AMD Radeon RX 9060 XT 16GB review: Finally, a powerful $350 GPU
Once again, the key for AMD is offering much more RAM than NVIDIA for far less money. Sure, AMD is using slower GDDR6 RAM with 320 GB/s of memory bandwidth, compared to NVIDIAs' 448 GB/s GDDR7. But in practice, actually having a larger memory pool is more useful when you're dealing with massive textures and 3D models in modern games. So if you're looking for the best gaming bang for the buck under $400, the 16GB Radeon RX 9060 XT is undoubtedly a winner.
The Radeon RX 9060 XT delivers a solid amount of 1080p and 1440p gaming power for just $350. It's not the fastest card around, but it's a great option for most people. $350 at AMD
The least powerful member of AMD's Radeon RX 9000 family (so far), the RX 9060 XT is basically the card you'd get if you can't stomach spending more than $500 on a GPU. Just remember that video cards tend to jump up in cost once they actually hit stores, depending on availability and the whims of retailers. The RX 9060 XT sports 32 RDNA 4 compute units, 32 RT accelerators and 64 AI accelerators. It can reach up to 3.1GHz speeds with its boost clock and consumes up to 180W of power. In comparison, the $549 MSRP RX 9070 has 56 compute units, while the $599 RX 9070 XT has eight more.
AMD is pitching this GPU as a replacement for the RX 7600 XT, which launched at $329 last year. That card also had 16GB of VRAM, but it had a slower 288 GB/s bandwidth. While the 7600 XT was more focused on entry-level 1440p gaming, the 9060 XT's hardware improvements make it far more capable. AMD claims it's 46 percent faster than the 7600 XT across more than 40 games, and in my testing that figure checks out.
For this review, I tested Gigabyte's RX 9060 XT GPU, which has three cooling fans pointed at a typical copper heatsink. There's no fancy vapor cooling chamber or any extravagant materials on the card, but at this price range I didn't expect much. Devindra Hardawar for Engadget
Across games and benchmarks, the 9060 XT clocked in just about where I expected: noticeably slower than the RX 9070, and either on-par or slightly slower than the NVIDIA RTX 5060 Ti. In 3DMark's Steel Nomad benchmark, the 9060 XT scored 200 more points than that NVIDIA card, and it was also 90 points ahead in Timespy Extreme. But NVIDIA had a 1,000 point lead in the Speedway benchmark and it was more than 2,000 points ahead in the Luxmark HDR 4 test.
Despite those figures, I'm still impressed that the 9060 XT can stay in the same league as NVIDIA's RTX 5060 Ti while costing a lot less. AMD's ray tracing performance has also improved quite a bit from its previous generations, scoring 45 percent faster in the 3DMark Port Royal benchmark. NVIDIA still has the lead in ray tracing in games though, especially with the one-two punch of DLSS 4 AI upscaling and multi-frame generation. NVIDIA's RTX 5000-series cards can produce up to three interpolate frames for every natively rendered frame, while AMD is still stuck with generating a single extra frame with its FSR 3 and 4 (Fidelity Super Resolution 4) upscaling.
None
3DMark TimeSpy Extreme
Geekbench 6 GPU
Cyberpunk (1440p RT Overdrive DLSS/FSR 3)
Blender
AMD Radeon RX 9060 XT
8,192
91,617
80fps
1,560
NVIDIA RTX 5060 Ti (16GB)
8,100
139,756
136 fps (4X frame gen)
4,220
AMD Radeon RX 9070
10,997
113,012
117 fps
N/A
AMD Radeon RX 7600
5,526
N/A
20 fps
1,013
The Radeon RX 9060 XT was able to run Cyberpunk 2077 in 1440p ray tracing "overdrive" mode at a respectable 80 fps, whereas NVIDIA's RTX 5060 Ti hit 136 fps thanks to multi-frame generation. Still, there's plenty of wiggle room to crank out more frames: Stepping down to 1080p got me 120 fps, and I saw 90 fps while using the less demanding ray tracing "ultra" preset in 1440p. In terms of raw performance, without any upscaling fanciness, the 9060 XT hit 114 fps in 1080p and 90 fps in 1440p.
The Radeon RX 9060 XT didn't run very hot during my testing, but that makes sense since it's purely focused on 1080p and 1440p gaming, During 3DMark's Steel Nomad stress test, which repeats the benchmark 20 times in succession, the card only topped out at 54 Celsius. I'm used to seeing video cards reaching upwards of 70C under load, so the 9060 XT was a nice surprise. While idling, it sits at 42 Celsius. That's higher than cards with vapor chamber cooling, but it's about average for cards in this price range. I didn't hear any annoying fan noise under load either, since it barely broke a sweat. Devindra Hardawar for Engadget
There isn't really much to complain about with the Radeon RX 9060 XT, so long as you're not expecting too much from a sub-$400 GPU. Still, it would be nice to see support for AMD's new FSR 4 AI upscaling spread out more quickly. When the RX 9070 XT launched, the feature was only available in a handful of games like Call of Duty: Black Ops 6, and now it's supported over 60 games. That's still about half of the games that NVIDIA's DLSS 4 supports (which so far includes over 125 titles).
On the RX 9060 XT, the FSR 4 user experience still isn't as simple as activating DLSS on NVIDIA cards. I had to enable it manually in AMD's Adrenaline software, then flip it on in Call of Duty: Warzone to activate it. On the bright side, I saw an average of 254 fps while playing a match in 1440p with extreme graphics settings, FSR 4 and frame generation flipped on. With frame gen enabled, but no FSR 4, performance dipped to 174fps. And if I removed both of those features, it fell to 110 fps. Clearly, FSR 4 and frame gen helps quite a bit, I just wish it were in more games.
Assuming retailers don't go wild by overpricing the Radeon RX 9060 XT, it's a solid option if you're looking for an affordable GPU that's dedicated to 1080p and 1440p gameplay. At $350, the 9060 XT is well below the RTX 5060 Ti's $429 retail price, and certainly much cheaper than models being sold for more than $500.
To reach a lower $300 price point, AMD also unveiled an RX 9060 Ti with 8GB of RAM. But honestly, I'd recommend avoiding that entirely unless you only play ancient games in 1080p. It's worth spending a bit more so you can play newer games with less headaches, and it's practically a requirement if you want to play in 1440p. Devindra Hardawar for Engadget
The Radeon RX 9060 XT is a reminder of a world where we didn't have to pay more than $400 to get a capable GPU. We already know more than 80 percent of PC gamers play at 1440p or lower resolutions, according to Steam's data. So for the vast majority of players out there — the ones who aren't using monitors with crazy high refresh rates or 4K+ resolutions — it's all the GPU you really need.
Hashtags

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


CNBC
a few seconds ago
- CNBC
AMD reports weaker-than-expected earnings even as revenue tops estimates
Advanced Micro Devices reported quarter earnings on Tuesday that missed estimates. The stock slid about 5% in extended trading. Here's how the chipmaker did versus LSEG expectations for the quarter ended June: For the current quarter, AMD expects sales of $8.7 billion, plus or minus $300 million, versus expectations of earnings of $8.3 billion. AMD reported net income during its fiscal second quarter of $872 million, or 54 cents per share, increasing from $265 million, or 16 cents per share in the year-ago period. Nvidia's overall sales rose 32% in the period from $5.84 billion a year earlier. AMD is the second-biggest maker of graphics processing units (GPUs) for artificial intelligence behind Nvidia, which has the vast majority of the market. But big AI customers such as Meta and OpenAI are increasingly looking to AMD to provide an alternative to Nvidia's pricey chips, especially for inference, or when AI models are deployed to the public. During the quarter, AMD announced new AI chips called the MI400 that are expected to hit the market next year. OpenAI CEO Sam Altman committed to using AMD's newest GPUs. AMD is also grappling with chip export controls which have been placed on some of its AI chips because the U.S. government worries that powerful GPUs could be used by adversaries to surpass American capabilities or be used for military purposes. The MI308 was previously barred for export to China in April, which the company said cost it $800 million in the June quarter. However, the company said in July that it expected shipments to resume after the Trump administration signaled that it would approve waivers. AMD said its outlook doesn't include any revenue from its China-focused AI chip called the MI308. AMD's adjusted gross margin during the quarter was 43%. The company said it would have been 54% if not for export control costs. AMD's main business, aside from GPUs, is making central processors, called CPUs, which compete with Intel to power more traditional servers. Both are reported in the company's data center segment, which had $3.2 billion in revenue, up 14% on an annual basis. The other major segment for AMD is called Client and Gaming, which includes the company's CPUs for laptops and desktops, and its GPUs for 3D gaming. That was up 69% on an annual basis driven by strong demand for the company's latest desktop CPUs, which it calls AMD Ryzen Zen 5. Gaming revenue was $1.1 billion, up 73% on an annual basis due to increased demand for custom chips for game consoles and gaming GPUs, AMD said.


CNBC
a few seconds ago
- CNBC
Super Micro shares plunge 15% on weak results, disappointing guidance
Super Micro Computer shares slid 15% in extended trading on Tuesday after the server maker reported disappointing fiscal fourth-quarter results and issued weak quarterly earnings guidance. Here's how the company did in comparison with LSEG consensus: Super Micro's revenue increased 7.5% during the quarter, which ended on June 30, according to a statement. For the current quarter, Super Micro called for 40 cents to 52 cents in adjusted earnings per share on $6 billion to $7 billion in revenue for the fiscal first quarter. Analysts surveyed by LSEG were looking for 59 cents per share and $6.6 billion in revenue. For the 2026 fiscal year, Super Micro sees at least $33 billion in revenue, above the LSEG consensus of $29.94 billion. Super Micro saw surging demand starting in 2023 for its data center servers packed with Nvidia for handling artificial intelligence models and workloads. Growth has since slowed. The company avoided being delisted from the Nasdaq after falling behind on quarterly financial filings and seeing the departure of its auditor. As of Tuesday's close, Super Micro shares were up around 88% so far in 2025, while the S&P 500 index has gained 7%. Executives will discuss the results on a conference call starting at 5 p.m. ET.


Forbes
a few seconds ago
- Forbes
The Clock Is Ticking On AI Security In A Quantum World
AI now runs in courtrooms, hospitals, airports, banks and several industries, becoming the crown jewel of many modern enterprises. However, protecting these AI systems in a quantum future is becoming increasingly difficult. Somewhere between the optimism of generative AI and the acceleration of quantum computing is a growing risk that few organizations are addressing today. While many worry about adversarial prompts and model hallucinations, experts say those are the least of our problems. David Harding, CEO of Entrokey Labs — a cybersecurity firm building quantum-resistant key infrastructure — warned that the real risk lies in how AI systems handle sensitive data. He argued that AI systems, and the massive volumes of sensitive data they ingest, may soon be the first victims of quantum-enabled cyberattacks. And most companies are walking into that future blind. The Quantum Threat Isn't Theory Anymore Earlier this year, Nvidia CEO Jensen Huang described quantum computing as reaching 'an inflection point.' While that statement sparked interest among investors, its implications for cybersecurity — particularly for AI-driven systems — haven't fully sunk in. As researchers push closer to building scalable quantum machines, long-standing encryption protocols such as RSA and ECC could be broken, making previously secure data fair game. In other words, the data feeding your AI today may be tomorrow's biggest liability. This isn't some distant sci-fi scenario. The groundwork has already begun. Nation-state actors are believed to be stockpiling encrypted data using what's known as a 'harvest now, decrypt later' strategy. Think of it like thieves stealing locked safes today knowing they'll get the keys tomorrow. Once quantum machines become powerful enough, they could retroactively decrypt troves of corporate secrets, defense communications and medical data, including everything passed through AI models today. 'Any electronic data is at risk from harvest now, decrypt later if it is not using digital keys resistant to today's AI attacks and near-term quantum attacks,' said Harding. 'Several countries including Russia, China, Iran and North Korea have well over 100,000 individuals solely focused on hacking our systems. Add automation into the mix, and the scale becomes nearly unmanageable.' Quantum threatens all digital systems, but AI amplifies the risk. These models don't just generate content — they ingest patient records, financial models, intellectual property and legal data. In autonomous systems, they make decisions. In others, they write code and trigger workflows. That puts entire AI pipelines — from training data to deployed agents — directly in the crosshairs. 'Quantum and AI-safe encryption has the same level of importance as the foundation of a building,' explained Scott Streit, Entrokey Labs' chief scientist. 'Without it, the structure collapses. There'd be no protection for customer data, IP or communications. In national security, satellites or precision weapons could be taken over.' Falling Behind The Curve Despite these risks, many enterprises still treat quantum computing as a future problem — something to solve by 2030. The U.S. National Institute of Standards and Technology (NIST) has laid out a path for adopting quantum-safe cryptography by 2035. But according to Harding, that timeline no longer reflects how fast both AI and quantum capabilities are evolving. 'The timeline is increasingly out of step with the pace of AI and quantum advancements,' said Harding. 'Some believe AI is already breaking into encryption systems.' And yet, most organizations continue to treat quantum-readiness as a long-haul IT project, involving years of consultations, infrastructure upgrades and vendor reviews. Harding refers to this pattern as 'cyber inertia' — an outdated playbook for a much faster threat. 'We're trying to solve a smarter threat with outdated answers,' Harding said. Streit added that 'AI can already create math that top mathematicians can't explain,' arguing that 'the only way to win is by using AI to secure AI.' To make matters worse, regulatory frameworks haven't caught up. Neither the EU AI Act nor NIST's AI Risk Management Framework say much about defending AI systems against quantum cryptographic threats, leaving a critical vulnerability unaddressed at the policy level. What's At Stake The financial fallout from a breach caused by quantum decryption is hard to estimate. But the principle is simple: What's considered secure today may not be tomorrow. That includes confidential model outputs, internal prompts, logged agentic decisions and sensitive metadata. Any of it could be exposed or tampered with. 'Think about how we respond to weather warnings,' Harding said. 'If there's even a 10% chance of a tornado, you don't wait. You get to shelter.' He added that this level of risk isn't something CISOs can handle alone. 'Quantum is a boardroom issue now — not just an engineering one. The scale of impact makes Y2K look like a warm-up act.' If Trust Fails, AI Fails While companies double down on AI performance, many remain dangerously naive about the risks embedded at its roots. As Harding put it, 'The question is no longer whether quantum will impact AI systems, but how quickly organizations can adapt before it does.' AI security depends not just on encryption, but on anticipating how fragile the entire ecosystem becomes when that encryption fails. If attackers can retroactively decrypt, reroute, or manipulate those systems, the blow to public confidence could rival or exceed any previous cyber event. Trust is what gives AI its power. Lose that, and even the smartest models would collapse. 'We've built an entire era of decision-making on architectures that might be more fragile than we thought,' Harding said. 'While companies chase optimization, adversaries are chasing the keys.'