Insomniac Games’ Marvel’s Wolverine is the hot new PS5 game right now, but that’s all it is — there’s no PC version in the works, and given Sony‘s reported decision not to bring single-player games to PC, that seems unlikely to change anytime soon. However, that doesn’t mean PC gamers are out in the cold after it was confirmed that an experimental PS5 emulator has been shown to be able to run the early portions Marvel’s Wolverine, though only at single-digit frame rates.
While the KytyPS5 emulator was already able to start the game up, that was all it was able to do. That was great if you just wanted to be able to move around the game’s menus, but it wasn’t much use beyond a proof of concept. Now, things have progressed considerably, with VideoCardz pointing to an X social media post showing the emulator rendering the game’s cinematics and gameplay.
The latest build of KytyPS5 can indeed play the game, but it does so at a single-digit frame rate. That’s a long way short of making the action-packed title playable on anything other than a PS5 or PS5 Pro. But it’s a start, and a much more promising one than was the case previously.
Latest Videos FromTom’s Hardware
Wolverine and Ghost of Yotei showing signs of life on PCIronic that Sony picked the absolute worst time to stop supporting PCAnyways, I guess we’ll see where things are in one month pic.twitter.com/bqljKpdZcsSeptember 29, 2026
The frame rate situation probably says more about the emulator than the game itself. KytyPS5 is still very much in the early stages of its development, which means that issues are to be expected — especially for a title that has only been available for a few weeks. The emulator’s developers are also focused on making it boot more games rather than making a handful work flawlessly, so it’s unlikely it will be ready for prime time for a little while yet.
None of this is to detract from the work that has been done to get this point, though. The rapid progress that Marvel’s Wolverine has seen is promising for the future of PS5 emulation on PC. KytyPS5 uses Vulkan 1.3 for its graphics output and uses an in-house system to decode the PS5’s RDNA 2 GPU instructions. It’s also available for both Windows and Linux right now, with an even more experimental version in the works for macOS.
This news also comes as another PS5 emulator has reached something of a milestone of its own. The developers behind SharpEmu, another experimental PS5 emulator, have been able to run a number of PS5 games at a playable 60 FPS. But even then, the success is modest —the developers only got six of 55 games to that point. A further six reached the point of gameplay, albeit at lower frame rates.
President Donald Trump, several Cabinet members, and House Speaker Mike Johnson sat down with key AI leaders for high-level discussions on AI, as recent incidents such as the publicized Hugging Face hack have brought forth issues about AI safety. The meeting resulted in a document titled “Joint Commitment on Frontier Responsibilities,” which states that “every company is responsible for developing its own technology safely,” and was signed by Sundar Pichai (Alphabet), Dario Amodei (Anthropic), Mark Zuckerberg (Meta), Greg Brockman (OpenAI), Elon Musk (SpaceXAI), Jensen Huang (Nvidia), and Trump himself. According to CNN, the president argued that this was the best way to move forward with AI, which he dubbed “Super Intelligence,” as it balances progress with safety.
Image 1 of 2
(Image credit: President Donald Trump/Truth Social)
(Image credit: President Donald Trump/Truth Social)
The pledge outlines four key points: (1) Implement robust internal controls to monitor the capabilities and alignment of its models during training and deployment around areas like cybersecurity, biosecurity, and chemical threats, and to ensure that its models do not hack or access technical systems in unintended ways; (2) Empower an internal team to ensure all of the controls, monitoring, and detection are operating as intended, and that any issues are remediated; (3) Partner with an independent external auditor or evaluator to carry out independent assessments of whether the controls, monitoring, and detection are operating as intended; and (4) Designate an independent committee of the board of directors to oversee and receive reports from the teams operating the controls and the internal and external auditors and evaluators, as well as to ensure any issues identified are remediated.
“I’m seeing tremendous self-policing, and they understand that they have to self-police,” Trump told reporters outside the West Wing. “These models are very complex and, again, I believe they’re going to be used for the good.” While this pledge has no law or regulatory pressure behind it, the president still called it “morally binding,” and that he believes that the AI companies can monitor themselves without the need for additional regulation.
On the other hand, Trump dismissed their concerns as hoaxes and called it a “sick conspiracy,” while Jensen Huang said that their fears are a “distraction,” and that AI labs should be accountable for any damage that their rogue agents may make. Chinese state media also said that move was “a response to Chinese competition,” and that “a global AI-safety framework that excludes China is not quite global.” Still, it mentioned the need to monitor AI development as humans could lose control of the technology.
This meeting apparently put most of the U.S. AI industry’s leadership and the Trump administration on the same page when it comes to AI safety. While this is just a pledge and not legally binding in any way, it has at least got the heads of the biggest AI labs and policymakers in Washington discussing the potential unintended consequences of AI development. How clashes with the pledge will be remediated, and the timeline in which that happens, remains an open question.
A pre-launch Geekbench 7 result, shared on X by INIYSA as one of OpenAI’s dots (it’s answer to Meta’s Muse), shows a multi-core score of 9,435, with indications of the hardware used by the new autonomous agent. The VM hardware appears to be a single nine-core slice of an AMD EPYC 9V74 with 9.73GB of memory, run on Ubuntu for the leak and Debian for post-launch runs. Our check of Geekbench 7’s public database found six runs on that configuration uploaded since launch, ranging from 1,512 to 1,614 single-core and 8,135 to 8,991 multi-core. Dots, powered by GPT-6 Astra, launched at OpenAI’s DevDay on Tuesday for Pro and Business Premium users.
OpenAI bills the autonomous agents as having “their own cloud computer, their own browser, and the apps you’ve connected” in its introduction. This is not a new concept: Cursor says its cloud agents get their own virtual machines; SpaceXAI says every Grok Bot shares one persistent cloud computer; and Meta’s Muse gives each agent a sandbox.
Maddie D. Reese posted on X on launch day the apps her dot said came preloaded. The lineup includes Chromium, Blender, GIMP, Inkscape, Kdenlive, Godot, FreeCAD, OpenSCAD, KiCad, QGIS, ParaView, 3D Slicer, with Python, Node.js, and Git. In the same reply, the dot said its cloud computer runs “Debian Linux.”
Fun note from Tibo at the closing q&a of DevDay:dots get their own Linux cloud computer, and that cloud computer comes preloaded with apps like Blender!I checked with my dot to get a fuller list of the preloaded apps and tools pic.twitter.com/JIt5VQiV2jSeptember 29, 2026
The post went up the evening of launch day, captioned “OpenAI dots Geekbench 7 Score,” with a screenshot showing a single-core score of 1,667 and multi-core of 9,435. The result in the screenshot shows that it was uploaded Sept. 25, or four days before the launch. The 9,435 is almost 5% higher than the post-launch best and about 10% higher than the median. The pre-launch run is potentially a bit of an outlier, but performance remains within the same ballpark. It is uncertain who ran the original benchmark.
Latest Videos FromTom’s Hardware
Tom’s Hardware previously reported that rival Meta’s Muse runs on AMD EPYC Turin with two cores and 8GB per sandbox. The 10 Muse runs in Geekbench’s database have median scores of about 1,041 single-core and 1,394 multi-core. The dot’s post-launch runs, with median scores of 1,570 and about 8,550, put it at about 1.5x the single-core and 6x the multi-core score.
Some of this gap is accounted for by the nine-core vs. two-core configuration; the rest of the gap likely comes from a difference in clock speed, partially offset by Muse’s newer architecture. Geekbench lists a 2.6GHz base clock for the dot’s EPYC 9V74 against 1.5GHz for Muse’s EPYC 9D25.
Muse’s runs also report 7.75GB of memory. Our independent run for this article came back with a single-core score of 1,000 and a multi-core score of 1,433, in line with the other Muse runs.
Data center CPU demand for always-on agents has surged as high-core-count CPUs are needed at sustained load. It is unknown whether the dots, as tested, hold their cores while idle. Such agents are rising in popularity: Muse reportedly passed 500,000 daily active users, and SpaceXAI plans standalone Nvidia Vera CPUs for Grok’s agentic workloads. It may only be a matter of time before a new CPU arms race begins. OpenAI has already said that, in the future, users will be able to add more dots and “scale the output of each dot by … increasing its speed.”
Get Tom’s Hardware’s best news and in-depth reviews, straight to your inbox.
Last year we moved into a new luxury rental building that advertised lots of amenities, including a pool and parking garage, but they don’t exist and the super told me the building can’t structurally support them. Management recently sent us a lease renewal with a $500 rent hike. Can we push back on the increase?
In New York, it’s atypical for leases to make mention of amenities, but according to Altagracia Pierre-Outerbridge, attorney and founder of Outerbridge Law representing residential tenants, condo owners and landlords, that’s the first place to start to determine your options.
“You could actually be entitled to a reduction in rent if the lease explicitly states the provision of certain amenities but fails to provide them,” Pierre-Outerbridge said. “The building would be in breach of contract.”
Where to start
If your lease does include amenities, your rent would in theory apply toward amenities, and therefore should be reduced proportionally, since they aren’t provided in practice. “But pricing out the exact reduction in rent for the lack of amenities is another thing,” Pierre-Outerbridge said. “You may want to speak with a lawyer who has handled cases like this that can advise on whether the reduction offered to you is commensurate with the lacking amenities.”
Unfortunately, most leases do not mention building amenities, so in your case, negotiating a rent decrease would be an uphill battle.
“We’re seeing a lot more fraudulent advertising in the New York rental market,” Pierre-Outerbridge said. “AI is making it all the more frustrating. People will jump on a listing that looks great in photos and rush through the viewing so that they won’t lose out on the apartment without confirming the amenities that got their attention are there.”
Further, what your lease probably does include is what’s known as a merger clause. “Luxury buildings typically include a merger clause in their leases,” Pierre-Outerbridge said. “A merger clause basically excludes any other written or verbal agreements and limits the agreement to the contract itself. So a landlord is not beholden to any promises made in an oral conversation, or in this case, an online advertisement.”
Take it up a notch
That doesn’t mean you’re out of options. “Just because the contract isn’t on your side doesn’t mean you can’t fight the rent increase,” Pierre-Outerbridge said. “Talk to your neighbors. They are probably just as upset and facing similar rental increases, and you can make more noise as a collective.”
Pierre-Outerbridge suggested writing to your landlord on behalf of the group to try to negotiate a smaller increase. “It’s easy for a landlord to stiff-arm one person, but that gets harder and harder to do to a group of tenants,” Pierre-Outerbridge said. “It’s a shame how many landlords try to inflate their listings and refuse to deliver on their promises, so don’t be afraid to let them know exactly how angry you are.”
If you aren’t sure where to start, a lawyer can help you figure out the best way to present your argument to your landlord on behalf of you and your fellow tenants.
“In all likelihood, your goal is for your rent not to go up at all,” Pierre-Outerbridge said. “But a $500 monthly rent increase, especially when they’ve misrepresented the building, is ludicrous. By pushing back and getting others involved, you can begin to assert some leverage.”
Altagracia Pierre-Outerbridge, Esq. is the owner of Outerbridge Law P.C, focusing primarily on tenant representation. The firm represents all sides in landlord-tenant litigation and transactional matters such as month-to-month holdovers, nuisance cases, licensee cases, harassment claims, repair cases, tenant buyouts, succession claims, DHCR overcharges and rent reductions and more. Pierre-Outerbridge has 15 years of experience litigating in Supreme, DHCR, and Housing Court. To submit a question for this column, click here. To contact Outerbridge Law P.C. directly, call 212-364-5612 or 877-OUTERBRIDGE, or schedule a meeting today.
Half the buyers out there are scared home prices are about to crash. The other half are hoping they will.
A recent survey from Clever found 58% of Gen Z buyers are actually rooting for a crash, just so homeownership feels within reach.
So, what do the forecasts actually say?
Every quarter, Fannie Mae surveys more than 100 housing experts on where prices are headed. The newest results are in. And spoiler alert: they’re not calling for a crash – not even the pessimists.
What the Newest Numbers Actually Say
The panel’s latest forecast has prices climbing every single year through at least 2030.
The panel’s average forecast is that prices will rise by 14.7% in the next 5 years. And here’s where it gets really interesting. If you split these experts into optimists and pessimists, even the pessimists still expect prices to increase about 6.6% by the end of 2030 (see graph below):
The takeaway? If you’ve been waiting for prices to fall, you may be waiting a while.
One thing to keep in mind though – these are national numbers. Prices in your area could run a little hotter or a little cooler than this, so it helps to know what’s happening locally, too. But the big picture is prices aren’t crashing. Historically prices usually rise.
How This Quarter Compares to the Past
Here’s something you probably don’t realize. This survey runs 4 times every year, so you can track the panel’s mood over time.
A year ago, the panel expected home prices to grow 2.1% this year. Now they’re forecasting 2.5%. That means the near-term outlook actually got more optimistic. But that’s only part of the story. The years after that shifted, too (see graph below):
Zoom out to 2027 through 2029 and the mood has cooled a bit. Each of those years is now expected to see a little less growth than the panel thought a year ago. That’s likely a reflection of where we are right now with everything that’s impacting the housing market.
But again, the overall takeaway here is every bar shows an increase in prices – the size of that increase has just moderated due to some of the factors at play.
A slower climb isn’t a bad thing, though. It’s a sign the market is settling into a more normal pace after a few wild years.
A little more growth here, a little less growth there. What hasn’t budged once is the idea that home prices will keep growing.
What It Means for Your Next Move
Now, percentages are great, but you probably care more about the actual dollars and cents of your move, so let’s graph that out, too.
Run the numbers on a $400,000 home bought in January, and the panel’s latest forecast puts you up about $58,000 in equity in 5 years just from price growth (see graph below):
That’s real wealth you could be building while others sit on the sidelines, waiting for a crash the experts don’t see coming. And with prices expected to keep rising, waiting could mean paying more for the same home later.
Bottom Line
Whether you’re bracing for a crash or hoping for one, the verdict is the same – prices are still expected to rise, not fall. Talk with a local real estate agent about what that means for your market and your plans.
President Donald Trump signed a directive telling federal agencies to start calling AI “Super Intelligence” after hosting a lunch at the White House with industry leaders discussing the technology and the risks that it carried. According to the executive order, the term artificial intelligence was born in the United States 70 years ago, and the capabilities achieved by AI today are much more advanced than what the original innovators envisioned. Because of this, he saw fit to have the federal government call that technology by the new term that he coined instead of referring to it as “artificial intelligence.”
“The terminology used by the Federal Government should reflect the transformative capabilities of these technologies and the limitless opportunities they create for the American people. Accordingly, the term ‘Super Intelligence’ more appropriately captures the promise, potential, and rapidly advancing capabilities of these technologies,” the order said. “It is therefore the policy of my Administration that, to the maximum extent permitted by law, the executive branch shall use the terms ‘Super Intelligence’ and ‘SI’ in place of ‘Artificial Intelligence’ and ‘AI’ and will not acknowledge the usage of ‘Artificial Intelligence’ and ‘AI’ in any applicable setting.”
Using the term “Super Intelligence” could also bring some confusion, as it’s the term that scientists use for a level of AI that has achieved cognitive superiority to humans and are better at them in every conceivable task. While the latest frontier models are extremely good at doing particular actions, they still are nowhere near the level of superintelligence described by experts. While the White House wants to win the AI race with China and has been pushing hard for the development of this technology, some industry leaders and politicians are concerned that an artificial superintelligence could get out of control and end humanity. Senator Bernie Sanders (I-Vt.), alongside Senator Greg Cezar (D-Texas), have even proposed a bill banning its development, with the included penalty at par with the illegal development of nuclear weapons.
Still, it seems that some of the executives of the biggest AI tech companies are following the President’s lead, with Elon Musk adopting the term while in Trump’s presence. As for the President, he told reporters that the change has officially been made “by the biggest, the smartest, the greatest people anywhere in the world at this,” likely referring to the AI executives who attended the lunch at the White House. “We’re using the word super because the other word is a fake word.”
In late August, Architect Labs claimed it had designed a chip that was almost entirely developed by AI, an industry-first achievement. AI is already used to optimize floorplans, placement and routing, verification, and other stages of semiconductor development. Generative AI can assist engineers with RTL code, whereas emerging agentic systems can operate electronic design automation (EDA) tools, analyze results, identify problems, modify designs, and repeat the process with increasingly less human intervention. Meanwhile, human engineers still define and develop architectures and make fundamental design decisions that determine what a chip does and how it works.
This creates a curious feedback loop. Today’s AI models run on processors designed by human engineers with growing assistance from AI; those models can then help design more capable processors for the next generation of AI systems. As EDA vendors and semiconductor companies give AI control over progressively larger portions of the design process, the industry is gradually moving from humans using AI tools to design chips toward AI systems participating in the design of the hardware on which their successors will run.
Can machines indeed design machines today? Probably not. But will they be able to do so in the future? That’s a big question with important ramifications — and engineers have been pondering it for longer than you might think.
Latest Videos FromTom’s Hardware
A brief history of AI in chip development
Cadence, Synopsys, and Siemens EDA, all leading developers of EDA software, alongside Ansys — a leading designer of simulation software — rolled out AI-enhanced versions of their tools in the early 2020s, before the generative AI boom took the world by storm.
(Image credit: Synopsys)
The first generation of AI-enhanced EDA software primarily used machine learning (ML) and reinforcement learning (RL) to tackle well-defined optimization problems. Given an existing design, a set of constraints, and particular targets, these tools could explore numerous implementation options to optimize placement and routing, and therefore power, performance, and area (PPA), while shrinking development time. Essentially, AI could find a better way to implement a design, but the design itself and its goals were still defined by engineers.
One of the key advantages of these tools was their ability to learn from previous runs and use accumulated data to guide subsequent design-space exploration, something which could reduce the number of iterations needed to meet PPA targets and, in some cases, produce results that would have required considerably more engineering time using conventional methods. Yet the autonomy of these tools is limited, as engineers define constraints, configure flows, run individual tools, analyze their output, and decide what to try next. AI accelerates or optimizes particular stages of chip development, but humans still control the overall design flow. And that’s changing with newer generations.
The latest AI-enhanced EDA tools are considerably more ambitious. Generative AI can write or modify RTL and verification code, analyze reports, identify potential points of failure, and even suggest fixes. Meanwhile, emerging agentic systems can operate multiple EDA tools and execute sequences of engineering tasks with very limited human intervention. Such an agent can analyze results, modify a design or its parameters, launch another simulation or implementation run, evaluate the outcome, and repeat the process until it reaches specified targets. As a result, AI is gradually moving from optimizing individual steps inside EDA tools to automating parts of the chip development workflow itself.
From big bang to architectures
In 2023 – 2024, both Cadence and Synopsys announced that hundreds of chip designs have been completed using their AI-enhanced Cadence.ai DSO.ai/VSO.ai/TSO.ai tools. Moreover, leading high-tech companies revealed details about how they used AI to complete their projects. Yet putting Google, Nvidia, OpenAI, and Architect Labs into the same bucket is not right, as they represent different degrees of AI involvement in chip development.
(Image credit: Google Deepmind)
Google, which was among the first high-tech giants to announce the use of AI to develop its AI accelerators, seems to be the least radical. Google’s AlphaChip uses reinforcement learning mostly for physical floorplanning: it places circuit blocks and optimizes layouts, but it does not invent the entire design or the architecture itself. Google DeepMind said in 2024 that AlphaChip had been used for the previous three generations of TPUs, meaning Google was well ahead of the general EDA industry with its tools. (The company has not shared AlphaChip’s progress in detail since then.)
Nvidia is more interesting because its internal AI systems tend to automate work traditionally performed by hardware engineers. The important distinction is that Nvidia trains specialized models on its own RTL, documentation, and unique accumulated engineering knowledge, something a merchant EDA vendor cannot access. This makes Nvidia an example of a chip designer that turns the engineering data behind its proprietary GPU and, more recently, AI accelerator, CPU, DPU, and network cards into training data for AI. Nonetheless, Nvidia itself draws a clear line between its automation and autonomous chip design.
OpenAI’s Jalapeño is probably the strongest real-silicon example we currently have. OpenAI says AI was directly involved in implementation, design-space exploration, verification loops, and arithmetic-circuit optimization, which enabled the company to go from initial design to tapeout in nine months. At Hot Chips, OpenAI disclosed some concrete PPA advantages: Compared with human baselines, AI-assisted designs improved a BF16 multiplier by 56%, an FP4 dot-product block by 21%, and an FP32 accumulator by 10%, while reducing the area of matrix and SIMD units by 10% and 8%, respectively.
OpenAI has not said that AI invented Jalapeño’s architecture, so humans have indeed remained responsible for it. But AI was extensively used to turn that architecture into silicon and optimize it … which brings us to Architect Labs.
Indeed, Architect Labs’ Redwood is qualitatively different. Two human architects defined the specification, after which Architect Labs says its AI autonomously generated and verified the logic, generated RTL, verified it, and produced firmware. To top it off, AI also assisted software development, including drivers and kernels. Meanwhile, unlike Google, Nvidia, and OpenAI, the company explicitly describes its system as one that performs machine learning co-design and says it can explore architectures, though without elaboration.
Redwood was ‘produced’ in under two weeks and runs billion-plus-parameter models, but there is a huge caveat: It has not been taped out. It has never actually been produced as an ASIC; the accelerator runs on an FPGA, and the claimed 3.4X performance-per-watt advantage over Nvidia’s Jetson Orin Nano is based on a projected Samsung 8LPP design.
Final words
Just a few years ago, AI in semiconductor development was largely an optimization technology that helped engineers find better ways to implement designs created by humans. Today, it can generate RTL, verify designs, operate EDA tools, and even explore architectural choices, although humans still define what ultimately gets built.
We are therefore considerably closer to machines designing machines, but nowhere near the singularity quite yet. For now, chips are helping design their successors. But they aren’t even close to deciding what those successors should be.
For many of us, the tech we carry to keep our other devices powered while on the go doesn’t get much consideration. Buy a power bank with enough ports, capacity, and output, and you’re good to go, right? And generally, that is right, though a few small things can make a big difference. Of the dozen-plus chargers we’ve tested so far, a couple take four to six hours to charge, which may be fine for some, but could be a deal breaker for anyone who needs a quicker turnaround on a fully charged power bank.
We’ve compiled 13 power banks, ranging from slim, single-port 5,000 mAh units outputting 22W to monster multi-port 25,000 mAh bricks capable of delivering 220W across multiple ports. Prices range from around $24.99 for the Cuktech CP12 (10K/20W) to $129.99 for the most powerful unit we’re testing, the high-end Anker Prime (A110B) 20K/220W. Some include integrated USB-C cables; some include a removable USB-C cable; and some include nothing else in the box.
Others have fancy displays that show charging and temperature information, while some use simple LEDs that flash or change color to show charging status. Looks and size matter, too. The designs of the banks we tested range from simple grey boxes barely thicker than a deck of playing cards to larger stand-up units, and even see-through chassis.
Latest Videos FromTom’s Hardware
What’s right for you depends on your needs, budget, and to some degree taste. In general, the higher the output and capacity, the more the device costs. How many ports do you want or need? Do you need to fly with it? Capacity and output are, of course, key deciding factors. But price, size, and safety features also matter. So there’s a lot to consider if you want the best power bank for you, and there are plenty of feature, design, and performance differences between our tested devices that we’ll share below to help with your buying decision.
What makes a power bank good or bad?
What makes a power bank good or bad?
The answer depends on who’s buying the device and how they plan to use it. For example, if you need more than one charging port or an integrated cable, certain models just won’t work for you. Make sure it supports modern charging standards like USB-PD (Power Delivery) and/or Qualcomm Quick Charge so it can fast-charge your device (all tested devices below have PD3.0 or above).
Having the proper output to fast-charge your device is another important consideration. Buying a 10K, 25W charger for your Asus Zenbook Duo, for example, isn’t the best idea, since the device can use around 90W (@ 20V) to fast-charge and run. For that, you’d want something with more listed capacity and more wattage output. If you’re only buying a bank for your phone, that same 10K/25W charger can easily give your phone or a small tablet lots of extra run time away from a wall charger.
Rated capacity versus actual capacity is another issue, but marketing materials don’t clearly state it. The main thing to know is that the rating you see on the box is based on the battery’s 3.7V rating. So the actual capacity varies depending on the voltage your device needs. The higher the voltage and output used for charging, the lower the actual capacity is. With 20V laptop charging, usable capacity can drop to around 20% of listed capacity, whereas with simple 5V charging it’s closer to 60% or more.
Get Tom’s Hardware’s best news and in-depth reviews, straight to your inbox.
(Image credit: Future)
Another worthwhile feature for power banks is pass-through charging, or the ability to charge the bank and output power to your device(s) at the same time. Most do support this feature, though we couldn’t find anything concrete on the Cuktech CP12 or Baseus PicoGo Air (though every current Baseus PicoGo series does support it). Belkin’s Ultracharge limits output to 15W while charging, however.
Name-brand devices from Anker, Sharge, Cuktech, Belkin, and Baseus, like the ones we’re covering here, typically have multiple safety features, including protection against overcharging, over-current/voltage, over-temperature, and more, to prevent premature device failure (or worse). You’ll get an assortment of these features from popular brands, but be careful with off-brand ones, as they may not support some (or any) of them, and could therefore be a hazard. If you see a no-name power bank, make sure it is at least UL/CE (or CCC if in China) certified and has some or all of these safety features before buying.
Test Equipment and Setup
Test Equipment and Setup
To gather our power bank testing data, we need several items to ensure we’re getting the right information. Our test setup includes a wall charger to charge the empty power bank, an Asus Zenbook Duo and an Anker portable power station to charge and get output from, a USB tester that shows the output voltage, amperage, and watts, along with the charging protocol, mAh, and Wh, and an IR thermometer to record the device’s surface temperature. We also need a high-quality cable that won’t limit the output (we chose one rated at 240W). To test the cable, we used a USB Cable Checker2 to verify its capability. Below is a list of all the parts we use in power bank testing.
Image 1 of 1
(Image credit: Future)
How We Test
How We Test
Before we plug the device in, we inspect build quality (does it feel like cheap plastic or metal), size, weight, and port count/type. We also look at capacity and output ratings, fast-charging capabilities (think PD, etc.), and any certifications (UL/CE) or safety features. From this information, we can get a good idea of the pros and cons, including cost/mAh, actual versus listed capacity, port count, temperature, protections, and more.
For testing, we first make sure the power banks are fully discharged before we begin. Next, we take the chassis temperature with the IR thermometer and connect the power bank to the Satechi desktop charger via a certified 240W USB-C cable, along with the Power-Z device. From there, we record the time to fully charge the device and the maximum sustained input wattage (which, as you’ll see, tends to drop due to temperature or at a point in charging where it slowly lowers the amperage/wattage, and sometimes voltage). We also record the maximum chassis temperature using the IR Thermometer.
(Image credit: Future)
After fully charging a power bank, the next step is to discharge it using a constant load. In this case, we use the Asus Zenbook Duo as a real-world load up to ~93 W while running Cinebench for the CPU and FurMark to stress the integrated Intel Arc B390 graphics. This isn’t your basic budget laptop. At over $2,200, it packs an Intel Core Ultra X9 388H (5.1 GHz turbo, up to 80W), dual 14-inch Lumina Pro OLED touchscreens, and a 99 Wh charger.
Here, we test maximum and minimum output in watts and determine watt-hours (Wh) and milliamp-hours (mAh), with data gathered from the Power-Z device while discharging. We also record temperature readings at the start and at the maximum.
Once we have that data, we can determine the device’s capacity efficiency, at least for testing the laptop and higher-voltage charges. As we mentioned earlier, mAh ratings on packages are inflated by voltage-conversion losses and are based on the internal battery cells’ 3.7V voltage. The higher the voltage (and, of course, amperage/watts), the less mAh capacity a battery bank provides.
In our testing, the Zenbook Duo accepts 20V and up to 93W when it can, which is a worst-case scenario for smaller-output chargers, normal for higher-output chargers, and typical for laptops and higher-powered devices. If you’re only charging a smartphone or small tablet at lower volts, the value will be closer to the listed specifications. But it will differ when your smartphone charges at a lower voltage, often 9V for fast charging. Remember, output wattage (and voltage) determines charging speed, and for some people (or in some instances), that matters more than capacity.
After testing all of the power banks, we compiled all the data into the charts below.
Tested Power Banks
Tested Power Banks
We captured some images of all the chargers during the test below. We also included a simple chart listing all the power banks tested, along with capacity, wattage, port/count type, safety features, and the current price (at the time of publication).
What surprised me most was how long a couple of these devices took to completely charge from empty. Cuktech’s CP12 and CP24 models and the Baseus PicoGo AC22 took over 2.5, 4, and 6 hours, respectively, to reach 100%. This is partly due to lower input wattage and/or how the device accepts power over time. Whether it’s the voltage used, temperatures and potential throttling, or battery health and safety features, each handles the activity a bit differently. Others, like the Anker Prime or Shargegeek 170, offer much higher output and charge even higher-capacity devices much faster. In fact, the Shargegeek took under an hour to fill its 24K mAh coffers, accepting the most power (almost 140W peak) from our charger. Again, charging may not be a major concern for many people, but it’s good to know how long to expect your bank to reach full, since some are notably faster than others at the same capacity.
Temperatures didn’t play as obvious a role as they did in our charger testing. Lithium batteries perform best between 20 and 30 degrees Celsius, and the hottest device we covered was the Baseus PicoGo, which reached a peak chassis temperature of 43 degrees. The coolest of the bunch were the Anker Prime and Anker Laptop chargers, peaking at 37 degrees externally. These are all warm to the touch, but not hot — not something I would be particularly concerned about leaving out on a desk or table, in your purse or backpack, or even just sitting on the couch or a bed while charging your device(s).
But of course, it’s always good to keep charging devices somewhere you can see them regularly. Because batteries can, of course, be dangerous when things go very wrong, and there are at least two batteries cycling when you’re using a power bank.
Discharging didn’t bring too many surprises. A rare couple, like the Shargegeek 170 and Cuktech P23, maintained output throughout the entire discharge tests, while most others dropped voltage, amperage/wattage, or both. This matches how many devices charge: they use a lot of power/voltage initially, then drop to lower values to protect battery health (or due to thermal throttling).
We see a lot of ads about charging X device to Y% in Z minutes, and the higher the sustained output (assuming your device can take it), the faster it charges. For those that did drop, the differences were somewhere between 33% and 50% of the peak output.
Image 1 of 4
(Image credit: Future)
(Image credit: Future)
(Image credit: Future)
(Image credit: Future)
The one thing I found surprising here was how quickly some power banks stopped outputting at their maximum. For instance, the Baseus PicoGo dropped from its peak just four minutes into testing, while the Shargegeek 170 pumped a constant 90+ Watts until it was empty – no subtle voltage changes, just 20V/90W full tilt until the very end. Belkin’s Ultracharge also maintained its full output over time.
Below are the results covering capacity, output, and the actual capacity from our testing. The Anker Nano (A1638) had the least capacity when charging a laptop, at 1,600 mAh out of 10K. It spit out over 40W of output for a short time and stepped down to 20W, about 20 minutes in.
The Baseus Picogo Air and its slim design managed the highest actual capacity, but this is more a function of its limited 9V output and a quick drop from 19 to 11W. This tiny little device is great for a phone, but not ideal for larger items like our laptop. The large 24K Shargeek 170 sits in the middle of the pack, since it can output the full 20V and 90+W constantly.
Image 1 of 3
(Image credit: Future)
(Image credit: Future)
(Image credit: Future)
Conclusion
Conclusion
After testing all 13 power banks, there really isn’t a single ‘best’ bank. It comes down to what you’re charging and how patient you are. If you’re mainly topping off a phone or small tablet, capacity efficiency matters more than raw output. The Baseus PicoGo Air (currently $49.99 on sale), with its tiny footprint, shines here, squeezing out the highest actual-versus-listed capacity because of its lower 9V charging output. It’s small, light, and fits easily into a bag or purse; just don’t expect it to do much for a laptop, but it will easily charge a modern smartphone.
For laptop charging, the Shargeek 170 (currently on sale for $79.90) and Cuktech P23 $109.99 stood out by maintaining peak output (90+W @ 20V) through discharge instead of stepping down like most other power banks. At under $80, the Shargeek is a steal compared to the Cuktech.
Although others will charge laptops fine, that consistency could mean the difference between a predictable fast charge and a slower charge over time due to pre-programmed behavior or thermal throttling. The Anker Prime at $129.99 deserves a mention for charging laptops, too, as it pairs high output, fast charging, and the coolest chassis temps. However, it is the most expensive power bank we’ve tested. While it and others do come with informative displays, it’s up to you if that’s worth the premium.
If you’re concerned about the speed of charging the power bank, stay away from the Baseus PicoGo (AC22), Cuktech CP24, and CP12 – they took anywhere from 2.5 to 6 hours to reach full, which is a long time to wait if you need your battery back in rotation quickly. The good news is that these also have pass-through charging and will charge and discharge at the same time. The bottom line is, for phones, go small and efficient like the Baseus PicoGo Air; for laptops, prioritize sustained high wattage and sheer capacity (Shargeek 170, Cuktech P23, or Anker Prime), and if you want a bit of everything, the Anker Prime is hard to beat given its balance of speed, output, and temps, but it is expensive.