How AMD's new FSR 4 tech could be the handheld gaming PC upgrade we've been waiting for
When you buy through links on our articles, Future and its syndication partners may earn a commission.
Love it or hate it, frame generation software continues to be a major feature of gaming graphics updates.
AMD's new FidelityFX Super Resolution (FSR) 4 frame-gen software offers massive updates for the Radeon 9070 and 9070 XT discrete GPUs, allowing AMD's new GPUs up to 3.7x better performance over the pure silicon power of the new RDNA 4 architecture.
While we aren't likely to see AMD release mobile variants of the new Radeon GPUs, we might see some of the RDNA 4 tech make its way to AMD's integrated graphics tiles on the Ryzen Z2 generation and the Ryzen AI 300 series chipsets.
While there are some divisive opinions around frame-gen software, it can be a game-changer for handheld gaming PCs. After all, AMD's AFMF 2 update did smooth out performance on the Asus ROG Ally X.
So, can FSR 4 do the same?
Based on AMD's benchmark data, FSR 4 upscaling alone offers significant improvements to the pure silicon performance of the RDNA 4 Radeon 9000 series GPUs across multiple games.
The Radeon 9000 series silicon offers 4K gaming speeds of 39-97 frames-per-second (fps), depending on the title. With FSR 4 upscaling, that range moves to 78-134 fps. With FSR 4 upscaling and frame generation enabled, the Radeon 9000 series offers frame rates of 144 to 233 fps for up to 3.7x performance gains.
Now, these benchmarks are provided by AMD, and are using the Radeon 9070 XT GPU with FSR 4 in "Performance" mode, in a test configuration with an AMD Ryzen 7 9800X3D, 32GB of DDR5 memory, and an MSI MEG x670E ACE motherboard. The games were run in 4K (3840 x 2160) at "Ultra," "Extreme," and "Maxed" presets.
While we tend to prefer independently verified performance benchmarks, these GPUs were just announced, so we'll need to wait until the review embargo lifts to see how they held up in testing by media outlets and analysts.
Based on these results, FSR 4 looks like a pretty solid update compared to previous versions of the software.
It's also interesting to compare the FSR 4 gains to Nvidia's RTX 50-series, as Nvidia's flagship GPU also struggles to game in 4K at 60fps with more demanding titles, relying on Nvidia's DLSS 4 frame-gen software to fill the gaps.
AMD announced the Ryzen Z2 series, a second-generation of gaming handheld chips, at CES earlier this year.
We've only seen one system with a second-generation Ryzen Z2 chip so far, the Lenovo Legion Go S, which houses a Ryzen Z2 Go processor. If you take a look at the Ryzen Z2 generation, the Z2 Go is the most budget-friendly processor, as it uses old Zen 3 CPU architecture, RNDA 2 GPU architecture, and houses just 4 CPU cores. So, the Z2 Go is not exactly the most thrilling.
The Ryzen Z2 and Z2 Extreme are far more impressive chipsets, though the Ryzen Z2 is very similar in terms of Zen 4 and RDNA 3 architecture compared to the Ryzen Z1 Extreme. However, software differences make a "night and day" difference between the Z2 and Z1 Extreme. While the Asus ROG Ally X will likely remain one of our favorite gaming handhelds, there's reason to look forward to a Ryzen Z2 series update.
If AMD can bring the RDNA 4 updates to FSR to older architecture, or if FSR 4 tech can influence an update to AMD's Fluid Motion Frames (AFMF) tech, that could help AMD level the playing field with Intel's Core Ultra 200V Lunar Lake. While Lunar Lake isn't a handheld chip, Intel does have internal teams dedicated to helping the handheld market, and the raw power of Lunar Lake's Intel Arc iGPU outperforms the Z1 Extreme as we saw on the MSI Claw 8 AI+.
While we don't benchmark with super sampling tech, we did witness Lunar Lake perform better than the RDNA 3.5 architecture of AMD's "Strix Point" Ryzen AI 300 series in a head-to-head demo of F1 24 at 1080p with medium settings.
FSR 4 is currently a Radeon 9000 series exclusive, but that may change in a few months if AMD can bring FSR 4 to older architecture.
We also could see FSR 4 become available on future AMD chips like the Ryzen AI 400 series and Ryzen Z3 generation if the software requires RDNA 4 architecture. Not that either chip has been announced or even hinted at by AMD, but the Ryzen AI 300 series will be a year old come Summer 2025, so we will likely see some AI PC chip updates from AMD later this year.
Regardless of personal feelings, frame-gen software can have a major impact on gaming experiences. Whether you choose to enable frame-gen is up to you.
I see a better case for FSR, DLSS, and XeSS when it comes to laptops and handhelds rather than desktops because of the hardware limitations of those form factors. But that's a matter of my personal opinion.
Valve's Steam Deck reminded me of what's most important in gaming — and it's not graphics
I tried gaming on the Lenovo Legion Go S and now I want to buy a Steam Deck instead
This ROG Ally and Steam Deck mount looks ridiculous — and its creators know it
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
Business Insider
an hour ago
- Business Insider
Sector Spotlight: Nvidia results highlight notable week for semiconductors
Welcome to the latest edition of 'Sector Spotlight,' where The Fly looks at a new industry every week and highlights its happenings. Confident Investing Starts Here: EARNINGS RECAP: Nvidia shares rose above $141 per share, up 5%, after reporting first quarter warnings on Wednesday. The company's Q1 beat was followed by inline Q2 guidance. Jensen Huang, founder and CEO of Nvidia, said: 'Our breakthrough Blackwell NVL72 AI supercomputer – a 'thinking machine' designed for reasoning- is now in full-scale production across system makers and cloud service providers. Global demand for NVIDIA's AI infrastructure is incredibly strong. AI inference token generation has surged tenfold in just one year, and as AI agents become mainstream, the demand for AI computing will accelerate. Countries around the world are recognizing AI as essential infrastructure – just like electricity and the internet – and NVIDIA stands at the center of this profound transformation.' Summit Insights upgraded Nvidia to Buy from Hold after its Q1 results and guidance. The risk of double-ordering on its Hopper generation AI GPU and the risk of China export controls are now priced into the stock, the analyst told investors. Summit also believes that the data center capex spending for the training market will remain robust and benefit Nvidia AI GP and data center networking businesses, the firm added. SECTOR NEWS: A jury determined following a trial in the U.S. District Court for the Western District of Texas that Intel's (INTC) 2012 licensing pact with Finjan covered VLSI's patents because the entities are affiliated under the control of asset management firm Fortress Investment, Bloomberg's Lauren Castle reported. The development sets up a potential escape for Intel from the more than $3B in patent-infringement verdicts against the company in the long-running dispute, the author noted. Nvidia (NVDA) disclosed in its quarterly filing that its CEO Jen-Hsun Huang adopted a trading arrangement to sell 6M of the company's shares between March 20 and December 31, 2025. Nvidia is also facing new accusations from U.S. lawmakers arguing the company is too close to China, criticism which may signal new challenges for the company, Olivia Beavers and Amrith Ramkumar of The Wall Street Journal wrote. Senator Jim Banks and Senator Elizabeth Warren recently wrote in a letter to CEO Jensen Huang that plans for an Nvidia facility in Shanghai risk giving China access to cutting edge technology. In an interview on CNBC's Mad Money, Jensen Huang said Nvidia's market share in China was 95% four years ago, but is now only 50%. He said it's critically important that AI is built out on an American technology stack, not Chinese. 'We want to be the preferred technology stack,' he noted. He plans to keep the dialogue going with the Trump Administration. He sees robotics as the next growth opportunity. 'We are an infrastructure company, not just a chips company,' he added. In a conference call following the company's Q1 results, Nvidia said it will consult the U.S. government if it develops new China chips and is considering options for China chips. The company added it has no new product for China data centers ready now and that limits are quite stringent on China chips. Nvidia contended that export rules will help foreign rivals and that it expects a meaningful decrease in China data center revenue. On Tuesday, Liam Mo and Fanny Potkin of Reuters reported that Nvidia is planning to launch a new AI chipset for China at a much lower price than its recently restricted H20 model and plans to begin mass production as early as June. The graphics processing unit, GPU, will be part of Nvidia's Blackwell-architecture AI processors and is expected to cost between $6,500-$8,000. Nvidia's suppliers, including Foxconn, Inventec, Dell (DELL) and Wistron, have made a series of breakthroughs that have allowed them to accelerate production of its flagship AI data center 'racks' and start shipments of the highly anticipated 'Blackwell' AI servers, according to Financial Times' Eleanor Olcott and Michael Acton, citing several people familiar with the matter. TSMC (TSM) executive Kevin Zhang said the company is still weighing whether it plans to use ASML 's (ASML) high numerical aperture, or NA, machines for its future process nodes, Reuters' Nathan Vifflin wrote. 'A14, the enhancement I talk about, is very substantial without using High-NA. So our technology team continues to find a way to extend the life of current (Low-NA EUV machines) by harvesting the scaling benefit,' Zhang said at a press briefing. 'As long as they continue to find a way, obviously we don't have to use it,' he added. The report noted that Intel has planned to use the high-NA machine in its future manufacturing process. A Swedish business consortium and Nvidia unveiled plans to build new AI infrastructure with Nvidia accelerated computing, networking and software in Sweden to transform and prepare the country's industries for the age of AI. Swedish industry giants AstraZeneca (AZN), Ericsson (ERIC), Saab (SAABF), SEB, in partnership with Wallenberg Investments, will build the system that will be operated by a joint company to offer secure, sovereign compute access to the industry partners. The intention is that the first phase of the deployment will be two Nvidia DGX SuperPODs featuring Nvidia's latest generation Grace Blackwell GB300 systems, making it the largest enterprise AI supercomputer in Sweden once operational. It will be used to run compute-heavy AI workloads to speed up processes such as training of domain specific AI models and large-scale inference, including reasoning AI. Oracle (ORCL) will spend around $40B on Nvidia's high-performance computer chips to power OpenAI's new giant U.S. data center, according to Financial Times. Oracle will purchase around 400,000 of Nvidia's GB200 chips, its latest 'superchip' for training and running AI systems, and lease the computing power to OpenAI, according to FT's Tabby Kinder and George Hammond, citing several people familiar with the matter. ANALYST COMMENTARY: Citi analyst Atif Malik raised the firm's price target on Nvidia to $180 from $150 and keeps a Buy rating on the shares. The company reported April quarter results in-line and issued a July quarter sales outlook $1B above Citi's expectations, 'clearing the final hurdle of the China H20 ban transition quarter,' the analyst tells investors in a research note. Moreover, Blackwell sales of $24B topped Citi's $20B expectations and management maintained mid 70's gross margins target on improving Blackwell profitability with no major tariff impact, the firm contends. Citi believes that with margins expanding, Nvidia shares should 'break its range bound trend' since mid-last year and will likely make a fresh 52 week high. Needham keeps a Buy rating and $54 price target on Semtech (SMTC) after its 'solid' Q1 results driven by data center related infrastructure shipments and LoRa-enabled solutions. Demand for data center related products remains robust as the company reported a record quarter, while the momentum in Data Center is expected to carry through the second half of FY26, driven by continued strength in FiberEdge and non-NVIDIA CopperEdge and LPO ramps, the analyst tells investors in a research note. Redburn Atlantic analyst Mike Harrison initiated coverage of Broadcom (AVGO) with a Buy rating and $301 price target. The company is arguably the 'pre-eminent' application-specific integrated circuits co-partner with a strong pipeline of future customers, the analyst tells investors in a research note. Further, the firm believes the company's networking products are prevalent through the different types of networking within artificial intelligence data centers, from the compute fabric to the back-end network, as well as the interconnects that permeate throughout. The consensus is not giving Broadcom sufficient credit for the strength of its ASICs pipeline, contended Redburn. Stifel noted that Nvidia recently published a blog post providing details on its new 800V high voltage DC power architecture for next-gen AI data centers and specifically named Infineon (IFNNY), Monolithic Power (MPWR), Navitas Semiconductor (NVTS), Rohm, STMicroelectronics (STM) and Texas Instruments (TXN) as power IC/solutions supplier partners. While three of the semiconductor companies mentioned have issued press releases in conjunction with Nvidia – namely Infineon, Navitas and TI – the firm noted there were few details beyond what was discussed in Nvidia's blog post. However, it expects all of the companies called out by Nvidia to benefit from the accelerating increase in power requirements for AI data centers that are expected to require megawatt-scale racks by 2027.
Business Insider
2 hours ago
- Business Insider
TSMC Stock Maintains High-Conviction AI Play Status With $220 Price Target
Taiwan Semiconductor Manufacturing Company (TSM), or TSMC for short, the world's numero uno pure-play semiconductor foundry, has surged nearly 20% over the past month. Despite this impressive rally, my outlook remains bullish. Confident Investing Starts Here: The company is well-positioned to benefit from powerful tailwinds, including relentless AI-driven demand, a dominant—nearly monopolistic—position in advanced chip manufacturing, ongoing geographic diversification, and a steady cadence of technological breakthroughs. Unveiling Next-generation Technologies in the Coming Years TSMC continues to be the principal manufacturer and primary supplier of chips for leading AI powerhouses, including Nvidia (NVDA), Advanced Micro Devices (AMD), Apple (AAPL), and Qualcomm (QCOM). Notably, TSM is also the producer of Nvidia's cutting-edge Blackwell chip series, which has recently gained immense popularity. TSMC's 3nm process currently represents the most advanced semiconductor technology in the industry, delivering superior power efficiency and performance. Looking ahead, anticipation is building around the company's upcoming 2nm and 1.6nm nodes, scheduled for launch in late 2025 and 2026, respectively. The 2nm technology, referred to as N2, remains on track for volume production in the second half of 2025. This next-generation process is expected to deliver a 10–15% improvement in processing speed at the same power consumption, or a 20–30% reduction in power usage at equivalent performance. Following that, the 1.6nm process is projected to further improve power efficiency by an additional 15–20% over the 2nm node. These advancements are especially timely, as data centers grapple with rising energy costs. The shift to more power-efficient chips is becoming not only a technological imperative but also an economic necessity. This positions TSMC as a key enabler in the ongoing global semiconductor upgrade cycle. Reflecting this momentum, TSMC has outlined strong long-term growth expectations. The company projects its AI-related chip revenue to grow at a compound annual growth rate (CAGR) of 45% over the next five years, while overall revenue is forecast to grow at a 20% CAGR during the same period. These figures underscore the company's pivotal role in powering the future of computing. Such growth potential hasn't gone unnoticed. Famous investor Cathie Wood 's Ark funds recently purchased 241,047 shares of Taiwan Semiconductor, worth $46.3 million, signaling firm institutional conviction. TSM Has Unrivaled Market Leadership TSM commands an overwhelming 64.9% global market share in the foundry segment, dwarfing its closest competitors— Samsung Electronics (SMSN) and Intel (INTC), according to Statista. Its unparalleled scale, deep client relationships, and technological edge create formidable barriers to entry. This dominant position grants TSM significant pricing power. Clients, many of whom have relied on TSM for decades, are unlikely to switch suppliers given the risk of falling behind in the rapidly evolving AI race. One of the major risk factors associated with TSM is its geographical location in Taiwan, which poses the risk of a takeover by China. However, TSM is prioritizing geographic diversification as a core strategic initiative, proactively addressing these risks by diversifying its global manufacturing footprint. In addition to its $65 billion investment in U.S.-based fabs, TSM has committed a further $100 billion to expand capacity globally. Its Arizona facility is reportedly operating at full capacity through 2027, highlighting robust demand. This geographic diversification not only reduces exposure to potential tariffs but also strengthens TSM's resilience to geopolitical volatility. Beyond the U.S., TSM is also establishing a new chip design center in Munich, Germany, as well as a manufacturing plant in Dresden, Germany, and new fabs in Japan. Taiwan Semiconductor's strategic investments beyond its home base reflect prudent risk management and reinforce its strategic shift toward a more balanced global presence. Pre-Earnings Numbers Point to Positive Outlook for Upcoming Earnings TSM is set to report its Q2 2025 earnings on July 17. The company is projected to report earnings per share (EPS) of $2.30, representing a 57.5% year-over-year increase. Additionally, Q2 revenues are projected to increase by 13% year-over-year, ranging between $28.4 billion and $29.2 billion, driven by high demand for its advanced 3-nanometer (nm) and 5-nm processes. On May 9, TSM unveiled impressive revenue figures for April 2025, marking the highest ever figure in any single month in the company's history. April net revenues galloped 48.1% year-over-year to 349.6 billion New Taiwan dollars (approximately $11.6 billion). It's worth noting that TSM anticipates 24% to 26% sales growth for FY2025, driven by strong demand for its latest nanochips amid the AI surge. Importantly, TSM trades at an attractive valuation compared to its peers. In terms of its valuation, TSM looks cheap. Currently, it's trading at an attractive forward P/E ratio of 21x, compared to much higher multiples of its peer group. Semiconductor company Advanced Micro Devices is trading at a higher forward P/E multiple (28x), while the AI prodigy Nvidia is trading at a forward P/E of 32x. Is TSMC a Buy, Sell, or Hold? On Wall Street, TSM stock carries a Strong Buy consensus rating based on seven Buy, one Hold, and zero Sell ratings over the past three months. TSM's average stock price target of $219.43 implies approximately 11% upside potential over the next twelve months. The semiconductor industry continues to experience strong growth, driven largely by the rapid adoption of artificial intelligence technologies. TSMC, with its unmatched manufacturing capabilities, deeply embedded customer relationships, and advanced technology roadmap, is exceptionally well-positioned to capitalize on this transformative trend. The company's timely investments in next-generation nodes—specifically 2nm and 1.6nm—alongside its global manufacturing expansion, align well with rising demand fueled by a broad upgrade cycle across the tech landscape. These factors collectively make TSMC a compelling long-term investment opportunity. With a favorable valuation and strong earnings momentum, the current environment presents an attractive entry point for investors looking to gain exposure to the accelerating AI megatrend.
Yahoo
2 hours ago
- Yahoo
China's first 6nm domestic GPU with purported RTX 4060-like performance has powered on
When you buy through links on our articles, Future and its syndication partners may earn a commission. Lisuan Technology, a Chinese graphics card startup, has announced via the company's official WeChat account that its forthcoming G100 graphics card has successfully powered on, marking a significant milestone in its deployment. The G100 purports to be China's first domestic 6nm graphics card. As China embarked on its journey towards technological independence, a wave of industry veterans joined the gold rush. Founded in 2021, Lisuan Technology is among the youngest startups in the graphics card sector, alongside Moore Threads (2020) and Biren (2019). Lisuan Technology has considerable backing, as it was reportedly established by industry veterans boasting more than 25 years of experience in Silicon Valley. The same can be said for Moore Threads, which was founded by Zhang Jianzhong, the former vice-president and general manager of Nvidia China. Little information is available regarding the G100, besides its use of Lisuan Technology's proprietary TrueGPU architecture. In contrast to some Chinese firms that license intellectual property (IP) from sources like Imagination, TrueGPU asserts that it is an in-house architecture developed from the ground up. Lisuan Technology previously stated that the G100 is created using a 6nm process node but did not reveal the manufacturer. Due to U.S. export restrictions, China cannot access the 6nm node, ruling out Samsung and TSMC as options. As a result, it is likely that the Chinese foundry SMIC is responsible for producing the silicon using its 6nm manufacturing process, which is also implemented for Huawei's latest Ascend 920 AI chip. With limited information, we can only rely on rumors regarding the specifications of G100. For example, it is claimed that the G100 provides performance similar to the GeForce RTX 4060. This claim generates significant skepticism, as the GeForce RTX 4060, despite being a last-generation product, is still regarded as one of the best graphics cards available; we have yet to see a Chinese-made graphics card rival it. Additionally, the G100 is rumored to feature ample memory and modest power consumption. The G100 reportedly supports popular APIs such as DirectX 12, Vulkan 1.3, OpenGL 4.6, and OpenGL 3.0, suggesting that G100 could be a decent gaming graphics card. Work on the G100 started in 2021, with Lisuan Technology originally aiming for a 2023 launch. However, financial difficulties obstructed these plans, and by 2024, the company neared bankruptcy. To support the struggling startup, Dongxin Semiconductor, its parent company, provided a substantial financial boost of $27.7 million, enabling continued development of the G100. Lisuan Technology has successfully obtained the first G100 chips from the foundry, and they are operational. The outcomes seem to meet the startup's expectations. As a result, the company has moved forward with software and hardware validation as well as driver optimization. Clearly, the G100 has considerable progress ahead before reaching the retail market. It is reportedly in the tape-out phase and is currently undergoing risk trial production. Completing a 6nm tape-out requires substantial time and investment, indicating that Lisuan Technology is at a pivotal point in G100's development. Lisuan Technology intends to deliver small quantities of G100 in the third quarter of this year. Nonetheless, given the timeline, mass production likely won't happen until 2026. Targeting the performance of the GeForce RTX 4060 isn't bad; however, the G100 needs to function as a reliable graphics card right from the start. It's unreasonable to expect Lisuan Technology's first attempt to compete with the likes of Nvidia, AMD, or even Intel. Creating a good graphics card from scratch demands considerable time and effort. Moore Threads has demonstrated that the software aspect is just as crucial as the hardware, given that new driver updates can significantly boost performance. We might see the first benchmarks for the G100 before the end of the year. Follow Tom's Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.
