Monday, February 3rd 2025

Intel Core Ultra 255H "Arrow Lake-H" Delivers 32% Single-Core Performance Improvement Over "Meteor Lake" Predecessor
Intel's Core Ultra 7 255H "Arrow Lake" processor has demonstrated impressive performance improvements in recent PassMark benchmarks, achieving a 32% higher single-core score compared to its "Meteor Lake" predecessor. The Arrow Lake-H chip recorded 4,631 points in single-threaded tests, significantly outpacing the Core Ultra 7 155H's 3,500 points while delivering a 15% overall improvement in CPU Mark ratings. The performance leap comes from Intel's architectural overhaul, implementing "Lion Cove" performance cores alongside "Skymont" efficiency cores on TSMC's N3B process node. This combination enables the 255H to achieve higher boost frequencies while maintaining the same core configuration as its predecessor—six P-cores, eight E-cores, and two Low Power Efficiency (LPE) cores.
Notable in this iteration is the absence of Hyper-Threading, resulting in 16 threads compared to the 155H's 22 threads. Arrow Lake-H maintains Intel's heterogeneous structure, incorporating up to eight Xe-LPG+ graphics cores derived from the Alchemist architecture. The neural processing unit (NPU) capabilities remain consistent with Meteor Lake, delivering 13 TOPS of INT8 performance. This positions the chip below Lunar Lake's 45 TOPS. Despite performance improvements, market success will largely depend on system integrators' ability to deliver compelling devices at competitive price points, particularly as AMD's Strix Point platforms maintain strong positioning in the $1,000 range. The battle of laptop chip supremacy is poised to be a good one in the coming quarters, especially as more Arm-based entries will force both Intel and AMD to compete harder.
Sources:
x86deadandback on X, via Tom's Hardware
Notable in this iteration is the absence of Hyper-Threading, resulting in 16 threads compared to the 155H's 22 threads. Arrow Lake-H maintains Intel's heterogeneous structure, incorporating up to eight Xe-LPG+ graphics cores derived from the Alchemist architecture. The neural processing unit (NPU) capabilities remain consistent with Meteor Lake, delivering 13 TOPS of INT8 performance. This positions the chip below Lunar Lake's 45 TOPS. Despite performance improvements, market success will largely depend on system integrators' ability to deliver compelling devices at competitive price points, particularly as AMD's Strix Point platforms maintain strong positioning in the $1,000 range. The battle of laptop chip supremacy is poised to be a good one in the coming quarters, especially as more Arm-based entries will force both Intel and AMD to compete harder.
9 Comments on Intel Core Ultra 255H "Arrow Lake-H" Delivers 32% Single-Core Performance Improvement Over "Meteor Lake" Predecessor
If you really meant the 140V I'd be really surprised as that means we get a 890M competitor, got a source for that? I'd like to know more.
www.notebookcheck.net/Intel-Arc-140T-graphics-delivers-60-percent-performance-boost-over-Lunar-Lake-variant.905446.0.html
www.techspot.com/news/105244-intel-mysterious-arc-140t-gpu-teases-massive-performance.html
www.tomshardware.com/pc-components/gpus/mysterious-intel-arc-140t-graphics-engine-significantly-outpaces-xe2-lunar-lake-igpu
Also, the score looks impressive this time around because we all remember how unimpressive Meteor Lake generation was.
Real life is the best benchmark. I'm sure those 32% will not even reach 13%. If not now, than later with the fixes for the errors in design and silicon for DRAM and CPU issues.
I configured a few hours ago another kernel. A hole page only for cpu migitations. There are other options hidden in submenus for DRAM issues.
Slower GPU! (vs. 258V)
(Less than) one third the NPU! (vs. 258V)
Three times the microarchitecture!
Three times disabled ISA extensions! ("Now available in E-cores!TM")
Three times CPU mitigations!
Three times Windows scheduler confusion!
Three times optimized path dispatch trouble!
Three times the fun!
This Intel insistence on not putting AVX-512 instructions in its home user CPUs creates a huge delay in the development of software optimized for these instructions and gives a huge advantage to other companies that manufacture CPUs with ARM and RISC-V architectures, which have equivalent instructions, to gain market share.
------------------------------
Intel needs to realize that people don't want these nonsense E-cores.
Much of the software is still poorly optimized for many cores, and even those that are optimized for multicore still overload 1 or 2 CPU cores. Therefore, Intel should abandon these nonsense/useless E-cores and put only P-cores along with 1 "Super-core" for every 3 P-cores.
Dozens of cores are only useful for server applications, not for home PCs.
This way:
Riddle me this: In today’s competitive CPU landscape, why aren’t AMD and the various Arm vendors bringing their supercores to market. They surely do have these is their evil lairs somewhere. Right, they do have those? ’Cause, AMD is surely much, much superior at CPU design than Intel is, so if there’s any chance at supercores being brought to market, it’s by them.
What we’ve got—are their supercores, in a way. They’ve got their full cores and they’ve got their compact cores. And Intel has got supercores, their continued big core product line, and they‘ve got their normal cores, brought up and much improved from their Atom predecessors.
So that’s where we’re standing today! You can even buy chips with many more supercores than you’ve proposed! It’s just that, it’s unlikely any vendor could do better cores than this “just like that” and offer anything better than what current full, or big, cores already do, especially when power (and heat dissipation) constraints, as well as placement difficulties (caches getting in the way of other units), will (I mean, ostensibly) render any brute-force-approaches invalid.