Go to page:

Arrow Lake was a dud. Intel transitioned from holding a compelling position against AMD among the best CPUs for gaming to being in a distant second place. While dealing with the instability controversy surrounding Raptor Lake Refresh, Intel released underperforming chips that, although architecturally interesting, were undermined not only by the competition from AMD, but also by Intel’s other 13th- and 14th-Gen offerings. Arrow Lake Refresh, officially dubbed Core Ultra 200S Plus, aims to change that narrative ahead of Intel’s true next-generation architecture, Nova Lake, which is on track for a release later this year.

For now, we have two new chips — the Core Ultra 7 270K Plus and Core Ultra 5 250K Plus. We’re looking at the Core Ultra 7 270K Plus today, which arrives for $100 less than the Core Ultra 7 265K, while packing four more E-cores and a 900 MHz bump in die-to-die clock speed out of the box. All of the knobs and dials for overclocking that Intel introduced with Arrow Lake are still present, but the bump in die-to-die frequency is now stock; you don’t need a Z-series motherboard to unlock it, which Intel says was a conscious choice given the pricing conditions Core Ultra 200S Plus is arriving during.

Although the Core Ultra 7 270K is a refresh by definition, it performs more like a reset. It’s arriving too late, and in a market that’s increasingly hostile to PC enthusiasts, but it feels like the Core Ultra 7 we should’ve seen from the start. The efficiency angle of Arrow Lake is out the window in favor of squeezing out higher performance, and Intel’s promising Binary Optimization Tool finds further performance gains in lieu of strictly more silicon. Further pushing the reset angle is the price. Intel has clearly recognized its growing position as the underdog in the desktop PC market, and it priced the Core Ultra 7 270K aggressively to make up some ground that’s slowly slipped away.

In applications, the Core Ultra 7 270K is hard to believe– so difficult, in fact, that I had to rerun applications on the full Arrow Lake stack to make sure my numbers were correct. In games, it’s decent. Intel is able to squeeze out a marginal lead over AMD’s competing Ryzen 7 9700X, but AMD’s X3D offerings still hold a solid lead, though at a much higher price.

We’ve also tested the Core Ultra 5 250K Plus, and you’ll see that testing reflected in our geomeans below. Our full Core Ultra 5 250K Plus review is live if you're interested in learning more about that chip.

Intel Core Ultra 200S Plus ‘Arrow Lake Refresh’ Pricing and Specifications

Although the Core Ultra 7 270K sports the Arrow Lake microarchitecture, it isn’t just an overachieving Core Ultra 9 285K. The specs are similar, but Intel’s Robert Hallock tells me that “it is not a binned Arrow Lake CPU. This is a new wafer, a new product code, etc.” Still, the Core Ultra 7 270K has the same core configuration as the Core Ultra 9 285K, with eight Lion Cove P-cores and 16 Skymont E-cores. The cache is the same, as well, with 40MB of L2 and 36MB of L3, as is the thermal design, with a TDP of 125W and MTP of 250W.

The main difference is clock speed, both of the cores themselves and in the interconnect between the various chiplets (Intel calls them “tiles”) that make up the Arrow Lake architecture. For core frequencies, the Core Ultra 7 270K Plus tops out at 5.5GHz, same as the 265K, while the Core Ultra 9 285K can climb to 5.7 GHz out of the box. However, the Core Ultra 7 270K has a 900MHz bump in die-to-die frequency, particularly speeding up communication between the Compute tile and SoC tile, where the memory controller lives. Intel also bumped the fabric speed by 400 MHz.

With Intel’s Core 200S Boost on Z-series motherboards, both the fabric and die-to-die frequency can climb to 3.2 GHz, regardless of whether you’re using a stock Arrow Lake or Plus chip. So, stock Arrow Lake chips can recover some of the performance on display here. Critically for the Plus parts is that you’re getting within 200 MHz of the boost profile out of the box. You don’t need a specific motherboard to leverage the improved speeds.

On the memory front, Intel has officially bumped the spec with Plus processors to 7200 MT/s, up from 6400 MT/s; though, even stock Arrow Lake chips have no issues maintaining 7200 MT/s with high-quality DIMMs. Intel has also teased early support for 4R (four-rank) CUDIMMs on select motherboards. It’s still early days for support, and we need a motherboard to play along, but that’s something that’s arriving with this Plus refresh.

The pricing is the story here, though. At $300, Intel has bumped the Core Ultra 7 270K Plus down a tier in pricing while bumping it up a tier in specs. “We could have produced something, you know, with the 8+16 config that is more costly, different branded… but we didn’t want to,” is what Hallock told me. Those would normally be hollow words, but given the performance here, particularly in applications, there really does seem to be a mindset shift within Intel. If that continues is a different question, but for this product, you’re getting more for less money, pure and simple.

Breaking down the Binary Optimization Tool

Tweaks in silicon are half of the performance equation here. The other half is Intel’s Binary Optimization Tool, or iBOT. There’s a stance that Intel is making up with software what it can’t achieve in hardware, but I don’t think that’s the right read of iBOT. It is, fundamentally, a lever that Intel can pull to increase IPC for a given workload. It’s something we’ve never seen before, and although iBOT on its own isn’t delivering some generational leap in performance, it shows a lot of promise.

Intel describes iBOT as translating “other x86” to “Intel x86.” You can think of it as a translation layer along the lines of something like Microsoft Prism, but we’re not moving from one ISA to another. Instead, Intel is optimizing instructions to better leverage a particular architecture. It’s able to do this using Hardware Profile Guided Optimization, or HWPGO. Within Arrow Lake Refresh chips — and Intel chips moving forward — there are registers to show what is happening when code is executing on the chip. That includes things like cache misses, branch mispredictions, and hardware interrupts.

When a developer is compiling their binary, there’s a toolchain of optimization that takes place where they look at these types of inefficiencies. Then, they can go back to the source code, make adjustments as necessary, and recompile. With iBOT, Intel is trying to eliminate those inefficiencies, but it’s doing so on a production binary. It doesn’t need to touch any source code. That’s because these “hooks,” as Intel calls them, work on shipping binaries. It’s able to see inefficiencies and make adjustments, but it does so at runtime on a production binary, not through source code.

Let’s use a cache miss as an example. Intel can see a cache miss happen, and it can investigate what went wrong. For instance, maybe a piece of data wasn’t tagged properly and was flushed from the cache. You’d have to go get that data again, and your performance would go down. iBOT allows Intel to tag that data properly so it doesn’t get ejected from the cache. Add up these small efficiency improvements, and you could squeeze out some extra performance. And, as Intel describes it, this would effectively increase IPC. Cache misses and branch mispredictions represent instructions that weren’t fully executed within a cycle, so fixing those issues makes IPC go up.

This post-ship optimization presents a lot of opportunities. I’ll tell you now that, in the handful of games iBOT is releasing with, you’re looking at somewhere in the high single-digits for an uplift. It’s not massive, but it’s an early demonstration that this concept has legs. Developers use different compilers and different toolchains, and those have evolved and will continue to evolve over time. iBOT allows Intel to take out at least some of the inefficiencies in those toolchains. It could apply to an older application running on a new architecture just as easily as it could to a newer application running on an older architecture.

iBOT is an opt-in feature; Intel tells me that it’s being cautious about rolling out the feature, trying to avoid claims that it’s playing dirty tricks to win favor in benchmarks. That isn’t the case, short of Geekbench, where Intel has a proof of concept for how iBOT can work outside of games. I’ll address that when we reach Geekbench in our productivity benchmarks.

Intel is modifying code running in real-time, and it’s been clear that multiplayer games aren’t initially supported in iBOT because of that fact. If there are broader security implications remains to be seen, but it’s something to keep in mind.

If there are security risks, they shouldn't reach deep. Intel says iBOT operates at the same level as user-mode applications. It doesn't have direct access to the hardware, and it's making system calls like any application would.

• MORE: Best CPU for gaming
• MORE: CPU Benchmark Hierarchy
• MORE: Intel vs AMD
• MORE: How to Overclock a CPU

Jake Roach is the Senior CPU Analyst at Tom’s Hardware, writing reviews, news, and features about the latest consumer and workstation processors.