Thanks! I was wondering about the test details and didn't see slide 87. So as usual they have compared with the weaker AMD 890M to get 16% better results.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Qualcomm revealed X Elite's benchmark scores
- Thread starter sunny5
- Start date
- Sort by reaction score
Thanks! I was wondering about the test details and didn't see slide 87. So as usual they have compared with the weaker AMD 890M to get 16% better results.
Aye here's what I think is going on (not 100% sure):
1. Intel is testing both their device (the S14) and the AMD S16 under performance mode best performance (although powerplan is also balanced? and they're different?). (Side note obviously the M3 Air has no such mode and at 20W is going to be a substantial performance deficit to both the HX370 and the 288V which are using more than 30W probably 40+W)
2. NBC reports that the S16 is power limited by default:
Intel likely overrode those defaults to make the S16 behave closer to the P13/P16 but if it did, it's likely the chassis isn't designed for that thermally hence why the default power of S16 is much lower than the P13/P16 which could affect the S16 results. To be fair, the Intel chip is in the ASUS S14 which may have the same or worse thermal limits (I don't know) and so it could be they chose the S16 as the most comparable HX370 chassis. So they might not have done anything terribly out of bounds or nefarious here, but given the history of first party benchmarks, especially from Intel ... well I'm a little suspicious. That and the tests being all done at "medium" presets.
Either way neither of the GPUs are comparable to the 10-core M3 at this power envelope or settings. These chips are using possibly double the power and possibly more, especially for the HX370. They're both much more similar to a full 18-core M3 Pro in power draw (if not using more power) - which also fits into the 14" Mac Pro chassis (which Intel classified as a thin and light) that they put the M3 into and they didn't the M3 in its Air chassis with an Intel chip also designed for that (fanless) power envelope.
Last edited:
Intel explained its new GPU architecture some time ago.
Slides: https://www.intel.com/content/www/u...intel-tech-tour-xe2-and-lunar-lake-s-gpu.html
Last edited:
I just want to remind people that native doesn't equate to well-optimised.
Something like the Resident Evil games seem very well-optimised for macOS, so hopefully someone will try and use that for a comparison.
Something like the Resident Evil games seem very well-optimised for macOS, so hopefully someone will try and use that for a comparison.
According to Intel's slide, the Intel SoC achieves 76 FPS and the AMD SoC 51 FPS in Resident Evil Village. Unfortunately, I can't find how much M3 MBA achieves under similar conditions (medium settings, at 1080p without upscaling).
Isn't part of the job of a GPU driver to optimise games for that GPU? Isn't that why AMD/Nvidia update their GPU drivers so often?
IMO, if you need to delegate game optimization to the drivers, that is a sign that your entire software stack has failed. I surely hope that with modern GPU APIs the need for driver optimizations is mostly behind us. Of course, as vendors sell more hardware by performing well in popular games, driver shenanigans will always be a thing.
Of course, others will make faster ARM chips than Apple's. If this one is not faster then some other one will be.
Apple's options are very limited because first, the chip has to work in a phone with a smaller battery. Then they put an SOC based on that chip in the Mac. Apple's first priority is always the iPhone.
Other companies can make a chip that is designed to run on a desktop PC that is plugged into AC main power. When you have basically unlimited power available from a wall outlet the chip can be MUCH faster.
These very fast ARM chips are already used in servers, they are overkill for most users, too expensive and they have unneeded amounts of computing power. Look at https://amperecomputing.com/briefs/ampereone-family-product-brief
Do we need 192 performance cores and 4TB or RAM? If you do, you can already buy that today. But Apple will NEVER sell anything like that
The question is not "What can modern technology do, but rather what will Apple decide to sell? Apple sells iPhones (and some other stuff as a side line)
Apple's options are very limited because first, the chip has to work in a phone with a smaller battery. Then they put an SOC based on that chip in the Mac. Apple's first priority is always the iPhone.
Other companies can make a chip that is designed to run on a desktop PC that is plugged into AC main power. When you have basically unlimited power available from a wall outlet the chip can be MUCH faster.
These very fast ARM chips are already used in servers, they are overkill for most users, too expensive and they have unneeded amounts of computing power. Look at https://amperecomputing.com/briefs/ampereone-family-product-brief
Do we need 192 performance cores and 4TB or RAM? If you do, you can already buy that today. But Apple will NEVER sell anything like that
The question is not "What can modern technology do, but rather what will Apple decide to sell? Apple sells iPhones (and some other stuff as a side line)
You are deeply confused.
So far, nobody has made an ARM chip faster than Apple's, for ST, by a long shot. (In fact nobody makes *any* kind of chip faster than Apple's, for ST, regardless of arch, including x86/x64.) Apple is multiple generations ahead of everyone. For MT, of course the number of cores will dominate. So Apple dominates as far up the food chain as it cares to compete, and no further. They're not trying to build DC servers (at least, they haven't been - Apple Intelligence may mean things change in the near future). So they don't have a 128-core chip, and they're not competing with Amazon Graviton or Ampere. But all of this is irrelevant to what we've been talking about, which is the consumer/prosumer/workstation market- basically, everything that's not a server.
Your claim about Apple's chip having to work in a phone is correct but irrelevant. The A chips aren't in Macs. The related point is that their *cores* have to work in phones. But as we've seen with the M3 and M4, they appear entirely capable of making a single core that works great both downclocked for phones and upclocked for Macs. (Or perhaps it's a single core family, with variants for the A and M chips - nobody really knows at this point.) Alternatively, they think that the cores they're designing are good enough for phones even as they extend them to be able to clock higher in Macs. So that point is either no longer correct, or also irrelevant.
Single-core performance of the Ampere SoCs are super slow. They work in a server but not in a personal computer.
They target entirely different markets and workloads.
Just take a look at this Geekbench 6 comparison of the newest iPad Pro and a 128-core Ampere Altra Max.
What?? That's why they made the M chips which stands for Macs. They're two separate chip families even if they have many things in common. Apple can make any M chip they want without any consideration to iPhones. If they were limited they couldn't make M2 Ultra with 24 CPU cores and 76 GPU cores because that certainly doesn't fit in an iPhone. 😄
27 hours of battery life is too good to be true but if true then 2024 is the death of ARM. I'll likely acquire one to compare against 4nm 8840HS which is delivering about 12 hours which I thought was long but 27 is crazy.
www.pcgamer.com
These new Asus Lunar Lake laptops with 27+ hours of battery life kinda prove it's not just x86 vs Arm when it comes to power efficiency
Intel's new Lunar Lake chips are helping Asus deliver battery life equaling, or even excelling, Qualcomm's ARM chips.
Oh I’m sure it’s 27 hours…when watching a 640x480 video at 5 nits on a loop with WiFi off.27 hours of battery life is too good to be true but if true then 2024 is the death of ARM. I'll likely acquire one to compare against 4nm 8840HS which is delivering about 12 hours which I thought was long but 27 is crazy.
These new Asus Lunar Lake laptops with 27+ hours of battery life kinda prove it's not just x86 vs Arm when it comes to power efficiency
Intel's new Lunar Lake chips are helping Asus deliver battery life equaling, or even excelling, Qualcomm's ARM chips.
Yes, I'm sure Apple will declare its entire HW product line obsolete by the end of 2024 when Lunar Lake hits the market with an iGPU as powerful as the base M3. In fact there are rumors that the iPhone event on Monday will be their last and they're planning to cancel the launch of M4 Pro/Max/Ultra later this year. "We are the Intel. You will be assimilated. Resistance is futile".
Last edited:
Apple's priority being iPhone doesn't imply they design only for the iPhone. For quite some time Apple has consistently had the fastest Arm cores in the world, and since the debut of M1, they're frequently competing for the fastest single thread performance over all ISAs. That's true right now - there is nothing out there as good as M4's P core.
Of course - Apple isn't in the server business. They make client devices, and optimize their silicon for client workloads.
If Apple built a 192-core Mac using that Ampere chip, a huge number of Mac applications would run better on a M1 Macbook Air, the slowest Apple Silicon Mac that will ever exist. The price of packing 192 cores into one monolithic die is that each individual core is pretty disappointing compared to cores designed like Apple's. Whenever you run an application that doesn't scale beyond a few cores, you want faster cores over an enormous number of slower ones.
If they end up taping out DC-specific silicon to support Apple Intelligence, I doubt it will look much like Ampere's products, or other traditional cloud-oriented DC silicon. Instead of focusing on CPU core count, they'd put in lots of NPUs, and as few CPU clusters as they think they'd need to 'manage' the NPUs.
27 hours of battery life is too good to be true but if true then 2024 is the death of ARM. I'll likely acquire one to compare against 4nm 8840HS which is delivering about 12 hours which I thought was long but 27 is crazy.
These new Asus Lunar Lake laptops with 27+ hours of battery life kinda prove it's not just x86 vs Arm when it comes to power efficiency
Intel's new Lunar Lake chips are helping Asus deliver battery life equaling, or even excelling, Qualcomm's ARM chips.
The 27 hour claim is absolutely achievable. With a 75Wh battery that is 2.7 watt average power draw, so enough to run a very power-efficient display on low brightness. Certainly a testament to how far the display technology got. Combine it with some optimizations and you can get the CPU consumption almost to zero when playing a video.
I would be more curious about the battery life under more realistic circumstances. Not many people watch low-res videos with screen dimmed all the way down and WiFi turned off. I’d like to see how these platforms perform when the CPU actually needs to do some work.
even if its true and achievable under limited circumstances , after decades of x86 evolving they can "beat" arm that is on the market , for real this time (not like in the past) for half a decade...good analogy
Last edited by a moderator:
I wonder if Apple has felt the need to make a comparison with Copilot+ PC because it feels it is losing some of its lustre.
Apple's next laptop presentation could be fun.
Apple's next laptop presentation could be fun.
video editing and playing games is not "best AI PC" is kind of miss-leading and Apple should wait until the release of the M4 Mac series. When Apple released the M3 didnt had this marketing message
Cannot wait to see the what M4 Pro and Max can do...that will put a big stain on the laptop segment based on what we can see from the current M4
Cannot wait to see the what M4 Pro and Max can do...that will put a big stain on the laptop segment based on what we can see from the current M4
Or, you know, people keep claiming they’re ’behind’ in AI and that’s not the case. But I guess “losing some of its lustre” is a more engaging narrative.
That's entirely plausible, but it's not the only reasonable possibility. If they want to start pushing Mac CPU cycles into the cloud, the way they're pushing some inference, they'll have servers with lots of big cores. When I said Apple Intelligence, I was talking about the entire PCC model, not just the AI stuff they're releasing next month (?). So sorry if that was confusing.
Honestly, I'm not sure I like the idea of pushing regular CPU work to Apple's DCs, even if I trust their privacy and security models (which I probably do). But it certainly does offer customers flexibility, and Apple a potential new revenue stream, so... I wouldn't write it off.
What makes you think it is misleading? From what I understand M3 outperforms most Copilot+ PCs in on-device inference benchmarks. Advertised TOPS number is one thing, ability to use it effectively in ML workloads is something else entirely.
the best AI PC as a mac...ML is not up there with current Copilot PC
For me ML=AI from pc world just under a different name. In terms of what AI can do under an Copilot PC is much more vast and responsive than the M3 Mac. Couple of months from now with M4 family with better NPU and official release of the macoS Sequoia it will be for sure a different story
Last edited: