There are so many things we approximate to make life simple. Wires, for example, have no resistance or other strange effects. Crystal oscillators output their exact frequency. But surely our model of how a computer stores and loads memory is accurate, right? You put data in a particular location and, later, you take it out. The [FEX-Emu] developers have a different perspective. Once you have caches and, perhaps, multiple CPUs, it isn’t that easy.

The basic problem is this: if one CPU (or, more accurately, bus master) writes to a location, will another CPU have access to the new value? X86’s Total Store Ordering model gives programmers strong guarantees about when loads and stores become visible, while ARM deliberately uses a weaker memory model that permits considerably more reordering for performance and efficiency.

An emulator can, in theory, compensate by translating ordinary x86 memory operations into ARM acquire/release operations, but doing that for nearly every memory reference can be expensive. Newer ARM extensions such as LRCPC help considerably, while Apple took a more direct approach by adding an x86-compatible TSO mode to Apple Silicon. That lets ordinary loads and stores behave the way translated x86 code expects with comparatively little overhead.

Things get much uglier with unaligned accesses and atomic operations. X86 software routinely performs accesses that ARM would consider badly aligned, and x86 provides surprisingly strong atomicity guarantees within a cache line. FEX sometimes has to catch alignment faults and dynamically patch translated code with barriers. Split-lock operations are worse still: some require excursions through the kernel and signal handlers and can be hundreds or thousands of times slower than the normal case. Qualcomm’s newer Oryon cores improve matters by supporting coherent cache-line atomics, while Valve has shipped a Linux kernel optimization that handles some troublesome unaligned atomics directly.

There’s another particularly nasty corner involving write-combined GPU memory. PC games frequently expect x86 ordering semantics while writing uncached buffers destined for a discrete GPU. ARM currently lacks a clean equivalent for some of these stores, and FEX measured worst-case bandwidth more than 800× slower, enough to reduce some games to below 1 FPS. UMA systems fare much better because drivers can often substitute ordinary cache-coherent memory.

It’s a long article, but a good illustration of why modern emulators are less about translating instructions and more about reproducing decades of architectural assumptions that software quietly depends upon. Of course, not all emulators or processor recreations are this accurate, and often that’s good enough. But sometimes you need a recreation that is truly cycle-accurate and behaves exactly like the original.

India has extended its anti-spam regime to require caller-ID and call-management apps to share users’ spam reports with telecom operators, prompting spam-blocking app maker Truecaller to call the ruling anti-competitive.

On Friday, the Telecom Regulatory Authority of India (TRAI), the country’s telecom regulator, amended rules governing commercial communications, making it mandatory for caller-ID and call-management apps that let users flag calls as spam or junk to send those reports to a blockchain-based platform maintained by telecom operators. The platform tracks commercial communications and enforces anti-spam rules.

The change, TRAI said, is intended to broaden the pool of spam reports available for action against spammers, effectively connecting reports collected by apps with the telecom industry’s enforcement infrastructure.

However, Truecaller told TechCrunch that it sees this requirement as a “one-way exchange” that is “anti-competitive,” arguing that it transfers commercially valuable data from call-management apps like itself to telecom operators.

India is Truecaller’s largest market, accounting for well over 350 million of its more than 500 million monthly active users globally. The Stockholm-based company uses community reports alongside automated detection and other signals to identify and block spam calls.

The rules come as India grapples with spam and fraudulent calls at enormous scale. In its report in February, Truecaller said its users in the country encountered around 42 billion spam calls in 2025, including calls that were blocked, labeled, or ignored. The company also stated that it blocked nearly 12 billion spam calls during the year.

It is not the first time Truecaller and the Indian regulator have been at odds over how spam calls should be handled. The Swedish company previously objected to restrictions preventing call-management apps from automatically labeling calls from certain government-designated number ranges as spam. It argued that the exemption could allow unwanted calls to escape its filters.

However, Friday’s amendments retain that restriction and have barred call-management apps from blanket blocking, filtering, or spam-tagging calls from designated number series used for promotional, service, and transactional communications. Individual users can still choose to block such calls on their own devices, the regulator said.

“While our data and user sentiment clearly show that spam has skyrocketed due to this free pass to spammers, we have been compliant with this since late last year,” a Truecaller spokesperson said.

Sumeysh Srivastava, a partner at New Delhi-based consulting firm The Quantum Hub, who leads its telecom-regulation policy work, said the latest change bridges two distinct layers: Telecom operators provide the underlying network and run the blockchain-based anti-spam system, while caller-ID apps operate on top of the network to identify and filter calls.

That raises technical and jurisdictional questions, Srivastava told TechCrunch, including what reporting standards apps will have to follow and how the requirement will be enforced against companies that are not themselves telecom operators.

A March draft proposed (PDF) using India’s IT laws to enforce the requirement. However, Srivastava pointed out that the new announcement did not say whether that enforcement mechanism was retained in the final rules.

It is also unclear how much information the apps will actually have to provide under the updated regulation. Kazim Rizvi, founding director of New Delhi-based policy think tank The Dialogue, told TechCrunch that requiring an app to transmit a specific spam report made by a user is materially different from requiring it to share the broader datasets, reputation signals, or analytical systems it uses to identify suspicious calls.

The rules will need clarity on what information must be transmitted, how users are notified or asked for consent, and how that data can subsequently be retained and used, Rizvi said.

TRAI did not respond to TechCrunch’s questions about what information apps would be required to share and whether the rule would also apply to spam-reporting features built into smartphone operating systems and dialers such as Android and iOS.

New rules for AI-powered calls

The amendments also address the growing use of software and AI voice agents to make calls. Calls made automatically, without a person directly dialing the number, will now fall under TRAI’s application-to-person (A2P) framework. That includes robocalls and calls using prerecorded or artificial voices.

Companies using such systems will have to declare their use and the phone numbers involved to their telecom operators in advance. Undeclared A2P calls will be treated as spam, TRAI said.

The key test, Srivastava said, is how a call is initiated, rather than simply whether it uses an AI-generated voice, leaving some uncertainty around AI-assisted calls that involve human initiation.

Satya N. Gupta, a former additional secretary at TRAI, told TechCrunch that the new rules do not restrict businesses from using AI or other automated calling technologies, but instead require them to disclose their use to telecom operators.

Telecom operators will also be allowed to levy a termination charge of up to 5 paise (about 0.052 cents) per minute on A2P calls. However, calls made using certain designated number ranges will be exempt.

Rizvi told TechCrunch that the new definition could also cover calls made using software even when a person is still involved, such as calls from contact centers and click-to-call services. “Without that distinction, the A2P category risks becoming broader than the regulatory harm it is intended to address,” he said.

This week was a little quieter than last in the tech news space — if you ignore the hullabaloo caused by various AI companies that their models might one day kill us all if regulation doesn’t arrive.

A bit heavier than an Apple iPhone event, but important nonetheless.

FiiO showed us the M25 and M25 R2R at CanJam London in July, but both digital audio players were still pre-production models with provisional specifications, no confirmed pricing, and enough unanswered questions to keep the Head-Fi forums occupied for months.

Two months later, FiiO has officially launched both models in China for ¥7,499, and the final production specification looks considerably more interesting than another expensive Android music player stuffed with flagship DAC chips.

The standard model is now officially called the M25 VERITA and uses AKM’s flagship separated-DAC architecture with one AK4191EQ digital processing chip and two AK4499EX DACs. The M25 R2R takes the opposite route with FiiO’s proprietary 24+5-bit segmented R2R PRO resistor-ladder DAC. Same player platform. Same Chinese price. Very different approaches to converting bits into music.

FiiO could have changed the finish, added another anniversary logo, and charged more money. Nobody would have been particularly shocked. Instead, it is giving buyers two genuinely different conversion architectures without forcing them into two different product tiers.

Related Reading:

One Platform, Two Very Different DACs

The M25 VERITA uses the AK4191EQ + dual AK4499EX architecture that has become familiar in some of the better premium DACs and headphone amplifiers. The AK4191EQ handles digital processing separately from the AK4499EX conversion stage, allowing FiiO to build its own analog and amplification stages around AKM’s flagship platform.

FiiO is also introducing four selectable AKM-based sound profiles with the VERITA. We have not heard the final production version, so we are not going to tell you that one setting suddenly makes poorly mastered music sound like Rudy Van Gelder was behind the console. We will reserve judgment until we actually listen to it.

The M25 R2R is substantially different. Its proprietary 24+5-bit segmented R2R PRO architecture uses a segmented resistor-ladder topology rather than conventional delta-sigma conversion, following the direction FiiO has already taken with products including the K17 R2R.

The upper five bits use thermometer coding, an approach designed to reduce the cumulative errors that can occur in conventional R2R ladder implementations. FiiO also provides selectable voicing options for the R2R version.

R2R does not automatically mean warmer, more analog, or better. The DAC architecture is only one part of the signal chain, regardless of what somebody with 47,000 forum posts might insist.

What makes the M25 family interesting is the ability to choose between these two architectures while keeping almost everything else constant.

More Power Than the Prototypes

FiiO has also revised the output specification since the prototypes appeared earlier this summer.

Both production models can deliver up to 1,700mW + 1,700mW into 32 ohms from the balanced output in Super High Gain mode, with THD+N specified below 1%. High Gain reduces that to 1,000mW per channel into 32 ohms, while the 4.4mm balanced output delivers 220mW per channel into 300 ohms in Super High Gain mode.

The single-ended output is rated at up to 500mW per channel into 32 ohms in the same maximum-power operating mode. Four gain settings are provided, ranging from Low to Super High.

That is serious output for something intended to leave the house. Sensitive IEMs obviously do not require anywhere near that level of power, but the M25 should have considerably more flexibility with demanding full-size dynamic and planar magnetic headphones.

FiiO M25 and M25 R2R Key Specifications

• Models: M25 VERITA and M25 R2R
• Processor: Qualcomm QCS6490
• RAM: 8GB
• Internal Storage: 128GB
• Expandable Storage: microSD up to 2TB
• Display: 5.5-inch, 1080 x 2160
• M25 VERITA DAC: AK4191EQ + dual AK4499EX
• M25 R2R DAC: FiiO proprietary 24+5-bit segmented R2R PRO
• USB Controller: XMOS XU316
• FPGA: High-capacity FPGA with customized femtosecond clock
• Headphone Outputs: 3.5mm single-ended and 4.4mm balanced
• Maximum Balanced Output: 1,700mW + 1,700mW into 32 ohms in Super High Gain mode
• Balanced Output Into 300 Ohms: 220mW + 220mW in Super High Gain mode
• Maximum Single-Ended Output: 500mW + 500mW into 32 ohms
• Gain: Low, Medium, High, Super High
• Equalization: 31-band PEQ and Auto EQ
• Bluetooth: Bluetooth 5.2
• Bluetooth Transmission: SBC, AAC, aptX, aptX HD, aptX Adaptive, LHDC, LDAC
• Bluetooth Reception: SBC, LDAC
• Local/USB DAC Resolution: Up to 32-bit/768kHz PCM and native DSD512
• Battery: 6,500mAh
• Charging: Up to 30W fast charging
• Desktop Mode: Supported with battery isolation
• Chinese MSRP: ¥7,499 for either version
• European MSRP: €1,249 for either version

FiiO rates battery life at approximately 9.5 hours single-ended and 8.5 hours balanced for the M25 VERITA, while the M25 R2R stretches those figures to approximately 14.5 and 12.5 hours, respectively. Actual endurance will obviously depend on gain, output level, screen usage, wireless connectivity, and the headphones attached.

Why Are These Different?

The most interesting aspect of the M25 is not the processor, power figure, or increasingly inevitable wall of PEQ options.

It is choice.

Most manufacturers decide which DAC architecture belongs at a given price and sell you that implementation. FiiO is effectively offering two versions of the same automobile with completely different engines and charging the same amount for either one.

That is particularly interesting because the DAC market has become far more diverse again in 2026.

We recently covered the $289 SMSL R100, which uses a pair of Analog Devices AD1861-J multibit DAC chips, alongside the $649 SMSL D400 MK2, which goes in the opposite direction with AKM’s AK4191 and AK4499EX platform. Schiit has just introduced the $699 Bifrost 3, abandoning the previous True Multibit architecture for its proprietary Mesh digital processing feeding ESS modulators.

FiiO itself has been particularly aggressive with R2R. The $989.99 K17 R2R combines its 24+5-bit segmented resistor architecture with 31-band PEQ, network playback, and up to 4 watts per channel from a desktop chassis. Its K13 R2R, BR15 R2R, WARMER R2R, and even DM15 R2R portable CD player demonstrate that FiiO is no longer treating resistor ladders as some exotic technology reserved for people who own cable elevators.

The M25 brings that argument into the premium portable category and makes the comparison unusually clean. Buyers can choose AKM delta-sigma conversion or FiiO R2R without changing the processor, memory, storage, display, EQ platform, or basic amplification capability.

Whether they actually sound dramatically different is another question, and one we would prefer to answer with headphones on our heads rather than a block diagram in front of us.

The Bottom Line

When we saw the M25 and M25 R2R at CanJam London, the most compelling part of the concept was FiiO’s decision to offer two fundamentally different DAC architectures inside essentially the same player.

The production models make that idea even stronger. Both receive a fast QCS6490 platform, 8GB of RAM, 128GB of storage, powerful headphone amplification, 31-band PEQ, Auto EQ, desktop operation, and enough connectivity to function as portable players or serious USB DAC/headphone amplifiers.

The decision comes down to the conversion stage: AKM’s flagship delta-sigma architecture in the M25 VERITA or FiiO’s latest segmented resistor ladder in the M25 R2R.

And because FiiO is charging exactly the same price for both in China, choosing between them may finally come down to something refreshingly old-fashioned in high-end digital audio.

Which one actually sounds better?

Pricing & Availability

Both the FiiO M25 VERITA and M25 R2R are listed for €1,249 pre-order on FiiO’s European website, with availability expected in mid-to-late November 2026.

FiiO has not announced official U.S. pricing or a North American release date.

Related Reading:

“The magnitude of the alleged infringement is not a defense,” the court told Suno in August.

Universal and Sony have now sued the AI music company a second time. They filed the complaint in the District of Massachusetts on Friday, as case 1:26-cv-14275.

It asserts 60,202 sound recordings. The first case, filed in June 2024, is still running.

The filing names twelve label entities, led by UMG Recordings and Sony Music Entertainment. Warner is not among them.

The judge effectively told them to file it

The labels first tried to fold the new recordings into the existing case. They moved to amend the original complaint to add more than 61,000 further works.

A forensic analysis of Suno’s training data had turned those up.

Suno opposed that motion. On 18 August the court denied it without prejudice, mainly because amending would disturb the case schedule.

The court did not question the claims themselves. The labels are “of course entitled to pursue valid claims for copyright infringement”, it wrote.

It then pointed at the exit. Proceeding in “parallel cases” best balanced the labels’ interests with judicial economy, the court concluded.

The new complaint says the labels are doing exactly what the court contemplated.

What the labels say Suno already conceded

The complaint leans on Suno’s own answer in the first case. The labels say Suno acknowledged building its models “by showing the program tens of millions of instances of different kinds of recordings”.

Those recordings “presumably included recordings whose rights are owned by the Plaintiffs in this case”, Suno said, according to the filing.

What Suno would not do, the labels say, is name them. They commissioned the forensic analysis for that reason, and say it found millions of their works in the training corpus.

Suno has not yet responded to the new complaint, which reached the docket on Friday.

Three licensing deals are the labels’ sharpest weapon

Suno has spent the past year signing the very agreements the labels say it should have sought first. The complaint lists three.

Warner Music Group signed in November 2025, and the complaint calls it a “former plaintiff”. BMG followed on 12 August, from a company the labels note had “never sued Suno”.

Believe signed on 8 September to supply training data. Nine days later Suno launched v6 on licensed catalogue.

The labels draw the obvious inference. Three agreements in under a year “confirm that a functioning market exists for licensing sound recordings to train generative AI models”, the complaint says.

Suno has tried to frame the deals differently. Chief product officer Jack Brody has said the revenue share is “not in exchange for training” and that the deals are “not really about the data”.

The complaint quotes Mikey Shulman, Suno’s co-founder and chief executive, saying the company “will be licensing works”. It calls the deals “a functioning licensing market, whatever nomenclature Suno adopts for litigation purposes”.

Suno’s own terms ask for what it did not give

The filing turns Suno’s contract with its users against it. Since January 2024 its terms of service have required every user to grant Suno a “worldwide, non-exclusive, fully paid-up, sublicensable” and “irrevocable” licence over anything they upload.

That licence covers reproducing, storing, modifying and creating derivative works from user content. The labels say it shows Suno knew perfectly well that training on someone’s audio needs permission.

Demanding a licence upstream while taking commercial recordings for free, the complaint argues, makes the infringement “knowing and willful”. Willfulness is what unlocks the higher damages tier.

The artist-name switch

Suno once blocked users from prompting it with the names of specific recording artists. Shulman cited that policy as proof the models were built to generate “completely new outputs”.

The complaint says the block is gone. “This design feature turned out to be so fundamental that Suno simply switched it off,” it reads, and users can now prompt with artist names through prompt augmentation.

The complaint also quotes an early investor. “If we had deals with labels when this company got started, I probably wouldn’t have invested in it,” the investor told Rolling Stone.

Suno, the investor said, “needed to make this product without the constraints”.

What it could cost

There are three counts. Two are direct infringement, split between post-1972 recordings and pre-1972 ones protected under the Music Modernization Act.

The third is circumvention. The labels say Suno “stream ripped” audio from YouTube, employing code to access, extract, copy and download files in breach of YouTube’s own terms and of section 1201(a).

The labels ask for up to $150,000 per work for willful infringement, plus up to $2,500 for each act of circumvention. Across 60,202 recordings the statutory ceiling alone runs past $9bn.

They also want an injunction requiring Suno to stop circumventing YouTube’s measures and to stop infringing, and they have demanded a jury.

The flood is the market-harm argument

Fair use turns partly on market effect, and the labels have reached for numbers rather than adjectives.

Deezer reported in July that AI tracks passed half of all daily uploads for the first time, roughly 90,000 a day, up from about 39% six months earlier.

Suno has told investors its users generate a Spotify catalogue’s worth of output every two weeks. The complaint also cites The Velvet Sundown, an AI act that reached over a million monthly Spotify listeners in summer 2025 before listeners learned it was machine-made.

Shulman’s own words do a lot of work in the filing. The complaint quotes him saying it is “not really enjoyable to make music now” and that “increasingly taste is the only thing that matters in art and skill is going to matter a lot less”.

Suno is worth more than when it was first sued

The complaint sets out the company’s finances in detail. Its latest round raised more than $400m at a $5.4bn valuation, against $2.45bn in autumn 2025.

The labels say Suno makes hundreds of millions a year and projects $1bn in revenue by 2028. Its top subscription tier costs $24 a month.

People who had worked together at Kensho Technologies founded Suno in July 2023. The company works out of Cambridge, Massachusetts, and Microsoft integrated it into Copilot that December.

It is not the only front

Sony ran the same play against Suno’s closest rival in July, suing Udio a second time over 30,000 more songs.

Europe has already produced a ruling. A German court found Suno broke copyright in a case brought by the collecting society GEMA.

Shulman said in January 2025 that it seemed “silly to throw a bunch of venture dollars at lawyers instead of sitting down and talking about how you could work together”. The labels note he said it seven months after they sued.

September brought a good batch of new Netflix TV series, spanning across different genres and tones. This weekend we’ve got a crime family drama with all the sharp banter you’d expect from Guy Ritchie, a true crime retelling that brings a female killer to the spotlight this time, and a bleak Nordic thriller that turned into a surprise global hit within days. Each of these picks earns every bit of its buzz, so add them to your watchlist right away.

We also have guides to the best new movies to stream, the best movies on Netflix, the best movies on Hulu, the best free movies, and the best movies on Amazon Prime Video.

Monster: The Lizzie Borden Story (2026)

Genre: Biography, True Crime, Drama
IMDb rating: N/A
Rotten Tomatoes: N/A

The fourth installment in Ryan Murphy and Ian Brennan’s Monster anthology tackles the infamous 1892 murders of Andrew and Abby Borden, told through the eyes of their daughter Lizzie (Ella Beatty), the prime suspect who was ultimately acquitted. Trapped in a household built on control and cruelty, Lizzie and the family’s maid, Bridget Sullivan (Vicky Krieps), retreat into a shared fantasy that slowly starts blurring with reality.

Ella Beatty gives an utterly captivating performance as the unsettling yet deeply sympathetic lead. I was impressed by the eerie, claustrophobic setting that makes the family house feel like a prison for Lizzie. I also appreciate how the series leans hard into gender, class, and repression as the real culprit behind the violence, instead of treating the murders as just another grisly mystery to solve.

You can stream Monster: The Lizzie Borden Story on Netflix.

The Gentlemen (Season 1, 2024 – Season 2, 2026)

Genre: Action, Comedy, Crime
IMDb rating: 8/10
Rotten Tomatoes: 79%

An aristocratic British officer, Eddie Horniman (Theo James), unexpectedly inherits his family’s massive country estate only to discover a massive weed empire hiding underneath. He gets sucked into a chaotic underworld filled with eccentric gang leaders, posh criminals, and dodgy deals. It serves up Guy Ritchie’s signature brand of razor-sharp banter, snappy quick cuts, and violence.

I absolutely loved the electric, slick dynamic between Theo James and Kaya Scodelario in Season 1. Their cool, calculating swagger makes it a fun watch. Season 2 shows Eddie expanding his operation into Italy, and the stakes escalate right alongside the ambition. I’ll admit the early episodes had me worried the show had lost some of its charm, but by the time it reaches its final stretch, the show pulls everything back on track, enough to make me genuinely eager for season three.

You can stream The Gentlemen on Netflix.

Blood Sacrifice (2026)

Genre: Crime, Drama, Thriller
IMDb rating: 6.3/10
Rotten Tomatoes: 90%

Blood Sacrifice is a newly released five-episode miniseries that follows Detective Thomas Berg (Jakob Oftebro), who reluctantly teams up with his estranged ex-cop father Alfred (Peter Andersson) after two police officers are found murdered near Stockholm’s summer archipelago. As more officers turn up dead in the same brutal fashion, Thomas realizes the killer is copying a case the original suspect couldn’t have committed, since that man is already behind bars.

From George Kay, the creator of Lupin, strips away usual Hollywood polish in favor of slow, methodical procedural teamwork. It became Netflix’s most-watched non-English series within days of release, and it’s easy to see why. The father-son dynamic gives the investigation real emotional stakes beyond just chasing a killer, and the show delivers a bleak, grounded Nordic noir that keeps you guessing until the final stretch.

You can stream Blood Sacrifice on Netflix.