AMD Radeon RX 9070 and 9070 XT review: Hitting NVIDIA where it hurts
AMD has finally done it. For years it's tried to undercut NVIDIA with slightly cheaper, but less capable, video cards like the Radeon 6700 XT and 7900 XT. And sure, it's still following that same strategy with the new Radeon 9070 and 9070 XT. This time around, though, AMD has produced far more capable hardware, especially when it comes to 4K and ray tracing performance. And there's hope that it could finally catch up to NVIDIA's DLSS AI upscaling with its new FidelityFX Super Resolution 4 (FSR4) technology.
These aren't perfect video cards, to be clear. But for $549 and $599, the Radeon 9070 and 9070 XT are far more compelling than AMD's previous lineup. That's particularly true since the 9070 XT is $150 less than NVIDIA's RTX 5070 Ti, and in our testing it's sometimes faster than that card.
AMD still has to prove that it can catch up to NVIDIA's DLSS, which has had several years to steadily improve its AI upscaling capabilities. In particular, AMD needs to match the performance of NVIDIA's multi-frame generation in DLSS 4, which has led to some surprisingly high fps figures while testing the RTX 50-series GPUs. AMD's Fluid Motion Frames technology is a start, but it's not nearly enough at this point.
The Radeon RX 9070 and 9070 XT are AMD's first RDNA 4 GPUs, a new platform sporting more powerful and efficient compute units, third-generation ray tracing accelerators and second-generation AI accelerators. There isn't a huge technical difference between the two cards: The RX 9070 features 56 compute units, 56 ray accelerators and 112 AI accelerators, whereas the 9070 XT has eight more compute accelerators and 16 additional AI processors. The XT model is clocked slightly higher, as you'd expect, and it also draws more power (304W vs 220W). Notably, both cards also ship with 16GB of GDDR6 VRAM, whereas NVIDIA skimped a bit and only put 12GB of RAM in the $549 RTX 5070.
As I was comparing our XFX-built review units, though, it was practically impossible to tell the RX 9070 and 9070 XT apart without reading the fine print on their labels. This is fairly typical for AMD cards with XT variants, and it makes sense since they're technically so similar. The XFX Swift GPUs I tested came in an elegant frost white case, with three large fans and surprisingly massive heatsinks. They're both 3.5-slot cards, so they also take up more room than the RTX 5090, which is a dense two-slot beast.
Thanks to AMD's efficient RDNA design, you also don't need to rely on complex power dongles to drive these cards. They both take two PSU connections directly, and XFX recommends 800-watt power supplies at a minimum. That's significantly higher than AMD's minimum 650W spec though, so it could be that XFX is just trying to play it safe with its massive cooling setup. (If you're planning to overclock these cards, though, you'll likely want more than a mere 650W PSU.)
I knew AMD was onto something special when I learned the Radeon RX 9070 averaged 60 fps in Cyberpunk 2077 in 4K Ray Tracing Overdrive mode. Sure, it needed FSR 3 upscaling to get there, but it was still better than the 50 fps I saw on the Radeon 7900XT a few years ago. The game occasionally dipped below 60 fps, but it was still very playable. 1440p was much smoother overall, averaging 117 fps. The Radeon 9070 XT, meanwhile, averaged 68 fps in 4K and 130 fps in 1440p.
Sure, these numbers are far behind the raw figures from NVIDIA's RTX 5070, which used the magic of multi-frame generation to deliver 115 fps in 4K with ray tracing and maxed out graphics. That card also hit 205 fps in 1440p. But as impressed as I was by the 5070, most of those frames were just meant to deliver the illusion of smoothness. During actual gameplay, I had a hard time seeing much slowdown with either Radeon card in 4K, and the difference between the NVIDIA card was practically erased in 1440p. Remember, fps figures aren't the entire story, even though NVIDIA wants you to believe otherwise.
I kept that idea in mind as I tested FSR 4, which surprised me when it averaged around 20 fps less than FSR 3 in Call of Duty: Warzone on both GPUs. The Radeon 9070 XT reached 250 fps in 4K with FSR 3, cranked up graphics and frame generation, whereas it hit 229 fps in FSR 4. AMD tells me that's no error, it's to be expected since FSR 4's AI upscaling is more focused on delivering higher quality graphics rather than a pure frame boost. I couldn't really see a huge difference while dodging bullets in CoD, but I did notice that finer textures like chain-link fences looked a bit clearer in FSR 4. (I also saw a few random glitched textures, something that was common in DLSS upscaled games early on. They weren't a major problem, but AMD clearly has to refine its upscaling model further.)
At the moment, enabling FSR 4 is a bit counter-intuitive as well. You have to turn it on in AMD's driver software, and then flip on FSR 3.1 in a compatible game. Warzone also required a reboot to fully enable the feature, but the game didn't prompt me to do so. And if you want frame generation, that's another option that has to be toggled on outside of the game. Hopefully this process will be smoothed out over time, along with wider availability for FSR 4. In addition to Black Ops 6, it's supported in the newly launched FragPunk, Civilization 7, Marvel Rivals and a handful of PlayStation 5 ports like the Spider-Man games. But it's nowhere to be found in Avowed or Dragon Age: The Veilguard, where you'll be stuck with FSR 3.
For games that don't work with FSR at all, AMD's Adrenaline software also has a 'HYPR-RX' mode that enables features like Radeon Super Resolution upscaling (a separate driver-level technology) and AMD's Fluid Motion Frames generation. Altogether, they led to me seeing 200 fps in Forza Horizon 5 using the RX 9070 XT in 4K with maxed out graphics settings, up from 85 fps natively. But again, those are just fps figures – AMD points out Radeon Super Resolution may not look as clear as FSR alternatives. (I didn't notice any weirdness in Forza, but I may have been distracted by the beautiful racing vistas in Mexico.)
GPU
3DMark TimeSpy Extreme
Geekbench 6 GPU
Cyberpunk (4K RT Overdrive DLSS)
Port Royal ray tracing
AMD Radeon 9070
10,997
113,012
60 fps (DLSS 3 w/ frame gen)
15,888
AMD Radeon 9070 XT
13,060
130,474
68fps (DLSS 3 w/ frame gen)
17,959
NVIDIA RTX 5070
10,343
178,795
115 fps (4x frame gen)
13,920
NVIDIA RTX 5070 Ti
12,675
238,417
153 fps (4X frame gen)
19,309
AMD Radeon 7900XTX
12,969
N/A
55 fps (DLSS 3)
14,696
When it comes to benchmarks, the Radeon RX 9070 and 9070 XT hold their own against NVIDIA's 5070 and 5070 Ti in most of 3DMark's tests. In the Steel Nomad benchmark, the 9070 scored 1,100 points higher than the 5070, and the 9070 XT beat out the more expensive 5070 Ti by almost 3,000 points in Timespy Extreme. I was also surprised to see the RX 9070 scoring almost 2,000 points higher than the RTX 5070 in the Port Royal ray tracing benchmark. Previously, ray tracing of any kind was AMD's Achilles heel.
Both of our XFX cards also stayed surprisingly cool during benchmarks and extended gaming sessions. They never went beyond 65C under load, and they typically idled below 40C. And since they never got very hot, I could barely hear their fans spinning up.
On paper, both of AMD's new GPUs are compelling alternatives to NVIDIA's midrange cards for 1440p gamers who occasionally dabble in 4K. The RX 9070 XT is particularly interesting, since it's noticeably faster and still comes in $150 less than the RTX 5070 Ti. But we're also dealing with a chaotic time in the PC gaming world, where GPU stock can disappear quickly and prices can rocket up quickly.
If you can nab either Radeon card at their listed prices, they'll be good deals. But it's not worth overpaying by too much for now. It also remains to be seen how the Trump administration's combative tariffs will affect pricing for PC hardware and electronics. Prices could easily jump by 20 percent or more to cover those costs.
While NVIDIA's DLSS 4 technology is more mature and leads to higher interpolated frame rates, there's also a good argument for going with AMD's cards since they have 16GB of VRAM. They'll be better suited to handling larger textures in games down the line, and there's also the potential for FSR 4 to improve as well.
It's clear now why AMD was focused on upgrading its mid-range Radeon cards first. There's not much point competing with NVIDIA at the extreme high end, like it did with Radeon RX 7900 XT and XTX. It makes more sense to focus on cards people can actually buy. The Radeon 9070 and 9070 XT also solve many of the problems I've had with AMD's GPUs in the past. They can go toe-to-toe with NVIDIA's cards, they have better ray tracing support and finally, they have AI upscaling. It remains to be seen if AMD will actually build on the promise of those features, but these cards are a hopeful start.

Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles


CNN
2 hours ago
- CNN
Why this key chip technology is crucial to the AI race between the US and China
In the largest single foreign investment in US history, Taiwan Semiconductor Manufacturing Company has unveiled a $100 billion investment, drawing global attention and prompting concern in Taiwan. TSMC, which produces more than 90% of the world's advanced semiconductor chips that power everything from smartphones and artificial intelligence (AI) applications to weapons, will build two new advanced packaging facilities in Arizona, among others. Here's everything you need to know about advanced packaging technology, which has seen exponential demand growth along with the global AI frenzy, and what that means for the struggle between the US and China for AI dominance. While the two countries have announced a temporary truce that rolled back disruptive three-digit tariffs for 90 days, the relationship remains tense because of ongoing feuding over chip restrictions imposed by the US and other issues. Last month at Computex, an annual trade show in Taipei that has been thrust under the limelight because of the AI boom, the CEO of chipmaker Nvidia, Jensen Huang, told reporters that 'the importance of advanced packaging for AI is very high,' proclaiming that 'no one has pushed advanced packaging harder than me.' Packaging generally refers to one of the manufacturing processes of semiconductor chips, which means sealing a chip inside a protective casing and mounting it to the motherboard that goes into an electronic device. Advanced packaging, specifically, refers to techniques that allow more chips — such as graphic processing units (GPU), central processing units (CPU) or high bandwidth memory (HBM) — to be placed closer together, leading to better overall performance, faster transmission of data and lower energy consumption. Think of these chips as different departments within a company. The closer these departments are to each other, the easier it is, and less time it takes, for people to travel between them and exchange ideas, and the more efficient the operation becomes. 'You're trying to put the chips as close together as possible, and you're also putting in different solutions to make the connection between the chips very easy,' Dan Nystedt, vice president of Asia-based private investment firm TrioOrient, told CNN. In a way, advanced packaging keeps afloat Moore's Law, the idea that the number of transistors on microchips would double every two years, as breakthroughs in the chip fabrication process become increasingly costly and more difficult. While there are many types of advanced packaging technologies, CoWoS, short for Chips-on-Wafer-on-Substrate and invented by TSMC, is arguably the best known that was thrown under the limelight since the debut of OpenAI's ChatGPT, which sparked the AI frenzy. It has even become a household name in Taiwan, prompting Lisa Su, CEO of Advanced Micro Devices (AMD), to say that the island is the 'only place that you can say CoWoS and everybody would understand.' Advanced packaging has become a big deal in the tech world because it ensures AI applications, which require a lot of complex computing, run without delays or glitches. CoWoS is indispensable to producing AI processors, such as the GPUs produced by Nvidia and AMD that are used in AI servers or data centers. 'You could call it the Nvidia packaging process if you want to. Almost anyone making AI chips is using the CoWoS process,' said Nystedt. That is why demand for CoWoS technology has skyrocketed. As a result, TSMC is scrambling to ramp up production capacity. In a visit to Taiwan in January, Huang told reporters that the amount of advanced packaging capacity currently available was 'probably four times' what it was less than two years ago. 'The technology of packaging is very important to the future of computing,' he said. 'We now need to have very complicated advanced packaging to put many chips together into one giant chip.' If advanced fabrication is one piece of the puzzle in terms of chip manufacturing, advanced packaging is another. Analysts say having both pieces of that jigsaw in Arizona means the US will have a 'one-stop shop' for chip production and a strengthened position for its AI arsenal, benefitting Apple, Nvidia, AMD, Qualcomm and Broadcom, some of TSMC's top clients. 'It ensures that the US has a complete supply chain from advanced manufacturing to advanced packaging, which would benefit the US' competitiveness in AI chips,' Eric Chen, an analyst with market research firm Digitimes Research, told CNN. Because advanced packaging technologies key to AI are currently only produced in Taiwan, having it in Arizona also reduces potential supply chain risks. 'Instead of having all eggs in one basket, CoWoS would be in Taiwan and also the US, and that makes you feel more safe and secure,' said Nystedt. While CoWoS got its moment recently, the technology has actually existed for at least 15 years. It was the brainchild of a team of engineers led by Chiang Shang-yi, who served two stints at TSMC and retired from the company as its co-chief operating officer. Chiang first proposed developing the technology in 2009 in an attempt to fit more transistors into chips and solve bottlenecks in performance. But when it was developed, few companies took up the technology because of the high cost associated with it. 'I only had one customer … I really became a joke (in the company), and there was so much pressure on me,' he recalled in a 2022 oral history project recorded for the Computer History Museum in Mountain View, California. But the AI boom turned CoWoS around, making it one of the most popular technologies. 'The result was beyond our original expectation,' Chiang said. In the global semiconductor supply chain, companies that specialize in packaging and testing services are referred to as outsourced semiconductor assembly and test (OSAT) firms. In addition to TSMC, South Korea's Samsung and America's Intel, as well as OSAT firms including China's JCET Group, America's Amkor and Taiwan's ASE Group and SPIL are all key players in advanced packaging technologies.


CNN
2 hours ago
- CNN
Why this key chip technology is crucial to the AI race between the US and China
In the largest single foreign investment in US history, Taiwan Semiconductor Manufacturing Company has unveiled a $100 billion investment, drawing global attention and prompting concern in Taiwan. TSMC, which produces more than 90% of the world's advanced semiconductor chips that power everything from smartphones and artificial intelligence (AI) applications to weapons, will build two new advanced packaging facilities in Arizona, among others. Here's everything you need to know about advanced packaging technology, which has seen exponential demand growth along with the global AI frenzy, and what that means for the struggle between the US and China for AI dominance. While the two countries have announced a temporary truce that rolled back disruptive three-digit tariffs for 90 days, the relationship remains tense because of ongoing feuding over chip restrictions imposed by the US and other issues. Last month at Computex, an annual trade show in Taipei that has been thrust under the limelight because of the AI boom, the CEO of chipmaker Nvidia, Jensen Huang, told reporters that 'the importance of advanced packaging for AI is very high,' proclaiming that 'no one has pushed advanced packaging harder than me.' Packaging generally refers to one of the manufacturing processes of semiconductor chips, which means sealing a chip inside a protective casing and mounting it to the motherboard that goes into an electronic device. Advanced packaging, specifically, refers to techniques that allow more chips — such as graphic processing units (GPU), central processing units (CPU) or high bandwidth memory (HBM) — to be placed closer together, leading to better overall performance, faster transmission of data and lower energy consumption. Think of these chips as different departments within a company. The closer these departments are to each other, the easier it is, and less time it takes, for people to travel between them and exchange ideas, and the more efficient the operation becomes. 'You're trying to put the chips as close together as possible, and you're also putting in different solutions to make the connection between the chips very easy,' Dan Nystedt, vice president of Asia-based private investment firm TrioOrient, told CNN. In a way, advanced packaging keeps afloat Moore's Law, the idea that the number of transistors on microchips would double every two years, as breakthroughs in the chip fabrication process become increasingly costly and more difficult. While there are many types of advanced packaging technologies, CoWoS, short for Chips-on-Wafer-on-Substrate and invented by TSMC, is arguably the best known that was thrown under the limelight since the debut of OpenAI's ChatGPT, which sparked the AI frenzy. It has even become a household name in Taiwan, prompting Lisa Su, CEO of Advanced Micro Devices (AMD), to say that the island is the 'only place that you can say CoWoS and everybody would understand.' Advanced packaging has become a big deal in the tech world because it ensures AI applications, which require a lot of complex computing, run without delays or glitches. CoWoS is indispensable to producing AI processors, such as the GPUs produced by Nvidia and AMD that are used in AI servers or data centers. 'You could call it the Nvidia packaging process if you want to. Almost anyone making AI chips is using the CoWoS process,' said Nystedt. That is why demand for CoWoS technology has skyrocketed. As a result, TSMC is scrambling to ramp up production capacity. In a visit to Taiwan in January, Huang told reporters that the amount of advanced packaging capacity currently available was 'probably four times' what it was less than two years ago. 'The technology of packaging is very important to the future of computing,' he said. 'We now need to have very complicated advanced packaging to put many chips together into one giant chip.' If advanced fabrication is one piece of the puzzle in terms of chip manufacturing, advanced packaging is another. Analysts say having both pieces of that jigsaw in Arizona means the US will have a 'one-stop shop' for chip production and a strengthened position for its AI arsenal, benefitting Apple, Nvidia, AMD, Qualcomm and Broadcom, some of TSMC's top clients. 'It ensures that the US has a complete supply chain from advanced manufacturing to advanced packaging, which would benefit the US' competitiveness in AI chips,' Eric Chen, an analyst with market research firm Digitimes Research, told CNN. Because advanced packaging technologies key to AI are currently only produced in Taiwan, having it in Arizona also reduces potential supply chain risks. 'Instead of having all eggs in one basket, CoWoS would be in Taiwan and also the US, and that makes you feel more safe and secure,' said Nystedt. While CoWoS got its moment recently, the technology has actually existed for at least 15 years. It was the brainchild of a team of engineers led by Chiang Shang-yi, who served two stints at TSMC and retired from the company as its co-chief operating officer. Chiang first proposed developing the technology in 2009 in an attempt to fit more transistors into chips and solve bottlenecks in performance. But when it was developed, few companies took up the technology because of the high cost associated with it. 'I only had one customer … I really became a joke (in the company), and there was so much pressure on me,' he recalled in a 2022 oral history project recorded for the Computer History Museum in Mountain View, California. But the AI boom turned CoWoS around, making it one of the most popular technologies. 'The result was beyond our original expectation,' Chiang said. In the global semiconductor supply chain, companies that specialize in packaging and testing services are referred to as outsourced semiconductor assembly and test (OSAT) firms. In addition to TSMC, South Korea's Samsung and America's Intel, as well as OSAT firms including China's JCET Group, America's Amkor and Taiwan's ASE Group and SPIL are all key players in advanced packaging technologies.

Yahoo
3 hours ago
- Yahoo
Trump's Secret AI Strategy Uncovered: Tech Insider Reveals Hidden Role of Elon Musk
BALTIMORE, June 07, 2025 (GLOBE NEWSWIRE) -- New briefing claims President Trump has quietly deployed the most advanced artificial intelligence system in U.S. history — and handed the keys to Elon Musk. In a recently surfaced briefing, tech entrepreneur and author James Altucher alleges that Donald Trump's first move back in the Oval Office was more than symbolic — it was tactical. According to Altucher, Trump's repeal of Biden's AI executive order set off a hidden chain reaction that gave Elon Musk the freedom to unleash a classified project known as Project Colossus — an AI facility located in Memphis, Tennessee that may now house the most powerful supercomputer on Earth. 'The Fastest Supercomputer on the Planet' 'It contains not just one or two… but 200,000 units of Nvidia's all-powerful AI chips… making it the most advanced AI facility known to man.' 'The fastest supercomputer on the planet.' — Jensen Huang, Nvidia CEO Altucher says the facility is growing fast — with rumors Musk plans to double its processing power within weeks. A Quiet Reversal. A Bold New Doctrine. 'In one of his FIRST acts as President… Donald Trump overturned Executive Order #14110.' That order — signed by Biden — had imposed regulatory constraints on advanced AI. Altucher claims the reversal was calculated, opening the door for private actors like Musk to move 'without delay or oversight.' Trump later followed with a $500 billion initiative dubbed Stargate, focused on expanding America's AI infrastructure. Beyond Chatbots — Toward Supremacy Altucher argues that what's being built isn't just faster software — it's a new class of intelligence entirely. 'AI 2.0… gives that knowledge to intelligent machines that I believe will solve our problems for us.' He says July 1 could mark a critical turning point, when Musk unveils a 10X expansion to Colossus — potentially redefining U.S. dominance in AI for decades. About James Altucher James Altucher is a computer scientist, bestselling author, and entrepreneur who has worked in AI for over 40 years. He has advised tech firms, built AI trading systems, and contributed to IBM's Deep Blue supercomputer project. His latest research focuses on the convergence of AI, national policy, and private innovation. Media Contact:Derek WarrenPublic Relations ManagerParadigm Press GroupEmail: dwarren@ in retrieving data Sign in to access your portfolio Error in retrieving data Error in retrieving data Error in retrieving data Error in retrieving data