Best benchmarks I've seen for these so far. Essentially great gains for Intel but still falling short in many ways to AMD 7000, AMD 8000 (most likely), and especially Apple M3.
https://www.youtube.com/watch?v=7wZjhlYqZ2I
https://www.youtube.com/watch?v=WH-qtuVRS2c
https://www.youtube.com/watch?v=Obtc24lwbrw
The Just Josh numbers you posted don't square with the ones above. It'll be a while before benchmarks settle, and as with all new devices there will be burps. We'll just see.
But taking a blind average of all of the above and applying to my own requirements perspective: I'm seeing a mid-range device with 80% the single-core performance of a M3 Max and significantly lower idle/regular-use power consumption. That sounds like a better device than a Macbook Pro to me.
Depends how much you focus on the AI parts of the announcement, I guess. I mean, the marketing makes this sound like their brand new, absolute highest end laptop CPU, not their mid-range laptop offering. This article is about an "AI Everywhere event" with a marketing presentation [1] talking about "leadership", "the latest LLMs" and "Local LLaMa2-7B".
Except when it comes to LLMs, it's critical to have lots of high bandwidth memory. The M3 Max can be configured with up to 128GB of memory at 300GB/s, whereas this caps out at 64GB of LPDDR5 and Intel haven't seen fit to mention memory bandwidth but for "LPDDR5/x 7467 MT/s" spec I can't believe it'll be any more than 100GB/s (which I guess is why they aren't mentioning it)
It's good to see Intel is paying more attention to ML, but they're clearly still playing catch up to Apple, let alone to nvidia.
[1] https://download.intel.com/newsroom/2023/ai/ai-everywhere-20...
Ultra is just catching up with amd 7000 and idle consumption is bad compared to m3.
How common is the use case of full CPU loads for hours on battery? Or eeking a 10th hour of light-use autonomy?
My use case doesn't often pin cores these days, but I imagine the same goes for the full CPU load question. It's not common, because it's not possible. Enabling that enables new ways of using it.
And outside of traditional computer use, those questions are both easily answered if you consider handheld PC gaming devices. People post excitedly about their strategies and successes when they squeeze out an extra hour of game time on both the high and low ends.
I started the day at 7:30, worked most of the day on it with multiple apps open - Chrome, Safari, Teams, Slack, VS Code, IntelliJ, DataGrip, Postman, Outlook, etc. Connected to a wireless keyboard and mouse, and streaming music to my AirPods for a few hours of the day. Most of my actively used apps were IntelliJ, Chrome, and Teams (admittedly not doing a LOT of building/running Java cause I was resolving build issues for a chunk of the day). I ended the day at 4:00 with 47% battery remaining.
That was unheard of on my last i9 MBP that could barely survive a few hours on battery, even when it was new.
No it's much worse than that for us Intel laptop users. Our machines make loud noises, overheat and throttle to the point where it affects Zoom video chats. The fan on my X1 carbon even comes on when booting the machine!
https://www.notebookcheck.net/Intel-Meteor-Lake-Analysis-Cor...
I wonder if using TSMC 3nm process as Apple did would have improved things considerably.
While it appears that with "Intel 4" Intel has succeeded to reduce the power consumption to be competitive with the CPUs made in older TSMC processes, like the AMD CPUs, history has repeated and exactly like at the launch of the "14 nm" Intel process in 2014 and at the launch of the "10 nm" Intel process in 2018 and 2019, Intel is unable to achieve in the new manufacturing processes clock frequencies as high as in their previous mature manufacturing process.
Because the P-cores of Meteor Lake have the same microarchitecture as the P-cores of Alder Lake and Raptor Lake, but a lower turbo clock frequency, they have a lower single-thread performance.
Nevertheless, the Meteor Lake E-cores are improved and the clock frequency when all cores are active is higher for Meteor Lake, due to the improved power efficiency, so the multi-threaded performance of Meteor Lake is better than in the previous Intel CPUs.
The best part of Meteor Lake is the GPU made at TSMC, which has 4/3 times more FP32 execution units than AMD Phoenix or Hawk Point at a clock frequency that is higher than 3/4 of the AMD frequency, so the Intel iGPU is faster than the AMD iGPU, and it is twice faster than the older Intel iGPUs.
The other nice feature of Meteor Lake is that the SoC tile (made by TSMC) includes everything that is needed when a computer is mostly idle, like when reading documents or watching movies, so in such cases both the CPU tile with 6+8 faster cores and the GPU tile with the 3D graphics can be completely shut down, leading to a great reduction in the power consumption for anything that can be handled by the two slowest cores and by the video display engine.
For consumers who do not need the higher performance of 45 W CPUs, laptop or SFF computers with Meteor Lake are a great choice.
While one model of 45 W Meteor Lake will be available some time later (Core Ultra 9), it will likely be too expensive and too hard to find in comparison with the 45-W AMD CPUs.