Technology

The IoT and AI are a likely partnership: IoT generates and captures data, often in large volume, AI is ideally placed to analyze it.

Combined, AIoT presents new opportunities, so much so that Transforma Insights forecasts no fewer than 9.1 billion AIoT connections at the end of 2033, a more than six-fold increase in 10 years.

The potential for AI in the IoT is far-reaching, and one standout application is enhanced security.

The security risk to the IoT

All connected devices are under growing levels of threat, but the IoT is particularly targeted. According to Beaming’s cyberthreat report into UK businesses, IoT devices were most frequently attacked in 2024. They are attractive targets for the data they exchange and their potential to be compromised.

The devices are often unmanned and generally sit outside corporate security perimeters. They may be in remote spots, where they could be subject to unauthorized physical access attempts, and often remain in place for long periods of time. Many, such as the IoT devices used in energy, transport, utilities and retail, transfer sensitive data of high value.

Businesses need confidence that data collected through the IoT—both real-time and historical—comes from secure and trusted sources, not least when it comes to developing and training AI models. In this, the IoT works with digital twins, which are digital representations of physical objects or systems.

The IoT enables the seamless flow of real-world data between the physical and the digital, while the digital twin’s attributes provide the features for AI modelling. Historical data collected through IoT is then used to train and refine the AI model.

How AI is helping secure the IoT

AI applies its automation and analytical capabilities to many tasks and priorities. It is making inroads into cybersecurity, something that has not gone unnoticed. Last year, the IEEE revealed almost half of the global technology leaders it surveyed (47%) expect vulnerability identification and attack prevention to be a top use of AI in 2026.

That may be some comfort to enterprises and connectivity and solutions providers grappling with the problem of protecting IoT devices and applications. This challenge is compounded by the fact that attackers are increasingly using AI themselves to automate phishing, accelerate reconnaissance and develop adaptive malware that evades traditional detection. However, it is more comforting still, that AI is already making a difference in the IoT, redefining how organizations protect their devices.

AI-powered anomaly and threat detection (ATD) is helping security teams identify threats like suspicious network traffic and botnet activity faster and improving resilience across large-scale IoT environments, something enterprises must strive for.

In the recent past, the focus on the IoT was arguably getting devices online. That is no longer the challenge; the test now is keeping them operational: compliance readiness and the flexibility to adapt to ever-evolving commercial and technological changes. It also means maintaining resilience.

IoT security must follow a clear defend against, detect and react approach to swiftly counter attacks. No one or two of these three measures are enough without the others.

How AI improves IoT visibility and incident response

AI-powered ATD detects anomalous behavior, such as remote code execution, abnormal port connection or a suspicious IP. These could indicate the beginnings of a cyberattack on an IoT device. It analyses the anomaly and can identify the attack type, be it distributed denial of service (DDoS), man-in-the-middle (MiTM) or an attempted device takeover.

ATD can then trigger direct action, if business rules dictate an automated response. This could take the form of threat isolation or referring the incident for full review.

Anomaly and threat detection protects over one million devices

ATD runs entirely in the mobile core network infrastructure, rather than through software agents on a device, so it can be retrofitted to existing systems.

Enterprises that have identified IP backdoors and Mirai botnet infections within hours.

With IP backdoors, ATD detects unusual outbound connections, or traffic, to suspicious IPs. Such backdoors may allow remote control or data exfiltration, both of which leave identifiable behavioral traces.

In the case of Mirai, anomalous behavior typically exhibits as spikes in outbound traffic, uncommon ports use, or repetitive scanning of external IPs. ATD can flag these irregularities in real time and trigger corrective actions, such as blocking or quarantining the device, blocking or throttling traffic or patching firmware.

Automation and analytics can shape the next phase of IoT security

There is a clear shift in IoT implementation and management. It is insufficient to plan for device deployment, sit back and gather the data. Without a strategy that accounts for the stresses, threats and changes that beset IoT estates, enterprises risk costly surprises like unplanned site visits and service disruptions.

AI, through automation and analytics, can shape the next phase of IoT security. Enterprises that detect, analyze and even automatically address, anomalous activity reduce the risk of cyberattack-related outages and inconvenient site visits to access devices.

Sending field technicians to maintain or repair devices can add significantly to total cost of ownership. Each truck roll, which incurs expenses for labor, fuel, vehicle wear and often missed productivity opportunities, can add up to over $1000 per site visit.

IoT downtime, meanwhile disrupts operations and can have a catastrophic reputational, as well as financial, cost.

How to balance innovation, data privacy and operational control

Enterprises are, for the most part, keen to innovate through AI but have understandable questions about data privacy and operational control.

It is important to know what AI does, in all process integrations, to understand why it does it and to have control that prevents AI deviating from its purpose.

On data privacy, ATD isn’t installed on IoT devices. Only packet headers from device cloud communications are mirrored from the mobile core to the ATD engine, with threat levels and AI-driven insights relayed through a customer portal.

Operational control is maintained through the business rules that dictate how the ATD engine reacts. The option to refer an anomaly for review, for example, gives enterprises the flexibility to incorporate human oversight, under predetermined circumstances.

This is especially useful when you consider there can be genuine reasons why a SIM may increase or cease communication, that an incident reviewer will understand.

AI-powered IoT security

AI is making a difference to the speed, efficiency and depth of response to cyberthreats. Automation and advanced analytics within IoT solutions’ security measures also help enterprises manage costs, by minimizing labor-intensive manual tasks and site visits, and reducing the risk of expensive cyberattack reparations.

For CISOs, CIOs, product and operations managers seeking to maximize IoT value and protect their enterprise IT domains from external threats, AI-powered ATD offers visibility and actionable insights to take IoT security to the next level.

We've reviewed and ranked the best antivirus software.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

Is AI taking our jobs?

Even as Artificial Intelligence displaces a significant number of roles in the United Kingdom, it’s clear the net-positive impact on the job market remains favorable.

A new study estimates that evolution in AI technologies could create as many as 610,000 new jobs by 2028.

What’s more, employees can look forward to not just more jobs but better ones; an analysis of over 500 million job advertisements from 15 high-income countries revealed that employers in the U.K. were willing to pay 14% more to workers with AI skills.

People with AI skills also have an expanding choice of roles, as organizations in every industry look to hire data scientists, machine learning engineers, computer vision specialists, AI trainers, human-machine teaming managers, AI ethicists and compliance officers and more.

AI is increasing the job market

In the highly digitalized U.K. financial services sector, 75% of organizations are already leveraging AI and another 10% plan to adopt within a few years. The wide range of AI use-cases spanning process optimization, cybersecurity, fraud detection, customer support, credit scoring and compliance management etc.is opening up a variety of opportunities for job seekers with relevant skills.

20% of education sector professionals in the U.K. are using AI in their current role to improve learning content, motivate students and accommodate their different needs, or even to complete administrative tasks. As roles combining technology and education evolve, there will be a need for skilled learning experience designers to create effective learning platforms, content creator managers for ensuring educational content meets quality standards, and specialists who leverage AI tools to create and manage online communities for learners.

Even in healthcare, where human expertise and in-person interaction are highly valued, there’s a huge need for AI professionals, in the near future, to build algorithms that can generate personalized treatment plans and predictive analytics models that identify disease patterns and risk factors. With telemedicine becoming common, the healthcare industry will require staff who can operate AI-enabled remote patient monitoring systems.

The use of smart health assistants, AI-based medical imaging and intelligent hospital workflow optimization tools will create a demand for medical professionals with varying degrees of AI capability and also for medical AI specialists, telehealth coordinators, and data analysts for managing and processing massive volumes of health data.

But where are the skills?

In a recent UK survey, 68% of IT decision makers said that insufficient skills and expertise was their top challenge in implementing AI projects, more problematic than integrating disparate data and fragmented systems. Training is the obvious answer, but it appears that 19 percent of the U.K. workforce – 6.31 million employees – have not been trained in how to use AI tools at work.

Organizations need to take urgent measures to prepare their employees to deal with the change that AI will bring to their work and workplaces. Besides training employees in functional AI skills that enhance productivity, employers should build capabilities in communication, empathy and problem finding to optimize the performance of human-machine teams.

Training with and for AI

Traditional training methods, involving scheduled classroom-based lessons with standard curricula, are ineffective for AI technologies, which are evolving at a blistering pace. Organizations need to establish an environment of continuous, lifelong learning where employees can learn new skills on the job, throughout their careers.

AI-powered learning platforms support this by providing rich training materials that employees can access from anywhere, and at any time, to learn what they want, when they want it. Importantly, training also addresses cultural barriers to AI adoption, such as apprehension about learning new technologies.

Next Steps

To successfully bridge the AI skills gap, employers should follow best practices, including integrating AI literacy in different functions to prepare employees across the enterprise for a future in AI and partnering with educational institutions to offer learning programs aligned with business and market needs.

The U.K. government can play a major role in building an AI-ready workforce by offering incentives to organizations investing in AI training and also providing internship opportunities for young people to gain practical experience in AI tools and technologies. This is also in the administration’s interest because AI has the potential to increase GDP by £550 billion over the next decade.

Improve your CV and your career with the best resume builders.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

For years, the artificial intelligence industry operated under a philosophy of rapid innovation. Whilst this benefits sectors like healthcare, the initial wave of optimism surrounding generative models is giving way to a challenging reality. Numerous national security agencies and research bodies are issuing warnings regarding the potential for these models to be weaponized.

At the heart of this warning is the realization that large language models do more than process text - they democratize technical knowledge and act as decision makers rather than tools. In the wrong hands, this capability can be applied to malicious cyber operations and the subversion of critical digital infrastructure.

Historically, executing sophisticated cyberattacks required years of specialized technical expertise in exploit development and network intrusion. Today, artificial intelligence bridges the knowledge gap for individuals who lack formal training but possess malicious intent.

Recent assessments by international security bodies, such as the Organization for the Prohibition of Chemical Weapons, highlight that the emergence of data driven molecular design models means that those with limited expertise can bypass monitoring from regulatory frameworks.

New frontier AI models such as Mythos could theoretically identify alternate synthetic pathways to design toxic chemicals using ordinary laboratory reagents.

Within enterprise environments, the exposure of software supply chains and cloud infrastructure also remains an acute vulnerability. Third party applications account for a high percentage of emerging security risks, and modern enterprise operations rely on a web of digital service providers and data aggregators that were barely visible a short time ago.

Every vendor represents a potential point of entry, and a single compromised credential at a small third party service provider can grant an attacker the freedom of lateral movement within a corporate or government network.

Current public skepticism is a response to the lack of transparency in how these frontier models are trained, monitored, and integrated. Companies developing models have a responsibility to ensure that their innovations do not compromise the stability of public systems.

This requires a commitment to the responsible deployment of these tools, prioritizing national security and architectural resilience over speed to market. The industry must move away from generic statements and focus on explicit, verifiable security practices.

A new framework for technological visibility

Organizations must adopt a rigorous approach to machine visibility and network defense. Traditional perimeter focused security tools are insufficient - defensive structures must shift toward continuous internal monitoring.

This means analyzing east west traffic within an organization network to scrutinize communications between systems and understand normal data flow patterns. Anomalies must become immediately apparent so security operations teams can act before an intrusion escalates.

Response times must also collapse. The traditional multi week window that attackers enjoy gives automated threats too much leeway to cause damage. Modern network detection and response platforms shrink attacker dwell times by identifying suspicious machine behavior in real time.

Because systems prefer structured layouts and consistent schemas, defender tools must leverage network telemetry to track how these models interact with internal data stores. Security teams need to see exactly how data is being processed, ensuring that unauthorized models are not mapping corporate assets.

Governments are responding to this reality with updated legislation, such as the strengthening of national cybersecurity laws in the UK. These updates expand the scope of statutory regulations to include essential digital service providers, managed service providers, and data centers.

Tougher penalties raise the cost of non compliance, and mandatory incident reporting requires organizations to alert regulators within tight windows, often 24 hours. These legislative changes acknowledge what technical experts have warned about for years - that cybersecurity breaches on critical infrastructure are a national security threat.

Breaches and automated attempts at exploitation are inevitable. The industry must treat advanced software infrastructure with the same level of caution as critical physical assets. In a world where automated systems can orchestrate complex network intrusions, the move towards more comprehensive security measures is essential.

We've featured the best endpoint protection software.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

If I'm going to watch yet another movie or TV show about a mafia-style gang in New York — because why bother in the wake of The Godfather and The Sopranos, right? — I want to come away from it having learned something new.

New MGM+ crime drama The Westies achieved this at conception, because I didn't know the first thing about the real-life Irish-American crime family of the same name. In fact, I didn't even know that they existed. But sure enough, they were known for racketeering, drug trafficking, and contract killing in Hell's Kitchen from 1960s-1980s.

I know what you're thinking: isn't Hell's Kitchen exactly where the Italian-American mob operated from in New York City. You'd be right in thinking that, and it's one of the main sources of dramatic tension across the show's eight-episode run.

Add in the fact that an FBI team has been tasked with tracking down our fictional version, led by family patriarch Eamonn Sweeney (J.K. Simmons), and you've got a proper old-school cat-and-mouse chase.

Interestingly, the biggest pro and con of The Westies is almost the same thing. While you're streaming it, you're engrossed — the storytelling itself is solid, dependable and superficially fun. But from the moment you switch it off, it will be completely erased from your mind.

Fans complained The Westies would be too tame just from the trailer — and I think they were right

After the above trailer for The Westies was released, I couldn't help but notice that the overwhelming fan response was to criticize scenes for being "too tame" compared to what mob behavior would have been like at the time.

Now that I've watched the entire season, I have to agree. Even if you don't know the story of the Westies, you're tuning into a gangster crime show expecting a severe level of violence — or even just an overarching sense of nasty brutality.

Of course, blood, guts and gore don't make a story any better objectively, but come on... we're trying to have some fun here. On top of that, it's difficult to invest in something new if you feel as though it is holding back, especially if it's aiming for a second season and beyond.

Still, the core storytelling is both engaging and well constructed. As Eamonn Sweeney's cronies face off against the local Italian-Americans, you're as engrossed in the inevitably fatal outcome as much as you are the complicated family dynamics.

The cast is stronger than its story

(Image credit: MGM+)

Bosch star Titus Welliver plays our ragged FBI lead Glenn Keenan, with an 80s "pornstache" so commanding that it might as well have had its own acting credit. When you can get past the facial accessories, you know you're in good hands... once Harry Bosch, always Harry Bosch, if you get what I mean.

If anything, I wish we had more of Simmons himself on screen. Sweeney isn't exactly the Don Corleone type, but it feels as though Simmons has somehow been a mobster boss in a former life. If you're selling him as number one on the call sheet, let's see that translated in each episode, please.

Really, I'm looking for holes when there aren't any — or at least I'm ripping tiny damaged flecks in the story's fabric into gaping tears. Because we've been so spoiled by shows such as Peaky Blinders and MobLand in recent years, we have to hold everything that comes after to an incredibly high standard.

Truthfully, The Westies just doesn't meet it. Not because there's anything wrong per se, but because the competition is so tough. It's kind of like studying really hard for an exam knowing you're going to get creamed by the kids who are naturally better at the same subject. Thems the breaks, so just enjoy the ride while it lasts.

The Westies begins airing on MGM+ from July 12.

• Zero Parades: For Dead Spies now has a free to play TTRPG
• It adapts the game's setting and mechanics for pen and paper
• It also comes with a free premade one shot to help you learn the game

Let’s face it, 2026 — with its component price hikes, sweeping layoffs, cancellations, expensive AAA titles, and Sony’s promise to stop making physical games — has been something of a year of Ls. But one of the few lights in this darkness, and game of the year contender, Zero Parades: For Dead Spies just announced brand new content is dropping: it’s a tabletop version of its gameplay, and it’s totally free.

If you head over to one-shot site you’ll find everything you need to get started including a simple core rulebook and a One Shot story set in the Zero Parades world 30 years before the events of the game.

The One Shot comes with everything you need to play your first session including location maps, premade characters, GM materials (to help whoever is running things) and adapted rules that bring the game’s mechanics to life in TTRPG form.

Scratching the itch

Anyone who has played Zero Parades: For Dead Spies or its spiritual predecessor Disco Elysium will have already felt the TTRPG inspired elements — all challenges in the games are already presented as dice rolls.

Just like in Zero Parades the pen and paper game has the same stats and Faculties — Action, Relation, Intellect — which determine how capable your character should be at certain actions, as well as details on how you can exert yourself to increase your odds of success — at the cost of raising your Fatigue, Anxiety, or Delirium respectively.

You also find items of clothing you can equip to make your stats better (and sometimes also worse), or tools which unlock actions like a camera allowing you to snap photos to use later — or to remember a sick spy moment you just pulled off, the decision is all yours.

As a major fan of both TTRPGs like Dungeons and Dragons and of Zero Parades I had been thinking of adapting the digital gameplay into something pen and paper for my regular playgroup. With this free release I don’t need to do any work at all — though I love there are the resources to create your own adventure and stories if you want to move away form the premade One Shot.

(Image credit: ZA/UM)

Zero Parades: For Dead Spies is everything you’d want from a follow-up to Disco Elysium. It keeps the post-Soviet vibes and aesthetics with a whole new story and setting. The gameplay is very similar too, with a few updates that fine-tune mechanics in a way that feels better and yet utterly familiar.

There’s also that general and gradually growing weirdness that made Disco Elysium so much of a joy to play through.

So many people I know who loved Disco Elysium had no clue Zero Parades is a thing they can play, and the perfect game to scratch the itch Disco Elysium leaves you with.

Plus if you’re reading this article the day it comes out the full Zero Parades: For Dead Spies game is 10%-off for everyone, and 20% off for owners of Disco Elysium in the Steam Summer Sale. What are you waiting for? Go and play it!

• The EU's General Safety Regulation is set to be expanded
• Driver Distraction Warning systems will need to be fitted to new cars
• Privacy advocates are already voicing data concerns

As of this week, the European Commission has stated that all new vehicles registered in the European Union will have to be fitted with Advanced Driver Distraction Warning (ADDW) systems to comply with new, more stringent safety legislation.

While European roads are among the safest in the world, the European Commission said in its announcement that "the number of deaths and injuries from road accidents is still too high." As a result, it is set to expand its General Safety Regulation to include several new requirements.

Among the new requirements are advanced emergency braking systems that detect pedestrians and cyclists, as well as "better forward vision" and camera-based driver distraction technology that audibly warns drivers to pay attention to the road.

Many new cars already feature this technology, which uses a small camera mounted behind the steering wheel or somewhere above the infotainment system in the center of the dashboard.

Constantly active, the system monitors the driver's eyes and facial expressions, intervening if it detects that the driver is looking away from the road for too long.

(Image credit: DS Automobiles)

Today's manufacturers' systems range from audible bleeps and on-screen safety notifications when it detects distraction, to the deactivation of automated cruise control and other functions.

Critics have already hit out at the decision, with the European Conservative describing the commission's move as the "latest annoying piece of EU overregulation," which arrives with little clarification on how the data will be handled.

These ADDW systems should operate in a closed-loop system, where all data is processed locally in the car and never uploaded to any third-party server.

However, since April 2018, all newly approved passenger cars and light vans in the EU have been required to include an eCall emergency system that automatically contacts emergency services in the event of a serious crash.

This, coupled with the fact that consulting firm McKinsey predicts that 95% of all vehicles will be connected to the internet by 2030, means it isn't a huge stretch to think that driver-monitoring data could eventually leave the vehicle.

Analysis: a data privacy nightmare

(Image credit: Volkswagen)

Analysis by Mozilla in 2023 examined the privacy policies of 25 car brands and found that every single one failed to meet its own privacy and security standards. Mozilla concluded that cars were "the worst product category we have ever reviewed for privacy."

In 2024, the Texas Attorney General opened an investigation into several automakers following reports that they had been collecting large amounts of driver data and selling it to third parties.

As the European Conservative points out, there is currently very little clarity from the EU over how driver data is handled, meaning there is the potential for these systems to eventually help determine insurance premiums or act as evidence in a court of law, for example.

Beyond this, ADDW technology can also prove extremely distracting when implemented poorly.

I've lost count of how many times a system has warned me to pay attention when all I'm doing is trying to find a simple function in a baffling series of touchscreen menus.

On the one hand, Euro NCAP says it will target "annoying" safety tech, but on the other, the European Union is mandating greater reliance on these systems. Now is a confusing time to be a motorist.

Asus ProArt RTX 5090: Two-minute review

The Asus ProArt GeForce RTX 5090 32GB GDDR7 OC Edition is a specialised flagship GPU built for a niche audience and isn’t going to appeal to those looking for a premium upgrade for a gaming build. It’s aimed at creators, AI users and compact workstation builders who want full RTX 5090 performance without the bulk of larger 3.5 or 4 slot coolers.

The ProArt uses the normal RTX 5090 spec: 21,760 CUDA cores, 32GB of 28Gbps GDDR7 on a 512-bit bus and PCIe 5.0 support. The card has a 2482MHz default clock and a 2512MHz OC mode, and a 1000W power supply is recommended.

The 32GB of VRAM helps when rendering, running larger AI models or heavily modded 4K games would exceed the memory available on lower-tier cards. The ports — HDMI 2.1b, DisplayPort 2.1b and USB-C — give excellent monitor setup flexibility.

The 304 x 140 x 50mm, 2.5-slot SFF-ready design is a key selling point, as most flagship RTX 5090 cards are much chunkier, while the slimmer ProArt leaves more room for nearby expansion cards, side-panel clearance and power-cable routing.

Installation is straightforward: a 1-to-4 adapter is included and the 16-pin socket on the card is angled back for easy cable routing, plus the included adjustable GPU support can be used to help prevent sag.

Build quality is top-notch, with no obvious flex, and the GPU avoids gamer-style RGB lighting in favour of a subdued creator/workstation look with brown wood-patterned laminate trim and subtle ProArt gold markings.

Cooling is provided by two large Axial-tech fans, a vapour chamber, liquid metal on the GPU and a double-vented backplate. In OC mode, the GPU peaked at 76°C and memory at 86°C, with a 36dB peak sound level, which is reasonable for a slimmer RTX 5090 cooler. Unfortunately there was noticeable coil whine under heavy loads, and at times it was louder than the fans.

Standard mode was about 5% cooler and 6% quieter, but most importantly, barely slower, while Quiet mode was about 7% cooler and 6% quieter than OC mode, but gave around 2% less performance.

placeholder image (Image credit: Future)

Overall performance is excellent, as expected from an overclocked RTX 5090. OC mode was about 3% faster than a stock RTX 5090, around 45% to 50% ahead of the RTX 5080 in compute loads, and up to 60% in gaming. The improvement is most obvious at 4K, in ray tracing and compute-heavy workloads.

The coil whine is frustrating given this is a card designed for creators, though it’s not so bad that you’d hear it outside a well-designed closed case.

Otherwise, the ProArt RTX 5090 OC Edition does a difficult job well. Yes, it’s expensive and specialised, but its slimmer design, quiet fans, 32GB of GDDR7 and top-end performance make it a capable choice for creators and high-end builders who can use what it offers.

That said, unless you can find it at a compelling price, it faces stiff competition from slim RTX 5090 cards that use AIO water cooling.

Asus ProArt RTX 5090: Price & availability

placeholder image (Image credit: Future)

• How much is it? It retails for about $4,099 / £3,799 / AU$6,499
• When can you get it? The Asus ProArt RTX 5090 is available now
• Where is it available? You can buy it in the US, UK, and Australia

The Asus ProArt RTX 5090 is available now, retailing for about $4,099 / £3,799 / AU$6,499 depending on the region and retailer. That makes it around the typical price for a premium RTX 5090, though it sits above many cheaper air-cooled cards with the same spec.

While the ProArt would always be an expensive card, AI data centre demand has really pushed up prices, so it’s a tough sell unless you really need the features.

The premium price mostly gets you the compact ProArt design rather than a big performance uplift. You still get the full RTX 5090 spec, a mild factory overclock, 32GB of GDDR7, USB-C display output, a slimmer 2.5-slot cooler and a three-year warranty.

The main competition is from other RTX 5090 cards, and if you don’t need the SFF-ready form factor, cheaper air-cooled models will get you very similar core performance for less, while larger premium or all-in-one (AIO) water-cooled RTX 5090 cards may offer lower temperatures, less noise or more overclocking headroom for similar money.

If the compact creator-focused GPU is perfect for your next build, the ProArt may be a compelling buy. If you just want the cheapest route to RTX 5090 performance, wait for the price to come down a little, or look at alternatives.

• Value: 3.5 / 5

Asus ProArt RTX 5090: Specs

(Image credit: Future)

• Full RTX 5090 spec and 32GB of GDDR7
• Slimmer SFF-ready 2.5-slot design

Asus ProArt RTX 5090 vs stock RTX 5090 vs RTX 5080

Asus ProArt RTX 5090

Stock RTX 5090

RTX 5080

Process Node

TSMC 4N

TSMC 4N

TSMC 4N

Transistor Count (Billion)

92.2

92.2

45.6

CUDA cores

21,760

21,760

10,752

RT cores

170

170

84

Tensor Cores

680

680

336

Render Output Units

176

176

112

Cache (MB)

96

96

64

Boost Clock (MHz)

2,482 default / 2,512 OC

2,410

2,620

Memory Speed (Gbps)

28

28

30

Memory Type

GDDR7

GDDR7

GDDR7

Memory Pool (GB)

32

32

16

Memory Interface (bits)

512

512

256

Memory Bandwidth (GB/s)

1,792

1,792

960

PCIe Interface

5.0 x16

5.0 x16

5.0 x16

TBP (W)

575

575

360

Recommended PSU (W)

1,000

1,000

850

Power Connector

1 x 16-pin, 1-to-4 adapter

1 x 16-pin, 1-to-4 adapter

1 x 16-pin, 1-to-3 adapter

• Specs: 4 / 5

Asus ProArt RTX 5090: DesignFutureFutureFutureFuture

• Slimmer than most flagship RTX 5090 cards
• SFF-ready

The ProArt design is the main reason to choose this particular RTX 5090 over a more conventional option. Asus has kept the card to 304 x 140 x 50mm (12.0 x 5.5 x 2.0in) and a 2.5-slot thickness, which is much easier to work with than the very large 3.5-slot and 4-slot RTX 5090 designs.

It’s also SFF-ready, which means it fits Nvidia’s size guidelines for small-form-factor enthusiast builds rather than just being smaller than average. That doesn’t make it a tiny GPU, but it does mean the ProArt is built to work in more compact cases that support SFF-ready graphics cards, as well as larger systems where you want more room for other expansion cards or easier airflow.

Visually, it avoids the usual gaming-card look. The brown wood-patterned laminate trim and small ProArt gold details give it a more subdued workstation feel, and there’s no RGB lighting to worry about.

Cooling is handled by two large Axial-tech fans, a vapor chamber, liquid metal on the GPU and a double-vented backplate. That’s a decent cooling setup for a thinner RTX 5090, though the slimmer cooler does mean it has less heatsink volume than some of the much larger cards.

FutureFuture

The card includes a dual BIOS switch for Performance and Quiet modes, a solid backplate and an adjustable GPU holder in the box. The 16-pin power connector is angled back toward the end of the card, which helps cable routing, but you’ll still need to be careful with bend clearance, especially in smaller cases.

Overall, the design feels practical, premium and nicely understated, but it only really makes sense if you specifically need a thinner RTX 5090.

FutureFutureFutureFuture

• Design: 4.5 / 5

Asus ProArt RTX 5090: Performance

placeholder image (Image credit: Future)

• About 3% faster than a stock RTX 5090
• Excellent cooling and low fan noise
• Very noticeable coil whine

The Asus ProArt RTX 5090 is at its best when you look beyond gaming and treat it as a compact high-end creator card. In Geekbench 6 Compute and Blender, it performs at the level expected for a tuned RTX 5090, with small but useful gains over the stock RTX 5090.

The gap is much bigger against older or lower-tier cards. Across the compute tests, the ProArt was about 50% ahead of the RTX 5080 on average, and around 40% ahead of the RTX 4090. The RTX 4090 is still a very capable GPU, but the ProArt’s combination of 32GB of faster GDDR7 and Blackwell architecture gives it a clear advantage in compute workloads and makes it a worthwhile upgrade.

While not the focus, gaming is still excellent, as you’d expect from an RTX 5090. Cyberpunk 2077 reached 98fps at 4K using Ultra Ray Tracing and balanced scaling, while Black Myth: Wukong hits 88fps at 4K with ray tracing, Cinematic settings and 50% upscaling. So while not the point of the ProArt, it’s a very capable gaming card.

FutureFutureFuture

Thermals are quite good for a slim RTX 5090. The ProArt peaked at 75°C, which is much warmer than larger 5090s with aftermarket cooling, which typically run about 10°C cooler. The VRAM reached 86°C, which is again higher than most 5090s but not in any way problematic.

Fan noise was also very low, with the ProArt peaking at just 36dB in my testing. That’s about the same as 5090s with big coolers.

All in all, Asus has done a good job tuning the ProArt to keep performance high and fan noise under control, even if that means letting temperatures sit a little higher than some larger cards. That trade-off is perfectly fine for a 2.5-slot RTX 5090, and the GPU and VRAM temperatures stay well within reasonable limits.

The catch is coil whine, which is an electrical buzz caused by components vibrating under heavy power load, rather than the fans themselves. In fact, most of the time it's louder than the fans themselves.

Some coil whine isn’t unusual on very high-power graphics cards, but it’s still disappointing on a premium creator-focused model where low noise is a key selling point.

Benchmark results

Asus ProArt RTX 5090

Stock RTX 5090

RTX 5080 OC

Synthetic Benchmarks

3DMark Steel Nomad

14,440

14,430

8,743

3DMark Speed Way

14,366

14,560

9,435

3DMark Port Royal

38,501

36,310

22,021

Compute benchmarks

GeekBench 6 Compute (OpenGL)

391,245

383,000

275,612

GeekBench 6 Compute (Vulkan)

408,212

355,000

259,633

Blender (aggregate score)

4,551

4,440

3,001

Thermals / noise

Peak GPU temp

76°C

72°C

67°C

Peak VRAM temp

86°C

89°C

70°C

Idle temp

43°C

35°C

45°C

Peak sound level

36dB

50dB

38dB

Gaming (average fps)

Cyberpunk 2077 - 4K

98

94

50

Black Myth: Wukong 4K

88

84

55

Cyberpunk 2077: Ultra Ray Tracing, balanced scaling
Black Myth: Wukong: Ray Tracing, Cinematic, 50% upscaling

• Performance: 4 / 5

Should you buy the Asus ProArt RTX 5090?Asus ProArt RTX 5090 Scorecard

Category

Notes

Score

Value

Very expensive but on par with other premium RTX 5090s.

3.5 / 5

Specs

A 5090 GPU with 32GB of GDDR7, USB-C display output and a mild OC mode.

4 / 5

Design

The 2.5-slot build and understated styling suit compact creator PCs.

4.5 / 5

Performance

Solid performance overall, with low fan noise and temps kept under control. The coil whine is frustrating though.

4 / 5

Final score

A capable specialised RTX 5090 for compact high-end builds, but the high price and coil whine keep it from being an easy recommendation for everyone.

4 / 5

Buy the Asus ProArt RTX 5090 if...

You want RTX 5090 performance in a slimmer build
The ProArt is easier to fit than many oversized flagship cards, while still giving you the mighty RTX 5090 GPU and 32GB of GDDR7.

You want a creator-friendly card
The subdued design, USB-C display output and included GPU holder make it a better match for a compact workstation or high-end creator PC.

Don't buy it if...

You just want the best-value RTX 5090
Cheaper RTX 5090 cards will get you very similar performance if you don’t need the ProArt size or features like the USB-C output.

Coil whine will bother you
Fan noise was low, but coil whine was very noticeable under heavy load, so it’s best in a case that insulates sound well.

How I tested Asus ProArt RTX 5090

• I spent about a week testing the Asus ProArt RTX 5090
• I ran synthetic, compute and gaming benchmarks
• I compared it against the stock RTX 5090, 4090 and the Asus ROG Astral RTX 5080 OC

Test System Specs

Here are the specs on the system I used for testing:

Motherboard: MSI Z890 Tomahawk Wi-Fi
CPU: Intel Core Ultra 285K
CPU Cooler: Corsair Titan 360 RX
RAM: G.Skill Trident Z5 DDR5 (2 x 16GB)
SSD: Samsung 9100 Pro
PSU: Corsair HX1000
Case: Thermaltake Core P3 TG Pro

I spent about a week with the Asus ProArt RTX 5090, testing it in OC mode and comparing it against other high-end cards. The main comparisons were stock RTX 5090 and RTX 4090 performance and the Asus ROG Astral RTX 5080 OC.

I used 3DMark for synthetic performance, Geekbench 6 Compute and Blender for creator workloads, and games including Cyberpunk 2077 and Black Myth: Wukong for real-world gaming results.

• Originally reviewed June 2026