Honestly, I feel like Intel is getting too much flak for Arrow Lake. While they obviously need to improve things like thread assignment with the new architecture, I feel like most of the blame needs to be assigned to Microsoft.
You're right about the criticism of being either too unfair or simply misinformed, probably largely due to a mix of unrealistic expectations and as usual users being fixated on synthetic or unrealistic benchmarks. There is no doubt that Arrow Lake is an advancement overall, despite there being noticeable regressions, which is
to be expected when there are large architectural overhauls, and the only way to avoid that would be to basically refine the same architecture forever.
As for scheduling etc., that's up to the OS to handle, not Intel. And while it certainly plays a role in some workloads, AMD has to deal with the very same issue. Interestingly enough, back in August when Zen 5 launched, MS was accused of screwing over AMD, but that's apparently long forgotten now. As is evident with Linux benchmarks, there is clearly some potential for all modern hardware running on a non-antiquated kernel, but don't expect MS to change that any time soon. (And please no more patchwork with workarounds, stability would actually be better.)
I work in IT, and we've had to roll back 24H2 on multiple computers because it breaks multiple unrelated pieces of software that are absolutely business-critical for us, so it's not just an issue with a single feature or part of Windows.
That's a quality issue with MS, and has been a growing problem for years.
…and they've moving to drop native support for x86, which I think will pair very nicely with moving to single-threaded cores and allow them to simplify core design immensely.
They are
not moving to drop x86, they are actually working on refining it.
I'm not sure what you mean by "native support", as all modern microarchitectures since the mid 90s have translated the ISA to micro-operations, which have effectively eliminated the arguments against x86. Moving to any of the current ISA alternatives would result in more instructions, less cache efficiency and more branching, to that's out of the question for high-performance generic computing.
What's much more significant, is the fact that most software doesn't use the x86 ISA efficiently. While many productive applications are compiled to use higher ISA levels, the OS kernel, drivers, libraries/runtimes are usually not, and still usually stick to x86-64/SSE2 (21 years old). That would be almost like running Windows XP on 8086/8088 ISA, let that sink in for a moment. If the entire software stack was recompiled for e.g. Haswell ISA level, it would probably unlock ~10% performance or more, and a lot more to be unlocked with manual optimizations of important libraries. (Linux distributions are moving in this direction, but the progress is slow.) Intel really screwed up with disabling AVX-512 on Alder Lake, otherwise we could have had a lot more performance by now.
It seems that ARL is broken at siilicon level, doubt they can fix it in new microcode in any significant way. Maybe this is caused pairing MTL SOC wuth ARL compute tile, and they do not talk in efficient way ?
Intel now must make new revision with fixed silicon, or say godbay to some 20% of their CPU share, as gamers and many causal gamers won't buy it.
It's not broken, unless there are hardware bugs that I'm not aware of.
As I've said since the beginning; the performance characteristics of Arrow Lake is what it is, no amount of firmware, BIOS, drivers or kernel workarounds are going to change that (not unless they want to pump dangerous voltage and clocks etc.). And as it's a very different architecture there will be some regressions, the bigger question is whether those regressions along with the gains matter to each specific customer. Keep in mind that 99% of users considering an upgrade will not be upgrading from Alder Lake/Raptor Lake.
There's no question that Arrow Lake is the more powerful overall, and too many dismiss a product because of use cases that aren't even relevant. It's not like anyone will game on a high-end GPU in 720p or 1080p. Realistically you'll be running 1440p/4K with something like a RTX 4060/4070, so the differences will be much less than people seem to think, and even with RTX 4090 it's ~2.9% in 1440p and ~1.5% in 4K, so it's not like you're going to suffer a bad experience. (And as Arrow Lake is computationally faster, more demanding games will skew in its favor in the future.) It ultimately comes down to which applications the user will run, which is also true when comparing to Zen 5, it's no point in buying the best CPU for workloads you'll never see.
As we can see from
Linux benchmarks, 285K is about ~12% faster overall vs. 14900K, which serves as a good "indicator" of what could be achieved with better software.
ARL names 285, 265 and 245 suggest they knew it is broken and won't be true successor for Raptors X900, X700 and X600. Maybe they do quick revision with fixed perf and names 290, 270 and 260.
It seems more like they've added some headroom to squeeze in an extra model or too down the line. While we wouldn't know until it happens, i.e. a 295K, or a refreshed lineup are options.
