Monday, July 7th 2025
Intel Plans "Arrow Lake Refresh" for H2 2025 with Higher Clocks and Upgraded NPU
According to ZDNet Korea's sources familiar with the matter, Intel will release the "Arrow Lake Refresh" CPU family in the second half of this year. While no exact timeline is given, we can expect to hear more about these CPUs in the coming weeks, and for the rumor mill to intensify. The new refresh generation will introduce slight frequency adjustments, including an increased base clock and maximum boost clock, while also incorporating a faster NPU. A base clock increase, paired with a boost to maximum core clock, will give Arrow Lake Refresh a significant advantage in gaming performance. Keeping the same LGA1851 socket will extend the support for the current platform for one more generation too, until "Nova Lake-S" arrives.
Given that Intel attempted an IPO for a limited set of gamers, we could see a more widespread adoption of a similar approach. As a reminder, on the CPU side, IPO fine-tunes P-core and E-core frequencies, ring-bus speeds, the UPI interconnect, D2D links between tiles, and both PL1 and PL2 power limits. For RAM, IPO raises transfer rates and tightens timings, pushing modules beyond their factory XMP or EXPO profiles. If Arrow Lake Refresh ships with an IPO-equivalent out of the box, gamers will enjoy a significant performance increase, ready to offset some of the early disappointment with the base Arrow Lake. Additionally, Intel will place a new NPU on its IO tile. Using the NPU3 on base Arrow Lake, Intel was unable to meet the basic 40 TOPS requirement for Copilot+ AI PCs. And with NPU4, that TOPS requirement is met, allowing Intel to brand these new processors as AI PC ready.ZDNet Korea also noted that Intel is preparing Nova Lake-S for the second half of next year, meaning that we are still months away from hearing more details about its platform. For now, we can only note that the Nova Lake-S, rumored to feature up to 52 cores, will be compatible with an LGA1954 socket and will include a mix of P, E, and LPE cores.
Source:
ZDNet Korea
Given that Intel attempted an IPO for a limited set of gamers, we could see a more widespread adoption of a similar approach. As a reminder, on the CPU side, IPO fine-tunes P-core and E-core frequencies, ring-bus speeds, the UPI interconnect, D2D links between tiles, and both PL1 and PL2 power limits. For RAM, IPO raises transfer rates and tightens timings, pushing modules beyond their factory XMP or EXPO profiles. If Arrow Lake Refresh ships with an IPO-equivalent out of the box, gamers will enjoy a significant performance increase, ready to offset some of the early disappointment with the base Arrow Lake. Additionally, Intel will place a new NPU on its IO tile. Using the NPU3 on base Arrow Lake, Intel was unable to meet the basic 40 TOPS requirement for Copilot+ AI PCs. And with NPU4, that TOPS requirement is met, allowing Intel to brand these new processors as AI PC ready.ZDNet Korea also noted that Intel is preparing Nova Lake-S for the second half of next year, meaning that we are still months away from hearing more details about its platform. For now, we can only note that the Nova Lake-S, rumored to feature up to 52 cores, will be compatible with an LGA1954 socket and will include a mix of P, E, and LPE cores.

78 Comments on Intel Plans "Arrow Lake Refresh" for H2 2025 with Higher Clocks and Upgraded NPU
When old E-cores are getting closer and closer to P-cores (in transistor count/complexity), and you have to make make "new" E-Core to pick up "magic energy efficiency" metric - you f***ed up, and should rethink your plans from this point onwards.
PS. I'm pretty sure most office workers would be just fine with quad/octa E-cores even today vs. one or two P-Cores + "LPE-core".
Adding lots of L3 doesn't make the CPU faster at logic, which is what makes the CPU responsive in games and applications, instead it mostly helps bloated code, especially when you exaggerate the conditions by running artificially low GPU loads etc. No one will run high-end GPUs at 720p/1080p/low details in reality, so we shouldn't care about edge-cases.
I would much rather have ~5% higher IPC than a whole chiplet of L3 cache, especially if it meant meaningful front-end improvements which for AMD are sorely needed.While I don't posses their binning data, it seems like Intel "screwed up" with the clocks of the Ultra 5 245K, which is the counterpart of Ryzen 5 9600X, and if it was tuned properly would be an excellent choice for gaming. I don't know whether this was a marketing decision of Intel or not, but the end result is Intel losing out on a lot of sales instead of shifting volume up to 285K.I don't play team sports, and strive to have a representative selection of both to validate software, and have had/used most except for the very latest. It's certainly not as simple as one being better at everything, and Zen 2 and 3 are certainly good where they shine. Zen to this day scale very well with batch workloads, and now extremely well with heavy AVX-512 loads. Intel consistently have had a lead in snappiness/responsiveness, which is why many applications "feels" better, even though most productive benchmarks only benchmark a big batch load, and is therefore often not representative of the time spent in the application. This also applies to gaming, and even benchmarks with .1% low measurements doesn't tell the whole picture about smooth rendering. There is also variance in input latency, which can be just as annoying, if not worse.
For now I'm fairly "annoyed" by Intel for constantly changing their hybrid designs, so in worst case I may have to get sample(s) from Alder Lake/Arrow Lake/Nova Lake to validate proper scaling of my software... :(
Cheers
LPE-Core (third type of "x86 core" that's getting added), is the "new thing" in Nova Lake because "usual E-cores" are just too high power (apparently).
It's NOT going to work how Intel wants it to work, because Microsoft's OS is too dumb to actually make it work right (Linux can, but that's besides the point).
www.techpowerup.com/review/amd-ryzen-7-5800x3d/16.html
It was ~$150 cheaper than the "equivalent" 12900k/ks and not much more than the 12700k which both required DDR5 to really shine which was INSANELY expensive at the time in comparison to a performant DDR4 kit. Near release it was 2-3 times the cost for RAM alone plus then the motherboard replacement being another decent chunk of change especially ones focusing PCI-e 5 connectivity.
I can still pick up a 5700X3D for ~£200 which is the same processor with just a slightly lower clock bin. I cannot build any platform for that kind of money no matter what you say.
www.techpowerup.com/review/intel-core-ultra-7-265k/19.html
Even now a 5700x3d is within 3-10% of equivalently priced options without the RAM/Mobo upgrade being needed at the moment and can either put those funds to GPU (which we all know are extortionate at the moment) or wait for both Intel and AMDs next gen offerings which should offer a decent performance upgrade vs current offerings.
So if you are on AM4 at the moment and have the ability/funds to upgrade to a 57/5800x3D but they dont stretch to a full rebuild you are mad to think of any other option if they are fairly focused on gaming. You do not however build a new AM4 rig in the TYOL2025 as there are better options out there both from Intel and AMD in everything else.
AMD has more longevity as you should still have at least one more generational upgade on current AM5 and their current options are one of if not the best option for gaming and they arent a slouch in productivity either.
Intel has SOME interesting options from things like 14600/14900 depending on what your focus is and the Core Ultra stuff if you are purely productivity focused but you have the knowledge that all platforms are EOL with no real upgrade options puts me off advising investing massively in either of their options.
uk.pcpartpicker.com/product/3ZKscf/amd-ryzen-7-5700x3d-3-ghz-8-core-processor-100-100001503wof
OK dude, thank you for confirming you need to pay 200% for an "equivalent" upgrade. Please read a complete post before spitting out the first pro intel thing you think of.
It very often puts FPS in the places you want it least (Average FPS + Highs) but does much less when the system needs to go back to ram and load things in (i.e. stutters) - so you can end up with wild swings in performance especially in badly coded UE5 games and console ports.
Nothing quite like looking at an FPS graph that says 180 but the game feels like Gsync is off at 80 FPS at 4K when fully GPU bound.
There are many types of stutter, and just as many causes. But one thing to be aware of is that some stutter may not be easy to measure in software, especially when it comes to "microstutter". I became aware of this many years ago while writing a rendering engine, when I noticed running a latency optimized Linux kernel resulted in a much smoother experience. Even running side by side next to a more powerful GPU, the difference was quite noticeable. (I hope to revisit this sometime on modern OS', and see how it behaves today, but I don't have high hopes considering how Win 11 "feels".)
I would argue that very frequent tiny stutter can be worse than an occasional slow frame, even if the slow frame is much more significant (this may be subjective). And some stutter may affect the kernel, and in such cases it may also affect the clock, and in such cases you simply can't trust the measured results. This is why the measured results may look "perfect", but you certainly can "feel" the difference. So in such cases you should trust your observation more than the "measurement".
I wish that we could have proper deep dive "reviews" of gaming setups done by skilled engineers using debug tools and possibly even customized tools to single out the various sources of latency in the system. But nevertheless, I know what the findings would look like; especially in software there is more crap than you can ever fathom.
300 EUR for a core 7 ultra and 130 EUR for a B860 mobo is not a bad price..
But didn't find anything under 300 in europe
It dipped below $250 for a new one not long ago -- seems like eu still very expensive for AL.
900 settings later i discovered windows has something called "MPO" as part of the DWM and if you disable it on the 9800x3d it stops the random microstutter and Gsync screen flashing...
When I tested with the Intel system - didn't even reinstall windows, I didn't install chipset drivers, I did nothing -- just plugged the AMD AM5 driver laden ssd, which stuttered on the AM5 boards, in the Z890 nvme slot and booted it. Was totally fine, no issues...