Sunday, January 28th 2024
Intel "Panther Lake" Targets Substantial AI Performance Leap in 2025
Pat Gelsinger, CEO of Intel Corporation, has outlined future performance expectations for the company's Core range of processors. In a recent fourth quarter 2023 earnings call he declared: "The Core Ultra platform delivers leadership AI performance today with our next-generation platforms launching later this year, Lunar Lake and Arrow Lake tripling our AI performance. In 2025 with Panther Lake, we will grow AI performance up to an additional 2x." Team Blue's Intel Core Ultra "Meteor Lake" mobile processors arrived right at the tail end of last year, as a somewhat delayed answer to AMD's Ryzen 7040 "Phoenix" APU series—both leveraging their own AI-crunching NPU technologies. Gelsinger believes that the launch of Lunar Lake and Arrow Lake Core product lines will bring significant (3x) AI processing improvements over Meteor Lake. He seemed to confident in a delay-free release schedule for the new year and beyond: "We are first in the industry to have incorporated both gate-all-around and backside power delivery in a single process node, the latter unexpected two years ahead of our competition. Arrow Lake, our lead Intel 20A vehicle will launch this year."
He proceeded to gush about their next node advancement: "Intel 18A is expected to achieve manufacturing readiness in second half 2024, completing our five nodes in four year journey and bringing us back to process leadership. I am pleased to say that Clearwater Forest, our first Intel 18A part for servers has already gone into fab and Panther Lake for clients will be heading into Fab shortly." Industry experts posit that Core "Panther Lake" parts could borrow elements from the next generation Xeon "Clearwater Forest" efficiency-focused family—possibly the latter's "Darkmont" E-cores, to accompany "Cougar Cove" P-cores. The Intel CEO is quite excited about the manufacturing outlay for 2025: "I'll just say, hey, we look at this every single day and we're scrutinizing carefully our progress on 18A. And obviously the great news that we just described those Clearwater Forest taping out, that gives us a lot of confidence that 18A is healthy. That's a major product for us. Panther Lake following that shortly."
Sources:
Intel Financial Report, Tom's Hardware, Wccftech, VideoCardz, Seeking Alpha
He proceeded to gush about their next node advancement: "Intel 18A is expected to achieve manufacturing readiness in second half 2024, completing our five nodes in four year journey and bringing us back to process leadership. I am pleased to say that Clearwater Forest, our first Intel 18A part for servers has already gone into fab and Panther Lake for clients will be heading into Fab shortly." Industry experts posit that Core "Panther Lake" parts could borrow elements from the next generation Xeon "Clearwater Forest" efficiency-focused family—possibly the latter's "Darkmont" E-cores, to accompany "Cougar Cove" P-cores. The Intel CEO is quite excited about the manufacturing outlay for 2025: "I'll just say, hey, we look at this every single day and we're scrutinizing carefully our progress on 18A. And obviously the great news that we just described those Clearwater Forest taping out, that gives us a lot of confidence that 18A is healthy. That's a major product for us. Panther Lake following that shortly."
56 Comments on Intel "Panther Lake" Targets Substantial AI Performance Leap in 2025
Panther Lake (top tier, KS rendition of the Core 9 Ultra) seems like an interesting upgrade from where I'm currently standing. But after around 9 months of ownership, I don't think I have even come close to taxing my i9-13900KS yet.
BTW, I did not vote on the poll as the option "None of the above" was not offered!
Regarding the poll it means nothing, Intel chips are a moving target regarding fabs and features, we don't actually know what's going to ship and when.
Whichever Intel CPU will offer 3-5x the performance per watt gaming from their joke of a 14th gen, on a new process like Intel 4 or Intel 20A will be interesting, since then it might actually compete with AMD and Apple.
So Cinebench should be considered as just that, one specific narrow benchmark, and never be used (alone) to extrapolate generic performance.
It can be used to approximate IPC, but then in a mix with ~20 or so other benchmarks, preferably a mix of workload types, and obviously workloads that don't run into other significant bottlenecks at the set clock speed, plus the CPUs must be completely locked at a clock speed far below any throttling. Another avenue is to use synthetic benchmarks for this purpose (a mix which is not just computationally intensive int and float operations, but also stresses the CPU front-end), but in either case there is always debate over which balance is the most fair.
Ultimately, the closest thing we have is a good broad range of CPU loads (not GPU loads like games), take the average of those within a couple of standard deviations to eliminate the outliers.
(No I don't have citations, I can't remember where I saw the numbers, if you really need to see it, Google)
With the old pat in lead, you get the old dog tricks again. "But Intel has AI in it's chips!"
www.techpowerup.com/317317/intel-meteor-lake-p-cores-show-ipc-regression-over-raptor-lakeyhea, I'm not denying that RPL isn't efficient, that's a fact. But doesn't mean that they have a bad IPC compared to the concurrence. They are just massively less efficient. RPL at 125w is as fast as zen 4 at 65w. You can absolutely say that RPL efficiency is shit, but not the IPC. I will stand on that hill unless proven wrong.
I prefer to talk about efficiency, since if you compare zen 3 to zen 4, the IPC uplift goes from meh to awesome depending on what you do, but zen 4 being able to clock that high compared to zen 3 and still being very efficient is what made it a win. Even if intel were to somehow manage to a 25% IPC uplift compared to RPL, it would be meaningless if they pulled that off with an arch that's "only" 5% more efficient, and still scaling poorly once you try to up the frequency.
www.guru3d.com/review/amd-ryzen-7-7700x-review/page-9/
RPL is a funny arch, if you look at TPU review, in ST it's actually efficient compared to zen4, but things start to crumble once you get into MT performance. Meanwhile, AMD efficiency in MT always looked like magic. There's something that AMD is doing right that Intel cannot match at the moment.
From what I understand so far, in 2024, every CPU will have AI acceleration. AMD isn't exactly being quieter about this :D. The difference is that Lisa mostly talks in conferences, when Pat talks everywhere.
stackoverflow.com/questions/14794460/how-does-the-arm-architecture-differ-from-x86