We tested Intel's new chips for cash-strapped hardcore PC users and they're impressive

REVIEW It's a tough time to be a PC enthusiast. Between the memory crunch and the AI boom driving up prices on storage, DDR5, and GPUs, it's gotten prohibitively expensive to build a PC.

Amid this turmoil, Intel hopes to win some goodwill from budget-conscious customers with its newly announced Core Ultra 200S Plus family of desktop processors.

The new chips boast higher core counts, more aggressive frequency curves, and, more importantly, are launching at much lower prices this time around. At $199 and $299 respectively, Intel's all-new Ultra 5 250K and Ultra 7 270K reflect a level of market awareness that we haven't seen from Chipzilla in quite a while.

Intel's Arrow Lake refresh is its most compelling value proposition in years. And, while the chips can't contend with AMD's cache-stacked X3D parts in gaming, they're significantly cheaper while also delivering strong performance in production workloads thanks to all those extra efficiency cores.

In this review, we'll be digging into the good, bad, and the ugly of Intel's latest generation of desktop CPUs covering everything from office productivity to HPC, and yes, gaming.

But first let's take a closer look at Intel's latest chips.

The chips

The overarching theme for Intel's Arrow Lake refresh is more cores per dollar. Both Core Ultra 5 and Ultra 7 Plus processors gain four additional efficiency (E-cores) over last gen.

With 24 cores (8-P and 16-E cores), this puts the 270K in direct contention with Intel's 285K. The new Ultra 7 doesn't clock as high, with a max turbo 200 MHz slower than the flagship. But, for many, those lower clocks are more than worth the lower MSRP.

The new Core Ultra 5 250K enjoys similar gains. The chip now boasts 6-P cores and 12-E cores for a total of 18 cores and 18 threads. Remember that there's no hyperthreading (SMT) this generation. At least as far as core count goes, the part is essentially a Core Ultra 7 265K that's had two of its P-cores fused off and its frequency tables remapped.

Apart from the higher core count, the parts don't look all that different from their pre-refresh siblings, but the architectural improvements actually go deeper than the speeds and feeds might suggest.

Intel has tweaked its boost algorithm so that, under sustained all-core loads, the chips can now maintain higher clocks overall. When all six of its P-cores are loaded up, the 250K can now maintain a 5.1 GHz core clock, up 100 MHz from the prior gen. Meanwhile, the 270K's E-cores now clock 100 MHz faster, while the all-P-core boost table has been updated to more closely resemble the 285K at 5.4 GHz, up from 5.2 GHz on the 265K.

Alongside the reworked boost tables, Intel bumped the die-to-die fabric frequency by 900 MHz. Arrow Lake saw Intel bring its chiplet – or, as Intel prefers, tile – based design philosophy to the desktop for the first time.

Not much has changed here. As we understand it, the tiles are still fabbed by TSMC and packaged in-house using Intel's Foveros 3D packaging tech. However, this means that Arrow Lake's memory controller is on a different tile than its P or E-cores. By increasing the fabric bandwidth, Intel aims to drive down memory latency, improving overall performance in the process.

On the topic of memory, Intel's Arrow Lake refresh natively supports DDR5 JEDEC speeds up to 7,200 MT/s, with 8,000 MT/s memory officially sanctioned via XMP profiles. That's a substantial uplift over Intel's original crop of Arrow Lake processors which officially topped out at 6,400 MT/s, though higher memory speeds were possible by overclocking.

And if you've got exceedingly deep pockets, Intel has also added support for 4-rank memory modules up to 128 GB each. We don't imagine many folks will actually take advantage of this capability, as just one of those DIMMs will almost certainly retail for several times the cost of a 270K.

According to Intel, these enhancements give the parts a whopping 83-103 percent multi-threaded performance advantage over AMD's entry-level 9600X and mid-tier 9700X processors, at least in rendering and synthetic benchmarks. It's easy to make comparisons like this when you slash the price of your products to undercut the competition.

The 270K's higher core count doesn't help nearly as much in gaming, with Intel claiming a four percent advantage in average FPS over the Ryzen 7 9700X. Meanwhile, Intel only asserts that its $199 250K ties AMD's 9600X across its suite of games.

BOT and the chipset driver

While some of the 200S Plus series' performance uplift comes from architectural refinements, much of the chip's gaming performance can be attributed to software enhancements.

Intel has a new Platform Performance Package, which bundles up all the libraries, performance profiling, power management, and application optimizations into a single installer to make getting up and running less of a headache.

Alongside the installer is what Intel is calling its Binary Optimization Tool (BOT), which leverages its compiler and profiler tech to reduce execution overheads and boost instructions per cycle for supported x86 binaries at runtime.

On average, Intel says BOT delivers an 8 percent uplift in gaming with some titles showing FPS gains of more than 22 percent.

BOT works by using the chipmaker's hardware and software profilers to analyze pre-compiled binaries for sources of cache misses, front-end stalls, and mis-predictions, and other hiccups that kill performance at runtime. This information is then used to develop alternative binaries optimized for Intel hardware which are then shipped as part of the Platform Performance Package.

"We don't see source code. We don't change source code. We do not reverse engineer. We do not recompile. Everything that the workload was originally designed to do stays in the binary," Robert Hallock, VP of Intel's enthusiast channel biz tells us. "It's akin to shader replacement for a graphics card. You've got a much faster, more optimized shader for that graphics pipeline, we're doing the same thing on a CPU."

Because of this, support is limited to select games at launch, and users will need to manually toggle it on in the Intel Application Optimization utility. Intel is also being intentionally cautious so as not to accidentally trigger anti-cheat software, which means it's not yet available for online titles.

Our test setup

Intel sent over two of its latest Core Ultra 200S Plus series processors for review along with a kit of speedy 7200 MT/s memory that was bristling with LEDs to help it shine.

For testing, we ran Intel's Core Ultra 7 270K and Ultra 5 250K using a kit of 7200MT/s DDR5 using the Intel default performance profile on our motherboard.

Our primary focus is to compare the 270K and 250K against Intel's older 265K. Intel was also gracious enough to send over a 245K to fill out our CPU roster. Unfortunately, it didn't arrive quite in time for publication; we'll be sure to update this story with additional insights when we can.

Alongside the Intel processors in our test suite, we also tested AMD's Ryzen 7 7800X3D and Ryzen 9 9900X using a pair of DDR5 6000 MT/s DIMMs. Technically these parts only support JEDEC speeds of 5200 and 5600 MT/s, respectively, but DDR5 6000 has been the recommended config for years as it allows for a 1:1 ratio between the memory clock and controller.

Neither of these chips are particularly price competitive with Intel's latest processors, but this is our first CPU review using our new test methodology and we happened to have them lying around.

As you'll see below, these parts do a nice job of highlighting the strengths and weaknesses of Intel's Arrow Lake refresh. In many cases, Intel's Arrow Lake Refresh parts punch well above their weight class.

For gaming benchmarks, we used an RTX Pro 6000 Ada. Yes, we're aware using a 48 GB workstation card is an odd choice, but with 10 percent more CUDA cores than a 4090, it was the best GPU at our disposal and the one least likely to introduce performance bottlenecks.

As much as we would have liked to conduct these tests on a 5090-class GPU and eliminate any possibility of CPU bottlenecks, given the current climate, we've got to make do with what we have, just like everyone else.

Setting expectations

Before we dive into production benchmarks and gaming, we'll take a look at how these chips perform in synthetic benchmarks to get a baseline.

While these benchmarks don't tell us much about how the chip will perform in real-world applications they can help us understand whether the parts are performing as intended and offer insights as to why one part may come out ahead of another at a particular workload.

Frequency validation

We began our testing by validating Intel's power and frequency claims. As we mentioned earlier, for Arrow Lake, Intel has not only boosted the core count, but the all-core boost frequencies across all of these parts.

In single-threaded workloads the 250K or 270K had no problem hitting their respective max turbo frequencies of 5.3 and 5.4 GHz in either integer or AVX2 heavy workloads.

Note: At the time of publication, our frequency and power validation benchmarks only run on Ubuntu 25.10 Desktop. This provides a less noisy and more repeatable environment for frequency and power monitoring, but we are investigating ways to replicate its functionality on Windows in the future. All other tests in this review were conducted using Windows 11 25H2, unless otherwise noted.

It was a similar story for multithreaded workloads. In integer testing, both chips quickly reached their claimed all P/E-core boost frequencies. This was largely true of our SIMD-heavy primesieve load. We did notice that the 270K's P-cores fell about 20 MHz short of their advertised boost, close enough that we're willing to give it to them.

At least in terms of clock speeds, Intel's Arrow Lake refresh parts do exactly what they claim.

Please note that, while we included package temps in these reports, these tests weren't conducted in a temperature-controlled environment. Instead, these data points were included to help identify any potential issues with throttling, which with our 360 mm all-in-one liquid cooler didn't end up being an issue.

Power

Over the past few years, Intel's desktop processors have earned a reputation for being power hogs. With the launch of Arrow Lake, Intel made a big deal about how much lower the chip's real-world power consumption was under load compared to prior generations.

Despite these claims, both Arrow Lake and the Arrow Lake Refresh boast the same 159 W and 250 W max turbo power limits as their older 14th-gen siblings. Depending on how religiously motherboard vendors adhere to these limits is also something to watch out for, as max power consumption could end up even higher if misconfigured out of the box.

At least on the MSI Pro X890-A WiFi used in our review, this wasn't a problem, with neither chip exceeding their max-rated TDPs.

Our CPU power profile bench measured the reported package power in both single and multi-threaded integer and SIMD (AVX) heavy workloads. While it would be more accurate to measure power at the EPS 12 volt cable, we currently lack the equipment to do so, and changes to how power is supplied to CPUs on some boards further complicates the matter. So, for now, we're stuck with the package power as reported by S-TUI.

At idle, both CPUs sip power, dropping to around 10 watts reported. Under a single-threaded load, power consumption jumped to between 20 and 24 watts for our integer load and 26-32 watts for AVX-heavy instructions.

Under an all-core integer load, power consumption increased significantly to 88 watts on the 250K and just over 150 watts on the 270K, while under an all-AVX primesieve stresser, both parts leap to the max advertised turbo power limits.

While the chips are still power hungry compared to AMD's 9000-series, we'll note that these all-core workloads aren't indicative of many real-world desktop applications. With 18 and 24 cores respectively, the likelihood of all of those cores being loaded up with an all AVX workload is relatively small.

The fact that Intel's chips can pull so much power when they need to also means that, on the odd chance you do run into such a workload, for example a Blender render or Handbrake transcode, using all those cores won't mean cutting back on clocks.

Memory bandwidth

Memory bandwidth is a major bottleneck for a lot of modern applications, so it's nice to see that Intel's new parts can actually utilize a decent chunk of the bandwidth DDR5 7200 has to offer.

Stream is a benchmark commonly used in the HPC community to get a sense of real-world memory bandwidth. While it's trivial to calculate the theoretical bandwidth of a system, memory subsystems and access patterns often make harnessing all that bandwidth difficult.

In our testing, the 250K and 270K performed essentially identical, delivering 88 GB/s in the Stream Triad benchmark out of 112.5 theoretical. Note that Stream doesn't account for write-allocate overheads, so a correction factor of 1.33x is applied to account for this.

Geekbench 6.6

In Geekbench 6.6 with BOT disabled, the 270K manages a single core score of 3,356 and a multi-core score of 23,372, giving it a 9-10 percent advantage over the older Core Ultra 7 265K. That's a very respectable uplift for a mid-cycle refresh, especially considering the price cut.

The chip's single core performance is arguably the more important metric here as it is more representative of day-to-day tasks. In this respect, the 270K performs with the margin of error of the Zen 5 cores in the 9900X.

The new Core Ultra 5 250K beats the prior-gen Ultra 7 in single-threaded performance and nearly matches it in multi-core. Both of the new Intel parts pull ahead of the 9900X and 7800X3D, which as we'll see later, is why any one benchmark can never give a full picture of a chip's performance.

Primesieve

There are few better burn-in tests than a primesieve. The idea is simple: count the primes up to a certain threshold, the first one done wins.

While not a common or particularly representative workload for the average Joe, primesieve can be useful for quantifying SIMD performance (SVE, SSE, AVX2, AVX-512, and so on). In this case, we use primesieve to count prime numbers up to 1 trillion, which is then expressed as millions of primes per second.

Once again, we see a similar dynamic as with Geekbench. Intel's core heavy architecture gives it a leg up in the multithreaded test, churning through 5,837 million primes a second. But while Intel dominates in multithreaded tasks, AMD Zen 4 and 5 cores show a distinct advantage for SIMD workloads. In the case of the 9900X, Zen 5's support for AVX-512 gives the part a 36.6 percent advantage in single-threaded SIMD workloads, which is what enables the part to keep up despite having half the cores.

The Core Ultra 5 250K still performs well here, but isn't a SIMD powerhouse by any means, but it's still doing better than the 7800X3D, which while supporting AVX-512 has to double pump its 256-bit data paths to do it.

7-zip compression/decompression

7-Zip isn't really a synthetic benchmark, it is an integer heavy data compression and decompression workload. What's important is that it can tell us just as much about how a CPU will perform compressing and decompressing files as it can about day-to-day tasks.

7-zip LZMA compression is another strong showing for the Core Ultra 7 270K. Single-threaded performance is with 5 percent of matching AMD's 9900X, while in multi-threaded workloads it again pulls ahead. Those E-cores may not be the fastest, but with enough of them and an easily parallelized workload, they certainly get the job done.

Compared to the 265K, the 270K is nearly 20 percent faster overall. Meanwhile, the 18-core 250K nearly matches the older Core Ultra 7.

Looking at decode performance, AMD's 9900X pulls ahead of the 270K, but its lead is easily within the margin of error – enough for us to call this a tie.

Moving down the chart, the four additional E-cores on the 270K really do make a difference in our all-core tests. The 250K once again trails, but not as much as you'd think. It's now within spitting distance of the 265K.

Speedometer 3.1

Since so many of the applications we use are now web-based, benchmarks like Speedometer can give us a sense of whether a chip is going to feel noticeably snappier in day to day use.

Across both Firefox and Chrome, the Core Ultra 7 270K is 9 percent faster than the 265K, with the refreshed Core Ultra 5 following not far behind. While it may be a measurable improvement, at least for the chips in our test suite, your preferred browser is going to have a bigger impact on performance than the CPU.

What this actually tells us

These benchmarks give a rough sense of just how much of an improvement Intel has managed to squeeze from the Arrow Lake platform since its debut. With the exception of SIMD-heavy workloads, Intel has managed to close the gap with AMD on single-threaded performance.

Combined with the additional E-cores, both chips should play in a much higher class than their predecessors.

However, synthetic benchmarks can only tell us so much. So let's take a look at how these chips perform in the real world.

Office and productivity

To see how these chips fare in more mundane tasks, we use LibreOffice to convert 50 image-heavy ODT files into PDFs sequentially. This is an exceptionally lightly threaded workload that tends to favor higher clock speeds.

As we saw with the canned benchmarks, Intel's Arrow Lake refresh parts roughly match AMD's Zen 5 cores on single-threaded performance, and that's reflected in the tight grouping here. While there is a measurable difference from chip to chip, it's not a meaningful one.

We expect this could change as we add more chips to our test suite, but for the moment, none of these parts are going to feel slow in everyday office tasks.

HandBrake x265 transcoding

Moving on to something heavier-duty, our HandBrake X265 transcoding benchmark really gives Intel's Core Ultra 7 270K a chance to stretch its legs, or rather its cores.

This test measures how quickly a 10-minute 4K 60 FPS H.264 video file can be transcoded to 1080p using the x265 video encoder at the medium preset and a constant quality of 18. We'll emphasize we're not using the QuickSync hardware accelerators in this test. While faster, these hardware accelerators trade quality for speed. Our test runs entirely on the CPU.

The 270K's higher all-core boost clocks and additional E-Cores contribute to a 9 percent gen-on-gen increase over the older Ultra 7 in transcode performance.

The 250K also performs well here, achieving performance within spitting distance of the 9900X and 265K. It's a good showing for a $200 part.

3D rendering

Blender is arguably the most popular and widely used of the 3D modeling and rendering software suites, which is why we prefer it over something like Cinebench, which is commonly used as a synthetic benchmark for measuring CPU performance rather than their intended purpose.

In this test, we measure the CPU's rendering performance in samples per minute (SPM) across the three scenes: Classroom, Monster, and Junkshop.

Blender is extremely well threaded, so it's no surprise the 24-core Core Ultra 7 270K takes top marks in this test, demonstrating a roughly 25 percent lead over the 265K and Ryzen 9 9900X.

The 250K's lower core count puts it toward the back of the pack, beating out the 7800X3D – that 3D V-Cache just doesn't help in this scenario – but it falls behind the 9900X and 265K.

Code compilation

According to cartoonist Randall Munroe, "the #1 programmer excuse for legitimately slacking off" is "my code's compiling."

Ergo, the logic follows that "my code will compile faster" is a great reason for why your boss should sign off on a new workstation.

As modern computing workloads go, code compilation is one of the more computationally expensive, often taking several minutes or even hours to compile large projects from source.

To test CPU performance in this task, we compile LLVM from source using Clang, the LLD linker, and Ninja via MSYS2 on Windows.

Code compilation is an integer heavy workload that scales quite well with core count and clocks. SIMD instructions like AVX or SSE usually don't come into play here.

As such, the 270K performs extremely well here, leading the competition with a compile time of 5 minutes and 30 seconds, beating out the 12-core 9900X and the 20-core 265K by roughly a minute and a half.

The Core Ultra 5 250K's 18 cores put it toward the back of the pack, but it still manages to beat out the 7800X3D in this task.

If your job involves compiling code from source on a daily basis but you don't have the cash for a Threadripper or Xeon workstation build, then the 270K might be worth considering.

HPC and simulation

Neither the 250K or 270K are workstation parts. They lack the PCIe bandwidth and feature sets you'd expect from a workstation-class product.

But with 18 and 24 cores respectively, we can easily see folks considering these chips over Intel or AMD's more expensive Xeon 600 and Threadripper processors for scientific applications as well as some local AI.

To test this, we benchmarked the chips against two common HPC simulations: GROMACS for molecular dynamics and OpenFOAM for computational fluid dynamics (CFD), both running in Ubuntu 25.10.

Molecular dynamics

GROMACS is commonly used to simulate proteins, lipids, and nucleic acids. In this case, we're measuring how quickly the chips can complete our Alcohol Dehydrogenase benchmark, which simulates a 134,000 atom system for 50,000 time steps and returns a result in nanoseconds per day, with more being better.

While this workload can be run effectively on modern GPUs, it still provides a baseline for CPU performance in molecular dynamics simulations. In this test, we see a roughly 10 percent uplift in performance from the 270K over the older Core Ultra 7. However, even with twice the cores, the chip still trails the 9900X by almost 14 percent.

The 9900X's larger number of performance cores and support for AVX-512 is likely the deciding factor here. Intel's consumer products lack support for the instruction set and their higher total core counts simply aren't enough to make the difference.

While this does seem to benefit AMD's 9000-series silicon, it doesn't help the 7800X3D, which trails all three of the Intel chips in this particular benchmark.

Computational fluid dynamics

By comparison, in the OpenFOAM 13 motorbike benchmark, which simulates the flow of air around a motorcycle, we see the Intel parts take the lead.

The 270K completes the simulation in about 11 minutes while the 265K and 250K follow about 19 and 33 seconds later. However, what we're seeing here isn't a compute advantage so much as a bandwidth one.

CFD is heavily dependent on fast memory. In addition to supporting DDR5 7200 MT/s out of the box, Intel's latest chips demonstrate far better memory utilization, achieving nearly 80 percent of theoretical bandwidth compared to around 66 percent on the AMD platform.

Local AI

Local inference on large language models like GPT-OSS 20B running in Llama.cpp on the CPU illustrates the relationship between compute and memory bandwidth quite nicely.

The faster memory subsystem on Intel's latest chips gives them a leg up in token generation, while AVX-512 support on AMD's Ryzen 9 CPUs translates into a faster time to first token.

LLMs aren't typically something we recommend running on the CPU unless you have to. However, the same isn't necessarily true of OpenAI's Whisper models in Whisper.cpp.

At least on Windows running the precompiled binaries, the 270K achieved a 6.3x speed up when transcribing a 10-minute audio clip, beating out the 265K, 9900X, and 250K at 5.4x and 5.2x real-time. The 7800X3D, while a killer gaming chip, came in last place in our testing.

Curiously we were able to achieve better performance on the 9900X by compiling Whisper.cpp from source on Linux, but weren't able to replicate this on Windows.

Gaming performance

The Register isn't a gaming-focused publication, but it's a workload most folks considering these parts are going to care about, so let's talk about it anyway.

Intel boasts that the 270K is its best gaming processor yet, and has seen fit to imbue some new capabilities, most notably its new binary optimization tool, in order to boost frame rates in a variety of titles.

The chipmaker claims to deliver a 13-15 percent uplift in gaming performance over its pre-refresh Arrow Lake parts. This largely aligned with the results of our testing, which found the 270K delivered an average of 9.4 and 15.5 percent higher frame rates than the prior-gen 265K, the biggest gains in titles that support their binary optimization tech.

We tested five titles at 1080p high settings. Unless otherwise noted, we disabled all ray tracing or AI upscaling features for much the same reason.

Kicking things off with Cyberpunk 2077, the 270K Plus delivers a modest 6.2 percent uplift in performance out of the box over the 265K, which jumps to 9 percent when BOT is turned on.

However, more impressively, the 250K not only bests the prior-gen 265K, but is just 3 percent slower than the 270K. For a chip starting at $199 (less if you opt for the KF variant), this is particularly impressive. Having said that, the 270K does manage to maintain higher frame rates at the low end.

The new chip's higher all-core boost clocks certainly appear to be helping matters, though this is definitely one of the weaker titles with support for BOT at launch.

Neither chip can catch AMD's nearly three-year-old 7800X3D in this title, though they do come close. The House of Zen's 3D chip stacking tech has aged remarkably well, something that's reflected in its price. The chip currently retails for between $349 and $379, if you can find it in stock.

All three Intel chips in our testing managed to outperform AMD's Ryzen 9 9900X in Cyberpunk. This appears to be down to the core-parking tech introduced with this generation, which effectively turns the part into a higher clocking 9600X to avoid die-to-die performance penalties when gaming.

In Total War Warhammer III, the 270K's higher boost clocks do give it a slight edge over its predecessor in our testing, though we're really splitting hairs here. The four additional E-cores and higher all-core boost clocks just don't make that much of a difference.

Once again, both chips trail the gaming-focused Ryzen 7800X3D in average frame rates, but actually managed to eke out a lead in the lows. Again, we wouldn't read too much into that because the memory configurations are different across these chips. As for the 9900X, AMD's core parking tech doesn't do the chip any favors in the title.

Moving on to F1 2025, the Core Ultra 5 250K managed to pull ahead of the older 265K in both average FPS and in the lows. The pricier 270K meanwhile delivers an 11 percent uplift in average FPS while maintaining higher frame rates in the lows compared to its predecessor.

Increased boost clocks and added e-cores definitely seem to be helping here. Unfortunately, Intel's architecture optimizations aren't enough to overcome AMD in this test, with the 7800X3D leading the 270K by 13.3 percent in average FPS. The 270K does, however, pull ahead in the lows.

Borderlands 3 is an older title, but it scales well with core count and is one of a handful of titles in our library that support Intel's BOT tech.

Once again, the Core Ultra 5 250K manages to best the older 265K by about 10 percent in average frame rates. The 270K's gains are even more impressive, jumping nearly 23 percent over the older Core Ultra 7.

Intel's binary optimization is doing its job here, delivering a 5 percent boost compared to leaving it off. The gains are enough that the 270K nearly manages to catch the 9900X in our testing, though the cache-strapped 7800X3D remains in a class of its own.

Shadow of the Tomb Raider is the oldest game in our test suite, but it scales even better with higher core counts and shows by far the largest gains of any BOT-enabled title.

Gen-on-gen against the 265K, the all-new 270K delivers a nearly 33 percent increase in average frame rates along with 24 percent higher performance in the lows. Turning to AMD, BOT enabled the 270K to overtake the 9900X and largely match the 7800X3D in performance.

The 250K, meanwhile, manages to achieve performance roughly equivalent to the more expensive 265K, with BOT off. Turning it on, the chip takes the upper hand, not only over the 265K, but the 9900X as well.

Intel's hardware and software co-design shines in this title, we just wish BOT performed this well in the other titles we tested.

This is by no means a comprehensive assessment of the Core Ultra 200S Plus-series parts in gaming, but should at least give you some sense of how these chips impact performance in scenarios where the GPU isn't the performance bottleneck.

We do think it's important to understand that, at higher resolutions and with more eye candy turned on, CPU performance is much less likely to be a bottleneck. This is especially relevant if you aren't rocking a 4090-class GPU or better, and you don't plan to upgrade your GPU in the future.

Summing up

The PC market is in the midst of an AI-fueled price crunch. Memory and storage can easily cost more than a CPU, and enthusiasts are having to make tough decisions on where to allocate their budgets.

Intel's aggressive pricing and meaningful gen-on-gen gains in both our production and gaming benchmarks make its Arrow Lake refresh parts a much better value than they were when its first round of Core Ultra 200S processors launched back in 2024.

Intel's $199 Core Ultra 5 250K strikes us as a particularly good deal that should allow budget-conscious PC enthusiasts to stretch their dollar a little bit further. At $299, the 270K is a tougher sell, at least for gamers. While it delivers higher performance than the Ultra 5, it's not 50 percent better.

If you were hoping for an Intel answer to AMD's cache-stacked X3D processors, this just isn't it. Worse, if the most intensive thing you plan to do with your PC is gaming, AMD's 7800X3D is much faster, and can now be found for as little as $349. That's $50 more than Intel, but the higher cost can be offset by using a less expensive 600-series motherboard.

But if your workloads also include anything else remotely multi-threaded, the 270K starts to make a lot more sense, punching well above what its price tag would otherwise suggest.

Unfortunately, if you do opt for Intel's Arrow Lake refresh, know that socket 1851 is a dead-end platform that's soon to be replaced by Intel's Nova Lake family of processors. Timing on those parts is still a bit fuzzy, but we expect to see them either late this year or at CES in January.

If you're the kind of person who upgrades their system every three to five years, this may not be a deal-breaker, but for those that upgrade more frequently or those factoring in the ability to reuse their motherboard across more than two generations, the 250K and 270K may not be for you.

It's also worth noting that AMD is also due for a desktop refresh, though recent reports suggest Zen 6 may not be coming to consumer platforms until next year. That still leaves plenty of opportunity for AMD to roll out their own budget-oriented parts. Maybe a 9700X3D or 9600X3D Micro Center special? Or perhaps we'll finally see that long-rumored 9950X3D2 dual 3D V-Cache dies.

In any case, if Intel's Arrow Lake refresh chips do strike your fancy, you can find them on store shelves starting Thursday. ®