21 GPU with 80 Compute Units. Our GPU Benchmarks hierarchy places the latest generation GPUs at the top and rates them as some of the best graphics cards currently available (if we use the term “available” loosely).
The Radeon Pro Duo (2016), a very limited-quantity dual-GPU card, was the last time AMD marketed a GPU priced at $1,000 or more, thus the 6900 XT is the most expensive AMD graphics card we’ve seen in a long time. With the RX 6800 XT taking on the RTX 3080, AMD’s goal is clear: to match the performance of the GeForce RTX 3090 while lowering the price by $500.
Of course, all of that is beside the point. Even eight months after its initial release, the RX 6900 XT, like all other RX 6000 and RTX 30 series graphics cards, is nearly impossible to obtain in stock. While the RX 6900 XT is just 55 percent more than the MSRP, it’s still a lot more than we’d recommend paying, according to our GPU pricing index.
We’ve updated this review with the most recent information while preserving the original performance findings. More current assessments, such as the RTX 3080 Ti, provide a more complete picture of how performance compares, but purchasing a new generation graphics card remains a major pain in the neck until prices return to something resembling normalcy.
While it’s great that the GeForce RTX 3080 and 3090 are cheaper than the Radeon RX 6800 XT and 6900 XT, there are a few issues when comparing the GeForce RTX 3080 and 3090 to the Radeon RX 6800 XT and 6900 XT. First, the VRAM of the RTX 3090 is more than double that of the 3080. It also includes a 20% increase in shader cores and a bigger memory bus.
Finally, it improved performance under professional workloads by 10% and increased the power goal by 10%. The RX 6900 XT, on the other hand, has 10% more shader cores than the 6800 XT. That’s all there is to it. There’s not much more to say about the specifications that we haven’t already said.
The RX 6900 XT includes the Navi 21 GPU’s full 80 CUs and 5120 shader cores, and the rest of the statistics, such as memory speed and TDP, are the same as the RX 6800 XT. That isn’t to imply that differences in silicon don’t exist.
Binning and sorting chips from silicon wafers aren’t new, but the best chips tend to emerge from the wafer’s center, where faults are less prevalent. Because the 6900 XT requires a fully functional Navi 21 chip, other aspects such as power and voltage requirements are almost certain to be improved.
What difference does it make in practice? If you want to overclock and get the most out of your computer, the silicon lottery can help you out. However, considering the 6900 XT already costs 50% more than the 6800 XT, that’s a significant price increase for a presumably tiny performance boost. Except for the product name, the Radeon RX 6900 XT looks similar to the RX 6800 XT.
However, the package isn’t quite the same. The RX 6900 XT is encased in a mouse pad inside the main box. If you need a mouse pad or AMD swag, it’s a pleasant little bonus. It’s also a rather large pad – not large enough to cover your entire desk, but considerably larger than a standard pad.
The RX 6900 XT is a powerful graphics card that weighs slightly more than the RTX 3080 but has similar proportions. It’s slightly thicker, taking up 2.7 slots, but it’s also 2cm shorter.
You might be able to squeeze it into a smaller build, but we advise caution because ejecting 300W (or more) of heat into a small case will degrade performance. If you’re building a smaller case, save some money and get the 6800 XT instead; we’re confident that once thermals settle, the two cards will run similarly.
The 6900 XT has a single HDMI 2.1, two DisplayPort 1.4, and one USB-C video output, just as the RX 6800 cards. Keep an eye out for third-party cards if that doesn’t satisfy your demands.
We’ve heard that AMD’s AIB partners will release RX 6900 XT cards (contrary to earlier rumors), which makes sense because anyone who builds an RX 6800 XT may utilize the same design for the RX 6900 XT. However, we don’t know when those cards will be available for purchase.
For power, there are two 8-pin PEG connectors, each rated at 150W. Power should be more than enough for regular operation when combined with the 75W from the PCIe slot. We noticed minor power spikes while starting up certain tasks (FurMark in particular), but they shouldn’t be a major issue with any good 850W power supply. If at all possible, we recommend using a single rail power supply.