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"
Yesterday, someone else said it better than me.
The "official" narrative is that Intel sucks, I really wonder how the criticism towards AMD would have looked today if Arrow Lake would've had slightly better gaming performance than Raptor Lake. :rolleyes:
www.pcmag.com/opinions/intels-raptor-lake-bug-struck-my-pc-a-year-later-heres-how-i-fixed-it#Crappy motherboards aren't really a thing unless you go and look for one, a majority of $200-300 boards will have decent VRM and M.2 heatsinks, then again M.2 heatsinks are cheap or come pre-installed on some SSD's. But I have to wonder just how many M.2 drives people actually need to fill every slot on the board, IMO a NAS would be a better option for mass storage.
What makes someone have to upgrade if the socket has better longevity other than FOMO? It isn't any different if the board had no upgrade path, if someone feels pressured to upgrade with every generation because of some extra USB ports or a 10GB NIC then at least having some socket longevity would save some cost over a whole new system. A 10GB NIC isn't exactly expensive to add as a pci-e card.
And I don't think they made a bad decision, if they don't care about longevity with upgrades, and want e-cores more with the extra I/O features then that was their own decision, not really wrong as either choice is fine.The narrative for Intel with how amazing e-cores are, a review is biased because they didn't use core 200 boost, or they didn't use fast enough RAM vs. an AMD system would be quite a bit stronger.
People are way too obsessed with AMD because of the socket and seeing the 9800X3D beating Intel's 9950x competitor (285K) in gaming. I really wish more tech reviewers would focus on comparing Intel chips with the non-X3D parts because thats what they're actually meant to compete withOnly found out recently the 245K is supposed to compete with the 9700x and not 9600x lol. AMD having two ryzen 9s threw me offGood point on the Linux benchmarks, they do seem a bit better compared Windows, skimming through them right now.I think the performance jump was ok, more than Zen 4 -> Zen 5. but yeah the degredation was an awful look for Intel$200-$300 boards are certainly nice. But apparently according to ThomasK, $200 is high end for a boardI do think the ram stuff is a bit of a cope tbh, but still nice that Intel can take advantage of fast ram ig. 200S Boost isn't even stable for me so no comments on that :sleep:. The Skymont E Cores really shocked me on how good they are though. P Cores OFF on my 245K and its still compiling 1.5x faster than my old i5-10400 with only E cores
So if Zen6 X3D beats Nova Lake with bLLC then AMD fans will be pleased.
For the MT crowd it's a bit different, we will see if X3D will bring a larger gain in non-gaming compared to Zen5, if it does then the 2 CCD with X3D has to beat Nova Lake in MT. Not easy considering both will have 48 threads max (or so the leaks say). I think the MT crowd is more pragmatic and less fanatical than the X3D gaming crowd, and so if Nova Lake is compelling enough they could have a different attitude towards it compared to Arrow Lake.My thoughts exactly.
I gave my board flak for it. That M.2 in the middle just doesn't belong.
I've explained there what benefits its removal would bring. But apparently I forgot to mention the freed up 4 lanes that could be used for something else.
As the other guy here randomly said: take a chill pill.
Timed Linux Kernel Compilation (Phoronix Test Suite):
245K 6P/8E: 77s
245K 0P/8E: 142s
my old 10400 6C/12T: 211s
EDIT just benchmarked P-cores only to give a better perspective
245K 6P/0E: 147s
Extremely interesting, the E cores are good enough that in multithreaded loads 8E beats 6P (maybe because these P cores lack HT? The 9600x would score around 108s and 9700x would score around 84s)
Like it only happened with 2 mobo/socket since 2008.
Most recently it was with the LGA 1700 socket in my case even if thats kind of weird to some.
I was an early adopter of the i 3 12100F coming from an aging first gen Ryzen system with a 1600x. You might wonder why the hell did I do that instead of upgrading the CPU in my AM 4 mobo.
Well its cause of multiple reasons.
First of all the game I was mainly playing and I was planning to play it for long term was single thread heavy and did not care much for the extra cores/threads and the 1600x had way too big drops when a lot of stuff was happening on the screen since it was an MMO.
Second reason was the prices in my country, even a second hand 3600 was about the same price as a 12100F and I did not feel like paying that much for an already aging CPU that wont even perform better in that game. 'Actually it did worse'
5000 serie was out of my budget range and the non X 5600 did not exist yet and on top of that my mobo was like the last one to get the BIOS update and I got tired of waiting for Asus to relase it..
At that point I've did some research and came to the conlusion that selling the AM 4 system and switching to a DDR 4 based Alder Lake is my best option as long as I can sell my AM 4 mobo+CPU and luckily I've managed to do that very easily.
I've used that 12100F for 2+ years with no issues until these shader compiling games became more common and it wasn't really cutting it anymore so I've upgraded to a second hand but brand new 12600KF for cheap like it was cheaper than a 13400F so even at stock speeds it was a better choice. 'I can't OC with my mobo after all since I've never planned for that nor cared about it'
Then I've upgraded yet again earlier this year to a 14600KF from the same seller since it was a way too good deal to pass on and he did buy my 12600KF so the whole upgrade cost me around $80 and this is my final CPU in this socket and mobo but I don't think that I would need anything faster anytime soon anyway.
Btw even if I've stayed on that AM 4 I would maybe have a 5700X3D or a 5800X3D so I wouldn't be in a better place + those X3D CPU's are stupidly overpriced around here so I'm not even sure about that so in the end I think I've made the right decision. 'E cores actually do help so I don't mind having those either'
As for my future next build if it ever happens, we will see but I'm open to both Intel and AMD depending on what they will offer in my budget range at the time.:)
In first three cases, first chip actually has only 8 lanes usable for devices:
4 lanes are used for uplink to CPU, and then other 4 for link to second chip, which has 12 lanes free, but with significant added latency. (and power waste)
That gives 20 PCIe lanes from chipset in X670, X670E and X870E boards.
X870 with that single B650 chip again gives only 12 lanes.
Then those lanes are/can be used for USB ports etc added controllers reducing amount available.
en.wikipedia.org/wiki/Socket_AM5#Feature_table
Hopefully AMD brings out new fully featured full size chip"set" with Zen6 to give good amount of I/O and with proper power efficiency to go against Intel's Z-serie chipsets.
(which have 6W TDP vs 14W of X670/X670/X870E)
We're now three pages in, and as usual it has derailed into a discussion about socket longevity, and the amount of copium and mental gymnastics needed to justify the prospect of keeping an obsolete motherboard with a brand new CPU is just comical. While there are edge-cases where it could make sense, these are exceptions, even among DIYs, not good enough to dictate buying preferences in general. (While more relevant upgrading paths like IO are ignored.) It is also an astute observation that AMD buyers have a much stronger desire to replace their product, sometimes almost immediately, which reveals something deeper about their true level of satisfaction with the product.
And I would imagine that a lot of these buyers would be much more satisfied with a Threadripper, since they are basically "paying for it" with upgrades every two years, but also any heavy workload that truly scales with >8 cores would be severely limited by power and IO. And this is the big problem with AM5; great "on paper" and in synthetics, but in real life it quickly gets too limited one way or another. And lastly, to those who are in the Linux world, you already know the performance scaling is much better for high core counts there, but you haven't really experienced how great and responsive a workstation can be until you've tried it with lots of RAM.I don't put much faith in "leaks" without any facts. The "consensus" for Zen 5 was ~30-40% gains up until real leaks arrived. And as I remind people of every cycle; not even the manufacturer knows the overall performance until a few months ahead of release. Even if the target is 20% IPC gains etc. from simulations and estimates, the real world capabilities may differ once they tune the engineering samples and have to work around all the numerous defects and timing issues.
As always, I'm hoping for as much as possible. I'm also very curious to see when APX is arriving from AMD.If Zen 6 remains on AM5, those options will be limited. Having more lanes off the chipset doesn't help much when the lanes to the CPU and the free CPU lanes are still limited.
What exactly makes a motherboard obsolete after just a few years? Not everyone needs a bajillion pci-lanes, pci-e 5.0, or an integrated 10gb NIC.IMO, that's just an opinion, those buying AMD only have the want to replace the CPU when the upgrade warrants doing so, ie. going from a AM4 2700X to a 5800X3D is an upgrade Intel wouldn't give you without needing to buy a new motherboard.I personally wouldn't, the cost of entry is too high, and given the astronomical prices on RAM, HEDT is out of reach for a majority of people.