Astell & Kern builds some outstanding digital audio players, and the company also knows how to charge us for them. But it’s taken pity on those of us without the wherewithal, and has developed this HC5 headphone amp/DAC.
It is a portable device intended to transform your bog-standard smartphone into a viable source of hi-res audio for anyone with the wired headphones good enough to take advantage of it. With high-end AKM digital-to-analog conversion components on board, Astell & Kern’s ‘digital audio remaster’ and ‘high driving mode’ technologies included, and both balanced and unbalanced headphone outputs, the HC5 is specified to get the job done. It also features half- a-dozen digital filters, as well as a huge range of volume adjustment and some slightly imprecise-feeling physical controls that serve to undermine the robust and tactile overall nature of the device just a little.
In operation, the Astell & Kern is slightly lightweight at the bottom of the frequency range — it doesn’t have quite the heft or substance that it really should have, and can lack a little apparent punch as a result. In every other meaningful respect, though, it’s a very accomplished device indeed. It’s dynamic, lavishly detailed, creates a large and well-organised soundstage, and communicates through the midrange like nobody’s business.
If you want to turn your humble smartphone into a device that bears comparison with some of the best three-figure digital audio players around, and if you have suitable headphones too, then you ought to be after one of the best DACs around. And you should absolutely hear the Astell & Kern HC5.
(Image credit: Future / Simon Lucas)Astell & Kern AK HC5 review: Price & release dateThe Astell & Kern HC5 headphone amp/DAC launched at the very end of 2025 and hit shelves in January 2026.
In the United Kingdom it sells for £399. You’ll need to part with more like $489 in the United States, while in Australia the going rate is around AU$779.
This makes it a reasonable (rather than 'cheap') option, of course — at only $85 / £85 (or around AU$120) the iFi Go Link 2 Max would be an example of something more affordable, while the $4,499 / £4,499 (around AU$8,999) iFi iDSD Phantom (albeit a desktop solution) proves this A&K is hardly at the costly end of the market either.
(Image credit: Future / Simon Lucas)Astell & Kern AK HC5 review: FeaturesApparently this is the first time that AKM’s flagship AK4191EQ digital processor and AK4499EX digital-to-analog converter have appeared in a product as small and as relatively affordable as this one. AKM suggests that splitting the d-to-a decoding into two stages preserves signal integrity — and Astell & Kern agrees. Mind you, that hasn’t stopped Astell & Kern bolstering the AKM chipset with its own ‘DAR’ (digital audio remaster) upsampling technology; it functions with both PCM and DSD content, increasing the sampling rate by quite a degree, and it can be switched off if you so desire. The upshot, though, is a DAC that supports 32bit/768kHz and DSD512 resolutions.
Astell & Kern’s ‘high driving mode’ technology is also making its debut appearance in an affordable product. The horizontal op-amp layout helps keep the physical size of the HC5 to a minimum, but also allows for significant driving power. It's worth bearing in mind, though, that the HC5 has no power supply of its own, and the prodigious amplification power it can muster will cut into the battery life of your source player more than somewhat.
There are a total of six digital filters for the user to explore, but this is the extent to which you get to influence the sound of the Astell & Kern. Perhaps oddly, there’s no sign of the EQ adjustment that’s commonplace on quite a few of the HC5’s nominal rivals.
The proof that the digital-to-analog conversion abilities of your smartphone or laptop are pretty rotten can be found in the number of extremely affordable USB DACs that are on sale now — TechRadar has rounded up the best of them, as I've previously alluded to.
But there’s no denying that spending $489 / £399 (or equivalent) on an Astell & Kern HC5 takes the sound of your laptop or smartphone to another level entirely. Provided you have the headphones to properly exploit its talents, anyway.
Load up a 24bit/48kHz FLAC file of Off Course by Oh Sees and all will rapidly become clear. When it’s dealt with by the HC5 the recording is more open and spacious, and yet somehow more singular and unified at the same time. The soundstage from which it comes is big in every direction, and the spaces and silences on it suddenly become almost as significant as the actual occurrences. Each participant has more than enough room in which to operate without crowding any other.
There’s real positivity and energy to the way the Astell & Kern delivers the recording, a sense of engagement and enthusiasm which is strongly at odds with the sound served up by a bog-standard smartphone. But at the same time the presentation is poised and balanced despite its direct and upfront nature — there’s never any suggestion the HC5 is not fully in control. This authority is especially apparent at the bottom of the frequency range, where the attack and decay of bass sounds is observed with such care that rhythmic expression is always confident and naturalistic. Down at the bottom end is where the Astell & Kern’s most significant shortcoming lurks, though. Just like the rest of the frequency range, the low end is lavishly detailed where timbre and texture are concerned — there’s more than enough information revealed and contextualised to let you know you’re getting a very full account of a recording. But there’s an appreciable shortage of body and substance to the low frequencies the HC5 generates. It can dig respectably deep but it doesn’t hit as hard, or with as much weight, as is ideal.
Otherwise, though, the frequency response is just as pleasing and convincing as everything else about the Astell & Kern. It’s articulate through the midrange, communicating the attitude and intention of a voice every bit as willingly as it reveals the details of tone and technique. The top of the frequency range is bright and similarly detailed, but carries enough substance along with it to prevent it sounding thin or aggressive — and this is true even if you like to listen at considerable volume.
There’s ample dynamic headroom available for when the attack or intensity or simple volume of a recording ramps up, and a similar facility with the less obvious (but no less important) dynamics of harmonic variation. It’s not enough for the HC5 to identify and reveal the most fleeting, most minor details in a recording - it is able to put them into the sort of context that means they serve as part of an overall picture, rather than treating them like some kind of academic exercise.
There’s ‘portable’ and then there’s the Astell & Kern HC5. Unlike quite a few nominal rivals, (some of which look like hip flasks) this headphone amp/DAC is sufficiently small (64 x 32 x 16mm, HxWxD) and light (46g) enough to be taken out even by those not wearing a coat with sturdy pockets, or carrying a bag of some kind.
The aluminium construction is impeccable, the proportions are almost instinctively suitable for a palm, and those concave sides only add to the tactility. Obviously Astell & Kern is a past master at this sort of thing, but it’s nice to be reminded it can do it for a £400 product as readily as it can for something costing £4K.
At one end the HC5 has a USB-C input — it’s supplied with both USB-C / USB-C and USB-C / Lightning cables, so you should have no problem connecting your source of music. At the other there are 3.5mm unbalanced and 4.4mm balanced headphone outputs.
Next to the USB-C input there’s a relatively large, knurled dial that takes care of volume — the HC5 features a frankly excessive 150 steps for ultra-granular control. It feels just slightly wobbly and a little less premium than the main body of the device.
That’s also true of the little button on the side of the HC5 that is basically your control for everything that isn’t volume; it feels just fractionally imprecise. Thankfully the display it works in conjunction with, which occupies fully half of the front of the device, is crisp and bright and legible. Here’s where you’ll turn the digital audio remaster algorithm on or off, and choose between the six available digital filters — and this, in essence, is what constitutes ‘set up’ where the Astell & Kern is concerned.
If — and it’s quite a big ‘if’ — you have wired headphones capable of taking advantage of the HC5’s unarguable sonic abilities, and if you have a streaming service subscription that includes lots of hi-res content, then there’s no arguing with the difference the Astell & Kern can make to your portable listening experience. Which has to count as ‘value’ in anyone’s language, doesn’t it? OK there are cheaper devices out there too, but none look or feel quite as good as this one — and there are also much more expensive ones out there…
You own good wired headphones
Because no good can come of plugging them straight into a smartphoneView Deal
You have access to truly hi-res content
The HC5 can deal with the biggest real-world resolutions aroundView Deal
You think ‘portable’ should mean ‘portable’
Small and palm-sized, the A&K can genuinely go anywhere without becoming a burdenView Deal
Your headphones are on a 6.3mm jack
You’ll need an adapter to use them with the HC5View Deal
You place a lot of value on the physical interface
The feel of the physical controls isn’t as premium as it might beView Deal
Your source player doesn’t have huge battery life
The A&K uses your player’s power, and quite a bit of itView Deal
Where price and performance is concerned, the most obvious rival to the Astell & Kern HC5 is the Chord Mojo 2. In absolute terms it probably has the edge over the HC5, and its EQ adjustment facility is worthwhile too — but it stretches the meaning of the word ‘portable’ quite a distance…
How I tested the Astell & Kern AK HC5At one end I connected the Astell & Kern HC5 to an Apple iPhone 15 Pro, a FiiO M15S digital audio player, and an Apple MacBook Pro — this way I was able to access a whole lot of content of various genres and resolutions, and also to assess the device’s performance against the extremely impressive D-to-A abilities of the FiiO DAP.
At the other end I connected a pair of Austrian Audio The Arranger open-backed over-ear headphones to the 3.5mm unbalanced output and a pair of Sennheiser IE900 in-ear monitors to the 4.4mm balanced alternative.
All Things Considered host Scott Detrow speaks with NPR's editor-in-chief Thomas Evans and Nina Totenberg about her reporting on the final day of the Supreme Court term.
Mark Zuckerberg met with Kalshi's CEO last year about a potential deal, but talks did not move forward. Now Meta is making its own prediction market app.
(Image credit: Patrick T. Fallon/AFP and Aaron Schwartz/Bloomberg via Getty Images)
Claude Sonnet 5 brings stronger agentic AI features, lower pricing, and updated safety protections. Here's what IT leaders need to know.
The post Claude Sonnet 5: Everything to Know About Anthropic’s New Model appeared first on TechRepublic.
TerraMaster has been making NAS hardware long enough to know that the upgrade cycle is everything. The F4-424 Pro arrived in early 2024 with a strong hand: an Intel Core i3-N305, 32GB of DDR5, and a build that put competitors under genuine pressure. Two years on, the company returns with the F4-425 Pro, and the result is a more complicated story than a straightforward generational step forward.
On the hardware side, the headline changes are meaningful. Dual 5GbE replaces the F4-424 Pro's dual 2.5GbE, which doubles the theoretical single-client throughput ceiling. The M.2 slot count increases from two to three. Both are welcome improvements that justify the refresh.
But there is a wrinkle. The processor moves from the Core i3-N305 to the Intel N350. The N350 is also an 8-core chip, and its maximum burst clock of 3.9GHz fractionally exceeds the N305's 3.8GHz. The difference is that the N350 is an Atom-architecture Alder Lake-N part rather than a Core-class one. Per-core performance and integrated GPU capability are both lower. The advantage is better power efficiency, but some will see this as a retrograde step.
The other major story for this platform is TOS 7. TerraMaster has rebuilt its operating system around an AI-first philosophy, with the OpenClaw assistant promising natural language control over 90% of common configuration tasks. That skirts the whole AI backlash, and those who don’t want to chat with their NAS, but equally, there are some that will embrace these features.
At £639.99, the F4-425 Pro sits in a remarkably competitive bracket. The Ugreen DXP4800 Pro offers a Core i3-1315U and a single 10GbE at £689.99. The Ugreen DXP4800 GT delivers dual 10GbE and ECC memory support at £589.99. The TerraMaster undercuts or matches both on M.2 count and brings a genuinely new OS story to the table. Whether that is enough depends on what the buyer most needs, but on spec alone, this isn’t one of the best NAS in this sector.
(Image credit: Mark Pickavance)TerraMaster F4-425 Pro: Price and availabilityThe F4-425 Pro launched on 23 June 2026, available direct from TerraMaster, as well as retailers including Amazon.com, Amazon.co.uk, and B&H Photo.
At the time of review, the 8GB model is priced at $640 / £640 from TerraMaster and Amazon. Online retailer B&H Photo wants $644.99. And all these prices are without drives, obviously.
One curiosity is that although the F4-425 Pro spec is for a system that uses the Intel N350 and comes with 16GB of RAM, TerraMaster also has a second SKU with the N305 CPU that its predecessor used, and 8GB of DDR5. This lower spec model is priced at $559.99.
The top SKU price matches the F4-424 Pro's UK debut cost, interestingly.
For the purpose of this review, we’ll focus on the N350 model, since that was the one that TerraMaster supplied us.
Given the price is similar to its predecessor, the networking upgrade from 2.5GbE to 5GbE and the additional M.2 slot make this solution an attractive option over the F4-424 Pro, even if neither is exactly a bargain.
What seems odd is that the release of the F4-425 Pro hasn’t made the previous F4-424 Pro any cheaper, unfortunately. More than a disappointment for budget-conscious buyers, TerraMaster is asking $687.99 USD for that previous design.
That pricing suggests TerraMaster thinks it's competing with itself to some extent, but recent releases in this NAS space strongly contradict that notion.
Ugreen had the DXP4800 Plus, added the DXP4800 Pro and now the DXP4800 GT.
Direct from Ugreen, the DXP4800 Plus is $583.99, the DXP4800 Pro is $639.99, and the new DXP4800 GT is on sale for $527.99. Given that all of these have more powerful processors than the TerraMaster F4-425 Pro, and the DXP4800 Pro has a 10GbE LAN port, TerraMaster's pricing seems oddly out of touch.
As this is a NAS review, it’s the law that I must mention Synology, even if this company has all but abandoned the SOHO NAS space. After an implausibly long delay between releases, Synology launched the Synology DS925+, a NAS that’s powered by the ancient AMD Ryzen V1500B. Oddly, given this brand's history, it is the cheapest option at $511.99. However, the DS925+ comes with only 4GB of RAM, its dual M.2 slots accept only Synology-branded modules, and the best LAN ports are only 2.5GbE.
Given these alternatives, and some others I’ve not mentioned, the F4-425 Pro seems overpriced and underspecced, a phrase I thought I’d never use in reference to TerraMaster hardware.
Item
Spec
CPU:
Intel N350 8-core 8C / 8T (Gracemont E-Cores)
GPU:
Intel UHD Graphics 770 (32 EUs)
RAM:
16GB DDR5 non-ECC SODIMM, expandable to 32GB via SO-DIMM swap
SATA Storage:
4x 3.5/2.5-inch SATA III
M.2 Storage:
3x M.2 NVMe (both PCIe Gen 3 x1
Ports:
3x USB-A 3.2 Gen 2 (10Gbps)
1x USB-C 3.2 Gen 2 (10Gbps)
1x HDMI 1.4
Networking:
2x 5GbE RJ45
OS:
TOS 7
File Systems:
EXT4, Btrfs
Maximum Capacity:
152TB (4x 32TB + 3x 8TB M.2)
RAID Modes:
TRAID, TRAID+, RAID 0/RAID 1/RAID 5/RAID 6/RAID 10
PSU:
12V 7.2A 90W external
Power Consumption:
45W under full SATA Load
Dimensions:
186 x 277 x 277 mm
Weight:
2.9kg (including PSU, but without drives)
Warranty:
2 Years
TerraMaster F4-425 Pro: DesignThe F4-425 Pro continues the black motif TerraMaster adopted with the 424 series, and I do prefer this look to the all-silver devices of the past.
The outer shell gives the chassis a denser feel than the silver-and-aluminium aesthetic of earlier TerraMaster NAS units. The front face is clean and largely featureless: four drive bays with activity LEDs behind a pattern of small holes, and TerraMaster branding on each bay.
Even without drives, this is a hefty item measuring 186 x 277 x 277 mm with a net weight of 2.9kg. That is compact for a four-bay unit, which is marginally smaller than the same type from Ugreen or Synology.
Drive installation remains tool-free for 3.5-inch mechanisms. But 2.5-inch mechanisms still require screws to seat, and aren’t the best design I’ve seen.
Firstly, there is no lock to stop an accidental press from disconnecting a drive. And, given how long TerraMaster has been making NAS, you’d think that trays with numbers on them might have made it onto its devices.
Admittedly, TerraMaster does provide labels for you to stick on the drive facias, but since they took the time to emboss "TERRAMASTER " onto the plastic fronts of the trays, you think they could also put 1, 2, 3, and 4 on them.
I can’t recall when I first complained about the lack of tray locks on TerraMaster NAS, but it was so long ago and so often repeated that its failure to fix this is evidently not unintentional.
Another place where TerraMaster does its own thing is with respect to the M.2 slots, where this new model has three and not the two that the F4-424 Pro came.
On Synology and Ugreen NAS, the M.2 slots are accessible behind a panel held in place with a couple of screws. But here, there are four screws to remove, and then the entire outer shell of the case slides off. This does provide great access, but I suspect this was done as a cost-saving measure, mostly.
Having three M.2 slots sounds wonderful for those with spare NVMe drives to populate them, although that’s tempered slightly by their PCIe Gen3x1 bandwidth allocation. But what I found slightly shocking inside is that the 16GB DDR5 SODIMM is in a single slot. Therefore, if you want to upgrade the RAM to 32GB, you will be forced to remove the existing 16GB module. I’ll talk more about memory later, because the memory controller on the N350 has some odd features.
(Image credit: Mark Pickavance)Moving on to external hardware features, having reviewed the latest Ugreen NAS recently, the number of ports on the F4-425 Pro seems on the low side.
On the front is a single USB 3.2 Gen 2 Type-A port, and there is no card reader of any variety. The rear has another two USB 3.2 Type-A ports, both of which are 10Gbps, and a single USB-C that is also the same USB spec. And, there are two 5GbE LAN ports and a single HDMI output.
There is no USB4 or Thunderbolt, and no PCIe slot to add a 10GbE card. You can channel-bond the two 5GbE ports, but there aren’t many switches that support 5GbE at this time.
Cooling is provided by a 120mm system fan that sits centrally at the rear. In smart mode, it adjusts speed to drive temperature. TerraMaster quotes 20.9dB(A) at standby with four SATA drives fitted, which is a quiet figure. And, this system is quiet in typical use.
Overall, my takeaways are that this platform is better in some respects than the N305-based F4-424, but each enhanced feature comes with a caveat, it seems.
Due to Intel's endless rebranding, the chip in the F4-424 Pro was a Core i3 N305, whereas the one in this NAS is a Core i3 N350. So like Thor in Ragnarok, it lost an eye, but gained something else.
In reality, these chips are remarkably similar, since they both use eight Gracemont E-Cores, have an identical 6MB cache, and a single memory channel for DDR5 memory.
But there are two important differences, the first being that the N350 is capped at 7W of base power, not the 15W of the N305. And to conserve power, the N350 can drop the base clock to only 100 MHz when the system is idle. There are also some enhancements to the GPU clocks to deliver a little more speed, but the Intel UHD Graphics 770 (32 EUs) doesn’t have gaming potential.
The switch to this architecture over the previous one focuses on making this NAS quieter, less power-hungry, and cooler. That makes perfect sense, even if it's slightly at odds with the headlong charge towards AI that TOS 7 is mustering.
What the processor change doesn’t address is that this platform is pinched for PCIe lanes, since both the N305 and N350 have only nine PCIe 3.0 lanes.
That goes some way to explain why the M.2 slots are only single-lane and can only achieve roughly 1GB/s, irrespective of what modules you put in them. As a side note, with these performance restrictions, there seems little point in using expensive NVMe drives in this system.
If we assume that each of the M.2 slots gets a lane, and each of the 5GbE LAN ports another one, and the SATA ports one lane, and the USB another four, then that’s ten, and we only have nine lanes. That infers a PCIe switch is used in this system, because otherwise the number of lanes and the systems that use them don’t stack up.
To be clear, it’s not like the NAS is massively over-subscribed on PCI bandwidth, but if you fill every port and slot, something is going to give at some point.
There is also something of an oddity with the memory model of this NAS. If you head over to Intel and look at the specifications for the N350, you might notice that Intel states the maximum amount of memory this processor supports is 16GB. Well, that appears to be wrong, because you can put 32GB in this NAS, and it will work. You can’t add any more, and because it has only a single memory channel and SODIMM slot, you can’t use two 16GB modules.
If you want more memory, you need to take the next rung up on the processor ladder. The Ugreen NASync DXP4800 Pro uses an Intel Core i3-1315U, and that has dual-channel memory enabling it to address 96GB.
(Image credit: Mark Pickavance)Dual 5GbE is one of the most consequential hardware changes over the F4-424 Pro. With link aggregation configured on a managed switch, the aggregate theoretical bandwidth reaches 10Gbps. Single-client throughput sits at up to 625MB/s per port, compared to 312MB/s from the F4-424 Pro's 2.5GbE ports. For 4K video workflows, large file transfers, and multi-user small office environments, that improvement is tangible.
However, it begs the question of why they didn’t simply combine the bandwidth and offer a single 10GbE port in the first place? If anyone wants another 2.5GbE line, perhaps for a network failover option, USB adapters that can use the 10Gbps USB-A ports are inexpensive.
The absence of a 10GbE port will disappoint buyers who want the maximum headline figure. The Ugreen DXP4800 Pro provides single-port 10GbE at $719.99, and the DXP4800 GT provides dual 10GbE at $559.99. TerraMaster is positioning 5GbE as a practical middle ground that delivers meaningful real-world improvement without requiring expensive 10GbE switching infrastructure. That argument has merit for some buyers, although 5GbE network switches that support channel-bonding aren’t especially common.
From a hardware perspective, the new N350 isn’t a huge move from the N305 used on the previous generation. What’s different here is that this hardware is better at managing power, heat, and the limited PCIe lanes available to it.
In a straight compute fight, the N305 might be better, but for running 24/7 through a hot summer, the N350 has some advantages that it might need.
TOS 7 is what TerraMaster is staking the F4-425 Pro's reputation on. Earlier TOS releases were functional but acknowledged as trailing Synology DSM and QNAP QTS in polish and app ecosystem depth. TOS 7 does not try to close that gap incrementally. Instead, TerraMaster has chosen to reframe the competition around an AI-first workflow that neither Synology nor QNAP currently matches.
The OpenClaw assistant is the visible centrepiece. A single click from the app centre installs it, after which the user can issue natural language commands for the full range of administrative tasks: RAID configuration, user account management, backup scheduling, security setup, and file management. The theory is compelling. A buyer who understands that they need a NAS but has never configured one could set up a working system through conversation.
Except there is a huge hole in this plan, since the buyer needs to understand how to install and configure OpenClaw, use LLMs and their API keys. And, those require you to fund your API of choice to accept those messages from OpenClaw, unless you have the expertise to run your own local models using Ollama, and direct OpenClaw to use that.
If you are willing to use AI but not pay for it, the fun doesn’t typically get started.
I tried my best to get the OpenClaw beta to run on this hardware during one of the hottest afternoons of the year, but failed miserably. I got Ollama installed as a Docker container and even loaded a model to use for local access, but I couldn’t get OpenClaw to work with it. Maybe on a cooler day, with better documentation, I could manage this, but given the level of personal experience I brought to this problem, it isn’t something anyone new to AI would want to embrace.
NAS manufacturers have a history of ambitious AI and automation claims that perform well in demonstration scenarios and inconsistently in everyday use. TOS 7 has been in development for over 300 days, according to TerraMaster, but that development timeline says nothing about robustness across edge cases, network configurations, and drive combinations that real users will bring to the table.
And having a tool like OpenClaw running on the NAS, with the ability to create users, shares, and folder structures, and to delete things, might not be as wonderful as it first seems.
(Image credit: Mark Pickavance)Ignoring TerraMaster's attempt to board the AI hype-train, the foundation platform for TOS 7 is solid. Docker and virtual machine support carry forward from TOS 6. The backup suite is comprehensive. Plex and Emby are available from the app centre.
DLNA compliance means out-of-the-box compatibility with smart TVs and media players without any configuration. HyperLock WORM protection addresses compliance requirements for business users. The security baseline, including AES folder encryption, OTP authentication, and the firewall, is mature.
Where TOS has historically needed work is in surveillance capabilities and app ecosystem depth. Whether TOS 7 addresses the Surveillance Manager's lack of a dedicated client with timeline and event marking is a specific testing priority.
There isn’t any debate, this is the best version of TOS yet, and it still has a few features that other NAS brands still haven’t delivered. One I especially like is the TRAID hybrid array model, which allows mixing drives of different sizes and yields more resilient capacity.
Only Synology offers anything comparable, and users of QNAP, Asustor, and Ugreen would gladly welcome such a feature.
A feature that none of those brands might embrace is that the HDMI port on this machine is effectively unused. It’s been a decade since TerraMaster first launched a NAS with an HDMI port on it, and in my review of that equipment, I commented that the port needed support by first-party apps for media playback.
Ten years on, the HDMI port remains useful for TerraMaster production staff to check whether the systems are booting correctly, but is of almost zero use to their customers. Everyone else integrates their HDMI if they have it, but TerraMaster stubbornly refuses to. There are some nefarious ways to get the HDMI to work using Virtual Machines, but that this annoyance was left to fester for so long is incredibly poor.
The other issue that TOS 7 doesn’t address is TerraMaster's somewhat confused approach to the first- and third-party application ecosystem.
When a NAS has four SATA bays, it effectively constrains its peak transfer performance to or from that array. Even with the fastest possible option, RAID 0 on four drives, with each NAS drive being like the WD Red Plus models I used for testing, the total array is only capable of four times the 199MB/s limit for those drives. That’s 796MB/s. which wouldn’t saturate a single 10GbE LAN port, if the F4-425 Pro had one.
More realistically, the RAID mode of choice is likely to be one with redundancy, reducing performance to 597MB/s, which fits rather neatly with what I’d expect from a 5GbE transfer.
But this NAS has two LAN ports, and without any enhancements, using channel bonding of the two 5GbE ports is unlikely to help the total amount of data read or written.
The only way to make things run faster would be either to use SATA SSDs, or NVMe storage, or use the M.2 for caching. I used the latter and was able to get more throughput.
One interesting bit about the M.2 slots and using them for caching on this system is that, typically, for read and write caching, two SSDs are required. But not on TOS 7, which will allow you to use a single SSD for both read and write caches. If you are wondering why others haven’t followed TerraMaster’s example, when you choose to do that on this NAS, you get plenty of warnings about how this can go wrong.
I tested a Crucial P5 500GB, and it achieved an NAS score of 859 MB/s. Therefore, to saturate both 5GbE LAN ports would probably require two NVMe drives in a RAID 0 configuration.
The lesson here is that even with all your storage ducks in a row, it might prove challenging to deliver sufficient performance for the dual 5GbE LAN ports unless you use caching liberally.
My views on the F4-425 Pro are a little mixed. It’s easy to get distracted by the all-singing and dancing TOS 7, and gloss over some of the obvious misplays. What I liked most was how efficient this platform can be, something that often gets overlooked when talking about machines that can run for months or years without ever stopping.
But the yin to that yang is that there isn’t a huge amount of power to throw at Docker containers or AI, and that’s where many brands are taking their devices.
I’m also confused why this machine ended up with two 5GbE LAN ports when its competitors are delivering 10GbE. The internal M.2 slots and the four SATA drives are all geared towards supporting a single 1GB/s data flow, so why would you split it into two?
And, for the billionth time of asking, where are the drive tray locks, the apps that can output via HDMI, and better ports than USB 3.2? After ten years of NAS development, two of these questions are long overdue for an answer.
Don’t get me wrong, there are good things here for those who don’t need computational power on the NAS and like cool running, but it feels like TerraMaster is trying to put off the evil day when its next NAS will need to deliver hardware with a wider appeal than the F4-425 Pro.
(Image credit: Mark Pickavance)Should you buy a TerraMaster F4-425 Pro?Value
Not a great spec considering the price
3.5 / 5
Design
Old enclosure, and no bay locks or numbers
4 / 5
Features
Thre M.2 slots and dual 5GbE LAN
4 / 5
Software
TOS 7 is mostly great
4 / 5
Performance
Constrainted by Gen3x1 M.2 slots and 5GbE LAN ports
4 / 5
Overall
AN efficient four-bay, but hardly excisting
4 / 5
Buy it if...You want something flexible
Four drive bays have enough capacity to hold plenty of data, and the dual 5GbE LAN ports can distribute it effectively. And, potentially, it could be used to run an AI Agent.
You want minimal power consumption
The CPU in this NAS is nearly as power frugal as an ARM processor, but can still run X86 code.
Those looking for a small four-bay NAS that can stay cool when the going gets hot should take a closer look.
You want Thunderbolt
Thunderbolt and USB4 ports enable connecting fast external SSDs or HDDs to add even more storage. If you want Thunderbolt ports or PCIe expansion, then the F4-235 Pro isn’t for you, as it only has USB 3.2.
You need TOPS
Those working on local LLMs often need TOPS, or Tera Operations per Second. And, depending on the processor, and if it has an NPU (neural processing unit), the performance for AI is often rated in TOPS. The N350 is rated at only 2 TOPs, which is way below what the latest Intel and AMD mobile processors offer. They often achieve over 100 TOPS. View Deal
For more storage solutions we've tested the best NAS hard drives around
The government of California will now be able to use Anthropic’s Claude AI with a half price discount.
A press release published by California Governor Gavin Newsom says the state's government will also have access to free workforce training, expert GenAI technical assistance and workflow input from Anthropic developers.
“This partnership is about using technology the California way: responsibly, transparently, and in service of people. AI should not replace the human work of government; it should help our workers move faster, solve problems more effectively, and deliver better results for Californians,” Gov. Newson said.
Claude comes to CaliforniaThe discount for Claude also extends to local governments at the city and county level, allowing state workers with lower budgets to gain access to cutting edge tech. Gov. Newson said the technology will primarily be used for drafting, summarization, and analysis, while also “supplementing day-to-day work and improving services for Californians.”
Californian state agencies will be able to access Claude through the Statewide Information Technology Shared Services (SITeS), a new portal that centralizes AI tools for government use. The Californian government has already worked alongside Anthropic to integrate Claude into numerous tools for state workers, such as the Engaged California tool that provides Californians with more of a voice in policymaking.
Claude Security and Claude Code are also being integrated into the workflows of the California Department of Technology and the California Governor's Office of Emergency Services in order to improve cybersecurity. The California DMV will use Claude to reduce wait times and improve services, and the Department of Healthcare Services will use Claude to assist in the state’s Medicaid program.
The state is home to 33 of the top 50 private AI companies in the world, including Anthropic. “As a California company, we feel a real responsibility to our home state. We’re honored to expand our partnership with California’s agencies and to put Claude to work for the people who keep this state running,” said Kate Jensen, Anthropic’s Head of Americas.
“Building AI responsibly and in service of people has been our approach from the start, and that’s exactly what this partnership puts into practice.”
If you like the idea of owning the very best action camera but have so far been put off by the price, then let me bring you some fantastic news. This excellent value GoPro Hero 13 Black bundle is on sale at Amazon for $379.99 (was $479.99).
That's a new record-low price for the highly rated 5.3K resolution camera from the industry's leading brand. Considering the camera ordinarily retails for $429 by itself, this bundle deal is even more impressive.
In addition to the excellent GoPro camera, the bundle features a handle, two Enduro batteries, two curved adhesive mounts, a 64GB SanDisk MicroSD card, and a handy carrying case. That's everything needed to get filming as soon as it arrives at your door.
Today's best action camera dealHere's a super-tempting record-low price for one of the best action camera bundles in the world right now. The camera offers an impressive 5.3K video resolution inside a rugged and completely waterproof body. A sweet deal just in time for your summer adventures.View Deal
Competition for the top spot in our best action camera guide is fierce, but the GoPro Hero 13 Black has held prime position for a long time now. Its commendation is in large part thanks to the excellent-quality 5.3K video, a superb range of accessories, including new auto-detected Lens Mods, and improved battery life and heat dissipation.
You can read more about why we love it in our GoPro Hero 13 Black review. Specifically, we praised its Bluetooth audio support, versatile mounting options, and ability to capture great-looking footage in well-lit areas.
I don't record much video for social media specifically, but if you do, then you'll love the 27MP sensor, which enables footage to be recorded in an 8:7 aspect ratio. Say goodbye to the awkward cropping of landscape footage for portrait outputs.
55 years ago, the legendary Gene Wilder invited us to join him in "a world of pure imagination." Now, a disembodied AI recreation of his sonorous vocals is inviting us to join him in his "my newest wonders beyond imagination."
But there's little-to-nothing imaginative about the use of an AI-generated version of Wilder's voice to help sell Netflix subscribers on a new Willy Wonka and the Chocolate Factory-themed reality show, and, quite honestly, I hate everything about it.
To catch you up, this new show, which is set to air in September on one of the best streaming services, Netflix, borrows sets (recreations), themes, and even one living actor who played an Oompa Loompa from the original 1971 film to create a reality show that it hopes will somehow, if not match, approach the charm and danger of the original film.
Not getting the pointWilly Wonka and the Chocolate Factory, based on the Roald Dahl bookCharlie and the Chocolate Factory, is a commentary on wealth, poverty, family, risk, and cruelty, all wrapped in a candy-covered fantasy.
The protagonist, Charlie Bucket, is so poor he lives in a ramshackle London cottage with multiple generations of family members, including a grandfather who has not left his bed in years. Meanwhile, Wonka is finally opening up his secretive factory to a select group of the public, chosen at random via golden tickets hidden in some of his candy.
I know, it sounds like a confection, but the first scene in which Wilder appears gives you a hint that this is not your typical children's story. When all the winners, including Charlie, arrive, Wonka (Wilder) unsteadily walks out while using a cane to greet them. As he approaches the gate, Wilder loses his grip on the cane and falls forward, but instead of face-planting, he does an effortless tumble and is suddenly upright and boisterous.
I tell you all this because there is so much complexity and nuance in that one scene and virtually all of it engineered by Wilder's performance. Throughout the film, his face and especially his voice drive the narrative and mood, yo-yoing it between playful host and sinister guide who keeps losing children to one self-imposed calamity after another.
Never before or since have I seen a character in a children's film swing so wildly from gentle and childlike to ominous. It's a wonder the Academy didn't recognize his performance.
Fast forward more than half a century to this streaming show, which uses Wilder's voice with the permission of his family; you can get your first sample of it in the new show trailer (above). Created by increasingly well-known ElevenLabs, the voice is, in my opinion, a poor recreation at best, but I guess I might also call it unmistakably Wilder-like.
It's also depressing. The thing about Wilder's real voice in films like Willy Wonka, The Producers, Stir Crazy, and Young Frankenstein is that it could be wildly emotional. Wilder as a hysteric was often the funniest thing you had ever seen. No one went off the rails like Gene Wilder, but also few other actors could express such warmth and compassion, and sometimes he did it all in a single scene.
At the end of Willy Wonka and the Chocolate Factory, Wonka discovers Charlie is stealing his everlasting gobstopper to deliver it to Wonka's chief competitor. His anger at Charlie is terrifying, but as soon as Charlie leaves the candy behind, Wonka (Wilder) transforms, and the compassion and joy in his voice is something to behold.
The AI Wilder voice in this trailer is devoid of passion, pathos, or really any recognizable human emotion. It's a soulless Gene Wilder.
Look, there's no doubt the power of AI is transforming life and culture. It lets us do things we never thought possible. It's a co-worker, a companion, a sleuth, a coder, and a creator, but we are now at the point where we must ask ourselves: Just because AI can, should it?
I guess we can ask a similar question about the AI-generated Val Kilmer in the unfortunately titled As Deep as the Grave. While I can understand if Kilmer had completed some filming and they wanted to bridge the gap in scenes filmed after his death, this full-scale recreation, even with the estate's permission, seems wrong. I love these actors, but none are irreplaceable, and using their likenesses and voices simply to generate nostalgia and sell a product feels very wrong, and certainly not the best use of AI (and all the resources behind it).
Surely, we can agree that it's often better to let the dead stay dead and not try to resurrect a simulacrum of their essence through the power of AI. The results are, to be honest, rather like the work of Baron Von Frankenstein: a monstrous creation that probably has no place on earth.
While many of us wouldn’t think twice about buying software from a global hyperscale software vendor, open source software (OSS) has flown under the radar as a majorly important part of the software landscape for decades.
Recent shifts in European attitudes, fuelled by geopolitical tensions and technological sovereignty concerns, has seen an increased focus on the desire to run open source software and back local vendors. A major advantage is the open nature of its development – anyone can inspect, modify and use it, and companies can distribute their own versions without the limitations and expenses of vendor lock-in.
It’s so important that around 70% of modern software stacks are estimated to rely on open source components in some form – EXANTE sees it as a plumbing system that keeps software stacks together.
But that very benefit could also be a disadvantage for corporate customers, because small groups of volunteers don’t have the same requirements as large corporations. While companies invest heavily in cybersecurity and resilience, visibility into open source software can be alarmingly low.
AI is making underfunding even more pronouncedVibe coding has further complicated matters, with developers now able to write code more quickly and malicious actors leaning on those same AI coding tools to discover vulnerabilities and develop exploits.
Without the commercial funding that hyperscale software receives, it puts OSS in a more vulnerable position. But given the importance of OSS for companies looking to build their own software stacks, many are now starting to see backing projects as an integral part of managing their risks.
To discuss why the responsibility, or burden, of open source software sustainability is increasingly being picked up by enterprise customers instead of individual developers and contributors, I spoke with EXANTE CTO Richard Forss, who explained how AI is changing software development and cybersecurity strategies globally, and why businesses should treat it as critical infrastructure.
EXANTE is also the company behind the Gecko Fund – a new €1 million grant programme to support critical open source software projects used across trading and financial data systems with grants of between €10,000 and €150,000 available – and the company believes that financial and technical support for open-source projects is now more important than ever.
“We believe the industry that benefits from these tools should play a role in sustaining them,” Gecko Fund founder Anatoly Knyazev asserted.
Open source is the foundation of modern fintech and brokerage businesses, but almost nobody talks about it. If you looked under the hood of any brokerage platform, ours included, you would find that about 70% of the critical stack is open-source.
It powers the databases, operating systems, messaging layers, cloud infrastructure – the bits that quietly move millions of transactions a day. Clients never see any of it, which is rather the point.
The reason we all use it is simple economics. Software is built in layers, and there is no sense in any one firm rebuilding the foundations from scratch. We use shared, battle-tested components for things like market data processing and risk infrastructure, and we spend our own engineering effort on the parts that are actually ours.
It lets teams move quickly and solve complex architectural and engineering puzzles that would otherwise be out of reach.
Because it works too well to notice. Open source behaves like a utility, the plumbing in the walls, and nobody thinks about the plumbing until it bursts. Millions of organisations rely on these tools every day, and for years, almost none of them paid a penny towards keeping them alive.
More accurately, open source drives an estimated $8.8 trillion in global economic value. Yet, nearly two-thirds of the developers responsible for maintaining these widely adopted systems receive very limited financial support.
Hundreds of thousands of organisations – from startups to huge enterprises – benefit from these tools every single day, but the actual responsibility for maintaining them stays with a handful of volunteers.
Everyone relies on the infrastructure, but historically, almost no organisation has taken responsibility for paying for its upkeep.
Because it is a single point of failure hiding in plain sight. We spend enormous amounts of money assessing vendor risk and operational resilience, scrutinising suppliers, war-gaming outages, and then the whole edifice can rest on a library maintained by one or two exhausted people you have never heard of.
Our first Gecko Fund grant went to Kryo, an open-source data serialisation tool used all over the world in high-performance and trading environments. For years, it has been kept going by two people on two different continents, with no outside funding at all. They do excellent work – and that is, of course, not the problem. The problem is that it is two people. If one of them burns out, or simply does not have a free evening to patch a serious flaw the moment it appears, the consequences do not stay neatly within their project. They ripple out into regulated markets that have no idea they were depending on them.
AI has put its foot on the accelerator, and it is pressing down on both cars at once. On our side, it genuinely helps. It spots bugs, speeds up code review, catches things early. The problem is that the people trying to break in have exactly the same tools.
Bad actors use AI to scan open-source codebases, identify vulnerabilities, and launch targeted exploits much faster than before. The time between a flaw being discovered and an active attack being launched is continuously shrinking.
This puts intense pressure on underfunded maintainers to deploy fixes instantly. Since AI scales these threats so effectively, the security of the underlying open-source ecosystem has become a systemic priority.
It’s not the famous projects. Mostly the dull ones. The risk lives in the deep, unglamorous code: I’m talking about the low-level libraries, the developer tooling, the APIs – developer tools that operate entirely out of sight. The components that never trend on anything and never will.
That is precisely why they go unfunded. They do not make headlines, so they do not attract sponsors or donations, and yet they are wired into thousands of commercial products.
It is the paradox of the whole industry. The more essential a piece of software is, the more invisible it tends to be. When one of those hidden libraries fails, it does not fail for one company. It fails for everyone at once.
Stop calling it a hobby and start treating it as infrastructure, because that is what it is. The first thing any firm should do is the unglamorous work of auditing its own supply chain and understanding exactly which open-source components it actually depends on. Most have never properly looked. Once you know what you are standing on, you have to hold it up.
But money is not the only currency – firms can contribute engineering time, security audits, testing environments, or documentation. The goal is to make the relationship mutual: if your business profits from these tools, your business should help keep them safe.
The corporate world needs to stop expecting critical infrastructure to be maintained by someone giving up their weekends. That is the change. Everything else follows from it.
It is starting to shift. There are sovereign technology funds now, government-backed efforts, industry groups forming around open-source standards. That is encouraging.
But finance, of all sectors, ought to be leading rather than waiting, because finance has more to lose than most if this plumbing fails. If we formalise how companies contribute and build proper structures for funding this work, then the people holding up the machinery behind global commerce might finally get the stable backing they need to keep the markets running. They have earned it several times over.
Olivia Rodrigo’s been keeping busy. Just a few weeks after dropping her third album, You Seem Pretty Sad for a Girl So in Love, the pop sensation is making her next move. And it’s a pretty big one that’s good 4 all of us: she’s collaborating with Lego on not one, not three, but five brick-built sets.
In fact, she’s the first musician to get multiple dedicated Lego sets, and they’re all distinctly Olivia Rodrigo, complete with her iconic shade of purple. And maybe most importantly, Olivia is getting her own Minifigures.
Fans who saw Olivia Rodrigo on the Guts World Tour — or streamed the concert on Netflix — will recognize the iconic moon she flew over the audience in the Olivia Rodrigo's Concert Moon set, which is available to preorder for $49.99 / £44.99.
Not only does it include an Olivia Minifigure that can be placed on the purple moon, but the base is also a record player.
(Image credit: Lego)You can even use the knobs on the right side of the record player to make the disc spin, causing the moon to rotate. As with most Lego sets, this one is highly interactive, with plenty of hidden details packed into the full build.
If you’re just after a Lego brick-built record, Olivia Rodrigo’s Vinyl is worth a look at $34.99 / £24.99. Made from 360 pieces, it lets you build a vinyl record that can be displayed using a stand on the back. The front features several elements, including a red megaphone — as used on the Guts Tour — a butterfly, stars, and a spot for an Olivia Rodrigo Minifigure.
(Image credit: Lego)The flagship set in the Olivia Rodrigo Lego Editions line is the Dual Guitar, which is made up of 1,228 pieces and costs $119.99 / £109.99. You’ll build an iconic purple guitar that stands more than 20 inches tall, includes two Minifigures of Olivia Rodrigo, and features hidden details.
(Image credit: Lego)Best of all, it’s half acoustic and half electric, offering the best of both worlds. Beneath the electric side, you’ll find spots for both Minifigures, along with easter eggs that Olivia Rodrigo fans will appreciate. If you've ever wanted to get "obsessed" with the details, this is probably the set you'll spend the most time exploring.
Rounding out the Olivia Rodrigo Lego Editions line are the $79.99 / £69.99 Olivia Rodrigo's Secret Storage set and the $49.99 / £39.99 Flower Bouquet set. All five are available to preorder now from Lego’s online store and will begin shipping on August 1, 2026.
Lego's Olivia Rodrigo's Vinyl is also available for preorder in the UK at £24.99.View Deal
Lego's Olivia Rodrigo's Flower Bouquet is also available for preorder in the UK at £39.99.
Lego's Olivia Rodrigo's Concert Moon is also available for preorder in the UK at £44.99.
Lego's Olivia Rodrigo's Secret Storage is also available for preorder in the UK at £69.99.
Lego's Olivia Rodrigo's Dual Guitar is also available for preorder in the UK at £109.99.