Thursday, November 13th 2025
Intel Officially Confirms AVX10.2 and APX Support in "Nova Lake"
Intel published its 60th Architecture Instruction Set Extensions and Future Features manual, confirming that the AVX-512 subset and its AVX10.2 superset will be supported on the next-generation "Nova Lake" processors. This puts an end to months of rumors circulating online that Nova Lake will not support AVX10.2 and APX, leaving only AMD with 512-bit vector acceleration. However, we will now have two x86 CPU vendors with a proper 512-bit processing in the desktop PC market. This is important as the regular CPU becomes a universal base platform, fulfilling its intended role with enhanced observability. Software can now be accelerated for any market from desktop to server.
Since AVX10.2 mandates both P-cores and E-cores to support AVX-512 acceleration, this is a sign that "Coyote Cove" P-cores and "Arctic Wolf" E-cores for Nova Lake will be running advanced vector processing, greatly accelerating what the CPU can do. Especially since Nova Lake is supposed to be a 52-core design, having a massive set of cores processing AVX-512 will finally give Intel a possible advantage in the desktop CPU feature set. In comparison, AMD introduced full AVX-512 support with its "Zen 5" cores across its product lineup, enhancing performance in optimized applications for both desktop and server CPUs. This marked the first time AMD implemented native 512-bit AVX support, eliminating the need to split 512-bit data into two 256-bit units for processing in two cycles. AMD will continue with AVX-512 and other features in "Zen 6," making this a start of year-long competition of advanced features in the desktop sector.
Source:
@InstLatX64 on X
Since AVX10.2 mandates both P-cores and E-cores to support AVX-512 acceleration, this is a sign that "Coyote Cove" P-cores and "Arctic Wolf" E-cores for Nova Lake will be running advanced vector processing, greatly accelerating what the CPU can do. Especially since Nova Lake is supposed to be a 52-core design, having a massive set of cores processing AVX-512 will finally give Intel a possible advantage in the desktop CPU feature set. In comparison, AMD introduced full AVX-512 support with its "Zen 5" cores across its product lineup, enhancing performance in optimized applications for both desktop and server CPUs. This marked the first time AMD implemented native 512-bit AVX support, eliminating the need to split 512-bit data into two 256-bit units for processing in two cycles. AMD will continue with AVX-512 and other features in "Zen 6," making this a start of year-long competition of advanced features in the desktop sector.


91 Comments on Intel Officially Confirms AVX10.2 and APX Support in "Nova Lake"
With Intel, it is also usually at least two gens (not sure why people repeatedly claim that Intel requires a new board for every new CPU gen) on the same board, sometimes three (refresh).
I currently run AMD but never really got the appeal of sticking a brand new CPU into ancient motherboards that only support half the feature set (if that) of the new CPU and where a lot of performance simply goes poof when you mix ancient tech with new tech.
But I guess a lot of uninformed AMD fans are easy to please in this regard. They're just happy that they can update their forums signatures and pretend they're on the latest gen when in reality their system is held back massively by the dated platform. Oh well... to each their own.
Me, I will probably upgrade from X870E to X970E for Zen 6, even if I wouldn't have to, because I'm a real™ enthusiast. It depends a little on what new features X970E is going to bring but if it has an improved IO controller and a few other niceties then I ain't gonna plug no Zen 6 into my "ancient" X870E board. No real™ enthusiast would do that.
As for the redundant sustainability argument that has been brought up: No enthusiast throws their hardware in the trash. We sell our stuff on eBay which brings down the cost of the new hardware and makes a used buyer a happy camper. This is a complete non-argument.
Intel usually changes sockets after 1 gen and a refresh, a refresh isn't a new CPU gen. Also thanks to the "real enthusiasts" buying new boards every CPU gen, prices are out of control on what used to be high end, and manufacturers listened when people buy boards covered in M.2 slots and dumb plastic shrouds covering everything, while real features like a POST code display and decent onboard audio are only on the ridiculously expensive boards.
But hey if the "real enthusiasts" want to flush money down the toilet for 2fps more in a game, then to each their own.Extra e-cores slapped onto the same architecture, not the first time Intel stagnated though, they stagnated through almost the whole 14nm+++++++++++++++ era.
But 14700K is not an arch uplift therefore the difference is rather small compared to 13700K (in 1080p gaming it's academic).
www.techpowerup.com/review/amd-ryzen-7-9800x3d/27.htmlThe only exception is the 14700K which brings more E-cores and L3 cache. But 13700K to 14700K is not really an upgrade.
www.techpowerup.com/review/amd-ryzen-7-9800x3d/25.html
Temps are better despite the marginally higher power draw but still it would be a real stretch to call this an upgrade.
Sidegrade yes, and depending on the resale price for the 13700K and the difference to get the 14700K it could have been worth it a while ago for some. Right now probably not, generally speaking.I'm readier. :rockout: Hopefully it's gonna slay!For some time now I've been trying to understand the whole "upgrade path" perspective/concept and imagine valid scenarios for it. I won't post my thoughts and examples here, maybe I'll make a separate thread, because as you said:Raptor Lake is a different arch than Alder Lake.
Looking superficially it might seem that the 13700K is just a clock bump of the 12900K while keeping the same core config and L3 cache.
But the L2 Cache is different and also other changes. Second and third paragraph below:
www.techpowerup.com/review/intel-core-i7-13700k/
That is why the difference between the 13700K and 12900K is larger (8.37% in apps and 6.49% in 1080p gaming) than that between the 14700K and 13700K (6.37% in apps and 0.78% in 1080p gaming).
www.techpowerup.com/review/amd-ryzen-7-9800x3d/27.html
So 12th gen to 13th gen is as legit as Zen4 to Zen5, but 13th gen to 14th gen is just a refresh.
By comparison, Zen 4 to Zen 5 is a more significant change, let's call it a "medium" sized architectural change. On paper, the changes might lead us to expect ~15-20% gains overall, if not more, yet the gains are fairly minimal (except for AVX-512 loads of course). I would love to hear the perspective from the designers whether how much of this are shortcomings of the design, and how much of this may be "laying the ground work" for the next iterations.
One interesting parallel in this regard is Arrow Lake, where the popular opinion regards it as a massive failure, yet the architectural design changes were needed not only to improve efficiency, but also for further expansion (This is one of the reasons why Nova Lake and successors with APX will get exciting). There is no question that Arrow Lake is a more capable architecture, and it only "falls short" of Raptor Lake in cases where the latter allows to boost into absurd clock speeds.
Intel seems not to be able to support or make any 512 bit calculations. Zen4 needs two cpu cycles.
So Zen5 with the niche 512 bit AVX 512 vectors will be a niche product again as intel made the 512 wide vectors optional. I see optional as we do not want to say no but we do not want to say yes to implement it. The advantage of zen5 is like the bulldozer xop instruction non existant in near future in my point of view.
Making a spec with optional is really nonsense, what intel does here with avx512 = optional. I see the point what intel marketing does, drop full 512 bit support and rename the instructions so no one sees the issue. Hiding the flaws with a worse substandard with similar or worse instructions, not to be able to implement AVX512 properly in the first place. I only judge on the marketing material and what i saw over the years from those gnu gcc release notes.
On the positive side, P and E cores seems to be able to do the same maths. I see it as flaw as intel did not said anything about the third types of Cores, what they are able to do or not. I think those were called low power E-Cores or something similar. They should have written clearly, all processor cores can execute any instruction which the cpu itself support, regardless which "power configuration" they are. I'll see it anyway with the gnu gcc release notes and when those processors hit the market what instruction get another time turned off by intel after consumers bought the processors for months / years.
edit: Most cunsumers won't bother with it. As those cpu instructions, sometimes called flags, are such niche topic. Just because current windows 11 pro does not use it, does not mean it has it's use. Why? Windows 11 run on 3rd gen intel core processors. (Whataboutism).
www.phoronix.com/news/Intel-AVX10-Drops-256-Bit
AM5 socket has wires for 28 PCIe lines. B650E and X670E provides 16x lines to GPU and 2 sets of 4x lines for NVME at v.5 Speed.
Zen 6 wont come with PCIe v.6 nor DDR6. So what limitation does those chipset have?
There is way to increase amount of PCIe lines. The Elephant in the room has its name. Promontory 21 made by Asmedia.
Most PCs comes to market with 1 NVME stick, 2 RAM modules (or even 1) and thats all. I rather wont mention those devices with sold-on eMMC and RAM modules which became so popular in corporates.
Everything over that is luxury. You want it, you have to pay for it.
PS: I have bough B550 board in 2020 and did not feel limited by it. I may work fine for another 4 years without severely limiting you have mentioned. 1x NVME stick, 1x HDD on SATA, v.4 16x GPU.
Nova Lake is probably going to have PCIe 6.0. AMD can't provide that because they're stuck on AM5 still.
But its good to bash something into oblivion for not providing it.
I have "good old" MLC in NVME.
Those "high speed" are TLC, QLC or even worse crap with limited amounts of (over)writes.Bifurcated or not, shared or not, fewer or more, .... that all depends on manufacturer.
Its simply:
You do not like, you chose another.
You want more, you have to pay more.
What can be connected to AM5 board:
tpucdn.com/cpu-specs/images/connectivity/amd-am5.pngWe will see.
I will not hold my breath on this.
Edit: No PCIe 6.0 listed
www.techpowerup.com/338075/intel-nova-lake-s-platform-and-8-000-mt-s-memory-configuration-leaks
We don't have a lot confirmed for Zen 6 yet, but Zen 7 will add AMX (which for now is relevant for HPC and "AI").
Some have speculated about higher clocks for Zen 6, but that must be a wild guess at this stage. But if it happens that Zen 6 doesn't add APX yet, what else could they do besides general tweaking and refinement? It would be interesting if they added another FMA-set, in a typical AMD design fashion of throwing resources in there, and optimizing them later. But at the very least I'm hoping for meaningful front-end improvements and better use of those extra ALUs added in Zen 5.Remember, that's always a shifting target, especially since many now essential tools are becoming performance bottlenecks due to bloat, namely the browser, basic office tools and the Windows OS itself. While many productive tools and utilities are pushing for some optimizations, others are just getting worse (*cough*Adobe*cough*). A lot of the performance improvements we are enjoying on the Linux side are primarily driven by the needs for servers.
And as always, there is a battle of bloat sneaking in everywhere (even in Linux), mostly due to misguided philosophies in regards to software development. And this is not helped either by so many abandoning real programs in favor of bloated web pages (often NodeJS based), or even wraps as "programs" running locally, of the horror. MS Office has been laggy for a while, but even LibreOffice has gotten too slow for some CPUs now (data ends up in the wrong cell, etc.). So with what should be "easy" workloads on the CPU are now commonly CPU and memory hogs, and can sometimes even slow down everything when bogging down the event system.
So for these reasons, having a fast CPU will continue to be important for a responsive user experience. APX, which the subject of this topic, will help in this regard.
I get what you’re saying on the user side, but I was never a proponent of “just get better hardware” when dealing with shit software. I know that it’s a pipe dream, but I would much rather that devs actually optimize their code and stop bloating things with pointless shit than operate on the “fuck it, we ball, hardware will bruteforce decent performance anyway”. Some increased requirements are fine, it’s expected, but what happens now is borderline unacceptable. It’s obviously that this is mostly me screaming into the void, but yeah, I would perhaps rephrase myself and say that OBJECTIVELY modern CPU performance has been enough for most users and their daily tasks, but the negligence of software provider make it seem like one needs to run a TR workstation just to write a text document.
Unfortunately it's been the attitude since Java was all the rage in the late 90s when CPUs were progressing quickly; why optimize when CPUs will be faster next year anyways? People back then didn't realize that the relative cost of e.g. a misprediction or a cache miss is actually increasing.Being an "old school" programmer, bloat has always felt "painful" to me, and am of the opinion that resources should always be wisely managed. But having followed it so long, it seems to me that bloat comes in "waves", along with whatever the new trend or gimmick is hip in CS.For me maintaining both older and newer systems for the purpose of software development, I've noticed a shift in recent years. E.g. Sandy Bridge was just fine for "normal office use" a few years ago, now it's painfully slow. Even on a Zen 3 LibreOffice is annoyingly slow vs. Raptor Lake, and all of this on Linux. On Windows (which I rare use at home) it's even worse; even a Raptor Lake is lagging in MS Excel, Visual Studio etc. is also sluggish. It's getting to a point where you almost need a speedy new CPU to do "simple" tasks effectively, but not necessarily a 24 core Threadripper, it's more about core speed (and having appropriate cooling to avoid throttling etc.).
There's a reason why I often speak about optimizations and improvements in software; it's to draw attention to it in hopes that people will realize how great computing can be. I do have some hopes for APX as mentioned, as this have the potential to greatly affect the cost of logic in code and advance the field of computing, but there is also the downside; many will see this as an excuse to do even more bloat than before. (This is also what's happening in gaming in the past 5+ years; lower overhead APIs have not led to better games, just even more bloat. And every big game is almost broken at launch... But I will not go deeper into that in this thread.)
A good upgrade is a least 90-100% extra performance for the same price you got your current setup as a bare minimum.