Monday, September 23rd 2024

Intel's 8P+32E "Arrow Lake Refresh" Rumored Cancelled

Intel's new 8P+16E "Arrow Lake" silicon powers not just the company's upcoming Socket LGA1851 desktop processors as the "Arrow Lake-S," but also its enthusiast mobile processors, as the "Arrow Lake-HX." The company is debuting this silicon later this year with the Core Ultra 200-series, which will see a ramp through 2025 across its desktop and mobile variants, but Intel is (or rather, was) slated to introduce an "Arrow Lake Refresh" microarchitecture as a follow-up, with a late-2025 debut and 2026 ramp. According to Olrak29, a reliable source with Intel leaks, this refresh has been cancelled.

The "Arrow Lake Refresh" microarchitecture would have been unlike "Raptor Lake Refresh," which was essentially the same silicon, but powering speed bumped SKUs under a newer 14th Gen Core branding, it would have been a physically larger chip, with more cores. Specifically, Intel would give the "Arrow Lake Refresh" an 8P+32E configuration, with eight "Lion Cove" performance cores, and eight "Skymont" E-core clusters for 32 efficiency cores. Intel claimed to have made huge strides with increasing the IPC of its E-cores with "Skymont," and the 8P+32E "Arrow Lake Refresh" would have been a multithreaded performance monstrosity.
There are many reasons Intel could have cancelled the 8P+32E "Arrow Lake Refresh." It probably finds its maxed out Core Ultra 9 285K competitive with the Ryzen 9 9950X in multithreaded productivity workloads; or it's simply cutting costs from having a new silicon release cycle, and will probably just release speed bumps for the 8P+16E silicon, like it did with the "Raptor Lake Refresh," regardless of how the 285K is competitive against the 9950X, or the possible 9950X3D.

Intel's public or leaked roadmaps see the "Nova Lake-S" succeed "Arrow Lake-S" on desktop, with its introduction slated for 2026, which should mean that the upcoming "Arrow Lake-S," and its possible refresh with an unchanged core-count, has to face AMD for at least 2 years. AMD recently released the "Zen 5" microarchitecture, which was received with mixed reviews, but could claw back with variants of the chip that feature 3D V-cache, under the Ryzen 9000X3D series. AMD confirmed that its succeeding "Zen 6" microarchitecture is on-track for a 2025 unveiling, with product launches expected either within 2025 or in 2026.

Meanwhile, Intel has bigger problems. Its underperformance in the stock markets means that the company is exposed to either amicable or hostile takeover moves, with the most recent utterance of the A-word coming from Qualcomm, according to a WSJ report.
Add your own comment

40 Comments on Intel's 8P+32E "Arrow Lake Refresh" Rumored Cancelled

#26
phanbuey
oxrufiioxoI'm so sad, I was so excited for more ecores..... Not sure what to do with my life now.
time to link 32 raspberry pis together and play DOOM II?
Posted on Reply
#27
john_
RogueSixThat would be kinda sad if true and another testament to how deep in shit Intel is. Intel Arrow Lake was originally supposed to be produced on Intel's 20A node before that node was scrapped/cancelled. Intel were forced to go with TSMC N3B for ARL because their own IFS is too incompetent to produce Arrow Lake.

I would have expected that Intel would bring the ARL refresh back home, i.e. ARL-R on the Intel 18A process node next year to demonstrate that the node is at least equal or more advanced than TSMC N3B.

So, my theory is that ARL-R has been cancelled specifically for that reason. Intel want to shy away from that comparison because they know exactly that TSMC N3B is kicking their 18A node's ass to hell and back. So much for catching up, or even outperforming, TSMC next year. Not gonna happen.
I have a somewhat more optimistic theory.
Intel sees that it's manufacturing is doing well, probably better than expected, so they might think they can bring their next CPU line(Nova Lake) closer for release, meaning a refresh of ARL wouldn't be making sense, especially when they will have to pay TSMC to build it.
Posted on Reply
#28
Eskimonster
TheEndIsNearNo more waste of sand
hahahahahha
Posted on Reply
#29
kondamin
I doubt there are enough memory lanes to feed 8+32 if the new e cores are as powerful as claimed
Posted on Reply
#30
bonehead123
DristunAgent 47, you have a new task... The leakers have gone rampant - they are now cancelling the rumours they had previously created themselves. This has got to end, their meddling is no good for the sanity of our Agency. The methods are up to your choice. Good luck, 47.
Yea, but now that they know that we know that they know that we know, is it REALLY that big of a surprise anymore ? :D
Posted on Reply
#31
TheinsanegamerN
kondaminI doubt there are enough memory lanes to feed 8+32 if the new e cores are as powerful as claimed
Intel is pushing 10000mt/s with arrow lake on the extreme end. That's 160GB/s.
Posted on Reply
#32
watzupken
32 E-cores is insane at this point in my opinion. There are trade offs with HT vs physical cores. Removing HT also means you need more physical cores to improve multithreaded performance. While performance is expected to be better with physical cores, it also requires more power.
Posted on Reply
#33
ratirt
Intel is not going to increase the number of the ecores and the CPU is cancelled?

Posted on Reply
#34
Minus Infinity
Vayra86You might be on the money here
Mobile is more important, so no way they take precious 18A supply away from Panther Lake for an Arrow lake refresh. 18A isn't even ready for prime time. They have a lot of work to do.
Posted on Reply
#35
azrael
nguyenYeah Intel is getting smarter, nobody like refreshes that bring nothing to the table except accelerated degradation :cool:
"The candle that burns twice as bright burns half as long ...and you have burned so very very brightly..." :D
Posted on Reply
#36
persondb
TheinsanegamerNIntel is pushing 10000mt/s with arrow lake on the extreme end. That's 160GB/s.
That is overclock, there is absolutely no guarantee on anything.

Yes, there will be probably be quite a few people running such thing, but it's absolutely not at stock and they can't sell a chip on such grounds.
Posted on Reply
#37
InVasMani
I wonder what the reason behind canceling it actually is. It could be any number of things given the recent events that have developed around Intel.
Posted on Reply
#38
Minus Infinity
InVasManiI wonder what the reason behind canceling it actually is. It could be any number of things given the recent events that have developed around Intel.
Cost cutting. Why would it be bad for a cpu lifecycle to be 2 years anyway? Buy Arrow lake or wait for Nova Lake. 40 core desktop would only be needed by someone wanting HEDT anyway and they'll buy TR/ Xeon W IMO.
Posted on Reply
#39
JWNoctis
A not-quite-relevant trivia: Early versions of the Larrabee were also supposed to have 32 of what could now be called E-cores.

A 32-core x86 processor in PC would have been pretty outlandish back in 2008, even though it was supposed to be a GPU.
Posted on Reply
#40
persondb
JWNoctisA not-quite-relevant trivia: Early versions of the Larrabee were also supposed to have 32 of what could now be called E-cores.

A 32-core x86 processor in PC would have been pretty outlandish back in 2008, even though it was supposed to be a GPU.
It did happen as Xeon Phi, the successor of Larambee. The first version released which was Knights Corner used a modified version of the original Pentium cores, with a notable addition being SMT 4(i.e. 4 hardware threads per core).

The next iteration of Xeon Phi which is Knight Landing then used a modified Atom core(Airmont) also with SMT 4 having chips with up to 72 cores. Those atom cores are what we know today as E cores or rather being a direct lineage from it as architecture wise it went

Airmont -> Goldmont -> Tremont -> Gracemont(Alder/Raptor lake E-core) -> Crestmont(Meteor Lake E core) -> Skymont (Arrow and Lunar Lake E-cores).

So it did already happened in the past though those were modified versions of it, with the mentioned Hyper Threading x4 and also having full fat AVX512 implementations.

For other consumer products, we could argue that the PS3 processor, the Cell engine, was also very similar. It had a single 'fast' core and 8 others slower(same clock but lower IPC due to simpler design) cores(but which had strong vector capabilities).
Though those cores didn't have direct access to the system memory and had only indirect access through DMA.

It doesn't seem like the idea of many very simple cores paid off as the Atom line basically developed into basically big cores with very complex out of order designs instead of the simpler in order designs of the past. And maybe that is why Intel cancelled 8+32, there was just too little tangible benefit to have that many cores for the intended consumers.

For those who actually can use such high core counts, they would likely be best served by the Xeon W or Threadrippers series.
Posted on Reply
Add your own comment
Nov 25th, 2024 00:16 EST change timezone

New Forum Posts

Popular Reviews

Controversial News Posts