Regardless of whether Intel would say it out loud, the Core Ultra 7 270K Plus and Core Ultra 5 250K Plus both represent an attempt to right the wrongs of the original Arrow Lake/Core Ultra 200S family. That bundle of chips was, necessarily, more power-efficient and cooler-running than the hotheaded 14th Gen models before them, though this came at the cost of hamstrung gaming performance. Rarely a desirable quality in a gaming CPU, that.

These two Core Ultra 200S Plus (or Arrow Lake Refresh) processors do, in comparison, achieve some appeal. They’re inexpensive and excellent multitaskers, and while they do still have efficiency on their silicon brains, Intel have looked to bump game speeds back up by rejigging their innards into a less latency-prone layout.

Alas, it’s not enough. Not only are Core Ultra 7 270K Plus and Core Ultra 5 250K Plus slower in games than AMD’s best chips, they once again fail to convincingly outpace Intel’s own back catalogue – the 2023 vintage 14th gen processors, included.

That’s especially unfortunate as the remedial measures Intel have taken for these chips are, on the face of it, sensible. Fingers may be pointed at the lack of major clock speed boosts: the 270K Plus’s 5.5GHz max boost clock is unchanged from the Core Ultra 7 265K’s, and the 250K Plus’ only rises to 5.3GHz, from 5.2GHz on the Core Ultra 5 245K. But previous Intel Core generations had been bumping up clock speeds for years, with little effect beyond increasingly troublesome temperatures.

Instead, the Core 200S Plus duo get 900MHz speed boosts to die-to-die (D2D) frequency, specifically addressing a problem with the original Core 200S models where the once-unified memory and compute controllers had been split into different tiles and spread across the processor. This is good for power efficiency, but introduces latency, as the disparate components need extra time to communicate with one another; an increase in D2D frequency should quicken this process, and get games performance back towards where it was when Intel chips were still monolithic.

These CPUs are also the first to wield Intel’s Binary Optimization Tool (apologies for the z, fellow Brits, that’s what it says here). A new component of the Intel Application Optimization (gargh) app, this tool supposedly speeds up certain games that were designed with older processors in mind. Or, rather, it speeds up how the Core Ultra 7 270K Plus and Intel Core Ultra 5 250K Plus can process them, producing game-specific increases in instructions per clock (IPC). That basically means the chips can get more done in each CPU clock cycle, uplifting performance without needing to raise clock speeds (and thus temperatures) themselves. It’s clever stuff, on paper.

In terms of testing all of this out, I may have slightly fired a .44 magnum cartridge directly through the ol’ metatarsals by being so disinterested in the original Arrow Lake chips that I never recorded proper benchmark results for them. For the requisite bar graph competition, though, I still think it makes sense to hold these new parts up to the 14th genners, which remain both the more likely subjects of “Should I upgrade?” ponderings and the most reliably high-performing Intel chips of the past few years. I’ve also thrown in the AMD Ryzen 7 9800X3D – if Intel are serious about the Core Ultra 7 270K Plus being their “fastest ever” gaming CPU, then they surely won’t mind comparisons to a top performer from the Ryzen camp.

Thus, gaze ye upon gaming benchmarks:

This is obviously not great reading, from the Arrow Lake Refresh perspective. The two new chips could be called an improvement, but only in the sense that they are now roughly as fast as the 14th generation, not repeatedly worse. And they still fall behind in Total War: Warhammer III, the Core Ultra 5 250K Plus not even equalling the years-old Core i5-14600K.

The only real win here is in Cyberpunk 2077, where the 250K Plus does actually beat its ancestor by a meaningful margin, and the 270K Plus just sneaks ahead of the former flagship Core i9-14900K. Credit where it’s due, that feat is repeated in Returnal and very nearly achieved in Warhammer III. Given the newer CPU is also a good 10% slower in F1 2024, mind, that “fastest ever” claim comes with a big, bold-type asterisk . It’s also, quite clearly, nowhere near the Ryzen 7 9800X3D’s level.

Binary Optimization, it turns out, is of limited usefulness as well. Part of that is just inherent to the tool’s lack of widespread game support (barely over a dozen, at launch), but even when it works, it’s not much of a magic go-faster button. In Cyberpunk 2077, enabling the tool nudged the 270K Plus up from 233fps to 243fps: a mere 4% boost, in a game that was already more generous to Arrow Lake Refresh than most. The 250K Plus also climbed from 224fps to 233fps, another teeny 4% rise, though this does at least show – if you’re more inclined towards mid-rangers – that Binary Optimization can at least get Core Ultra 5 hardware up to Core Ultra 7 standards of performance. In certain games. And conditions. And assuming you don’t just enable the same tool on an Ultra 7 as well.

All the same, I can’t bring myself to get as huffy about Arrow Lake Refresh as I did about Intel’s preceding misfire. Factoring into that is the apparently earnest attempt at making amends on game performance, if it was ultimately a doomed one. More practically, there are other things this CPU duo does much, much better.

Multitasking and productivity are two. Get your PC doing something that can take better advantage of multiple threads than games, be it editing, encoding, or if you’re into that sort of thing, livestreaming, and both the Core Ultra 270K Plus and Core Ultra 250K Plus will juggle the workloads far more deftly than their older alternatives. Even if the 270K Plus isn’t entirely Intel’s fastest gaming CPU, it might well be their highest-scoring mainstream CPU in Cinebench’s simulated benchmark tests, and the 250K Plus finally nails down an outright victory over the Core i5-14600K here too.

Again, I don’t have results for their original 200S equivalents, though smart money is still on the newer CPUs: both have an additional four E-cores (E for Efficiency) over their pre-Refresh equivalents. While these aren’t as big and beefy as their accompanying P-cores (P for Performance), they still add four extra threads for the CPU to play with, and with minimal extra strain on power or thermals.

Speaking of, both chips remain nicely cool, even on the budget-friendly Thermalright Peerless Assassin 120 SE air cooler I’d tested them with. Whereas the same setup had the Core i9-14900K’s P-cores peaking at 90°c in games, even hitting the danger zone of 100°c during Cinebench runs, the Core Ultra 7 270K Plus only went as high as 76°c while playing, and stuck to a safe 90°c at its benchmarking peak. The Core Ultra 250K Plus was even chillier, peaking at 68°c in games (down from 77c° on the Core i5-14600K) and 74°c in Cinebench (down from 82°c). With the exception of the Ryzen 7 9800X3D’s 88°c peak temp in Cinebench, both chips were also consistently cooler than that AMD bigshot.

The fact that these processors can run on cheap coolers is particularly welcome at a time when memory shortages are driving component prices – and by extension, the cost of new PC builds – painfully skywards. And that’s not to mention how aggressively priced these CPUs are themselves: at $300 for the 270K Plus and $200 for the 250K Plus, these average around $100 less than the equivalent Arrow Lake pieces, with the former notably coming in nearly $200 cheaper than the Ryzen 7 9800X3D. That, alone, could pay for a decently specced motherboard (if not, regretfully, a good set of RAM) to go with it.

This is expected to be the last generation of Intel CPUs to fit the LGA 1851 socket, so any new mobo won’t get to carry over any future chip upgrades. But then, the CPU isn’t the kind of component you’ll be swapping out frequently anyway.

The Core Ultra 7 270K Plus and Core Ultra 5 250K Plus do have their place, then, as affordable multithreading engines. It’s also encouraging to see that Intel really have got a grip on temperatures; the 14900K, especially, was way too ready to hit boiling point.

It’s just that for games, specifically, they’re still not preferable to a shiny new Ryzen, or even to upgrading from any reasonably non-ancient Intel platform. Right now, you’re not going to see worthwhile improvements over an existing 14th or likely even 13th gen Core equivalent, while AMD’s 3D V-Cache tech still has Intel licked at the enthusiast end.

On that note, the most prudent buying decision of all may be to simply wait. Intel have yet another CPU family, Nova Lake, set to launch later this very year, and word is they’ll have their own answer to 3D V-Cache in the form of BLLC (Big Last-Level Cache). Throwing vast memory stores at game framerates may not sound as elegant as something like D2D optimisations, but hey, it’s been working for AMD for years – and it’s clearly time for Intel to try something different, at least to themselves.

Source link

Arm CEO René Haas confirmed at Computex that ByteDance and Oracle have joined Meta as customers for Arm’s own data-centre CPU, validating the company’s shift from licensor to silicon vendor.

Arm chief executive René Haas confirmed at Computex on Monday that ByteDance and Oracle are among the customers using AGI, Arm’s first in-house data-centre CPU, joining Meta as the third and fourth named adopters of a chip that the Cambridge-based company is positioning as the structural alternative to Intel’s Xeon and AMD’s EPYC server lines.

The strategic shift the customer announcement validates is the harder of the two parts of the story. Arm spent the past three decades licensing CPU IP to chipmakers who then designed and sold their own silicon, AWS’s Graviton, Microsoft’s Cobalt, Google’s Axion, Nvidia’s Grace and now Vera.

AGI represents the company’s decision to design and sell finished silicon directly. The customers Arm is now announcing for that finished chip are not its existing licensees; they are the same hyperscaler and enterprise-cloud buyers its licensees would themselves have hoped to sell to.

The of EU tech

The latest rumblings from the EU tech scene, a story from our wise ol’ founder Boris, and some questionable AI art. It’s free, every week, in your inbox. Sign up now!

The Arm-AGI launch is therefore as much a vertical-integration play against its own customer base as it is a horizontal-integration play against Intel and AMD.

The customer roster tells the story. Meta was the first publicly named AGI customer, announced at the AGI launch event in San Francisco in March. ByteDance is the largest Chinese AI workload customer, which is significant given the company’s parallel custom-CPU programme on Arm and RISC-V tracks reported last week.

Oracle is the enterprise-cloud workhorse that pairs Arm-based servers with its database product line for customer deployments. The composition is meaningful: one US frontier-AI lab, one Chinese hyperscaler, one US enterprise-cloud provider. Arm has, on three named customers, demonstrated that AGI is being adopted across the three most strategically distinct categories of data-centre buyer.

The financial-projection backdrop matters. Haas said at the AGI launch that sales of the chip alone would bring in $15bn to Arm by 2031. The chip is co-designed with Meta and built on a chiplet design using TSMC’s 3nm N3P process, the same node Nvidia uses for Rubin.

The $15bn projection is, on the most aggressive read, a doubling of Arm’s current annual revenue base if achieved. The Monday customer announcement is the most concrete validation yet that the projection is supportable.

The structural read on the broader CPU market is the editorial point worth surfacing. Nvidia’s Vera CPU launched yesterday with OpenAI, Anthropic and SpaceX as named customers.

Snowflake’s $6bn AWS Graviton commitment last week added a major enterprise-data-platform customer to the Arm-server-side ledger. ByteDance is building its own custom Arm-and-RISC-V CPUs in parallel. Arm itself is now selling AGI to Meta, ByteDance and Oracle.

Every named major AI-data-centre customer on the public record is now committed to Arm-based silicon in at least one tier of their compute stack, either as a direct purchase, a hyperscaler-built custom design, or a co-developed product. The x86 incumbents have, on the available evidence, lost the hyperscaler-CPU war in the four weeks since Computex 2026 began.

The implication for Intel and AMD is severe. Both companies have historically derived the majority of their server-CPU revenue from hyperscaler purchases, and the hyperscaler portion of the data-centre CPU market is structurally the highest-margin tier. The current public commitments to Arm silicon, in aggregate, materially compress that segment of the x86 incumbents’ addressable market.

The compression is not yet visible in Intel’s and AMD’s reported revenue because hyperscaler purchasing happens on multi-year cycles, but the trajectory through 2028 is now clear.

The harder question for Arm is whether selling finished silicon to its existing licensees’ customers will compress its core IP-licensing revenue line.

The hyperscalers that previously paid Arm royalties through their own custom CPUs (Graviton, Cobalt, Axion) may now have less reason to maintain those programmes if Arm’s in-house AGI design covers the same workloads at comparable price-performance. The vertical-integration play, in other words, is its own competitive risk.

Arm shares were modestly higher in pre-market trading on the announcement. The AGI chip is shipping now to all three named customers.

[ad_2]

Source link

When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.

[ad_2]

Source link