New Intel laptop CPUs out - no need to worry.

[danwells]

Intel's Meteor Lake CPUs debuted today, and the first performance tests are out - one reliable source is PC Magazine... I initially did a double-take, because it was running a little bit ahead of the Mac they had it benchmarked against (about 20% in most tests, 40% in Cinebench R23, which likes Intel architectures). Then I took a closer look at WHAT Mac they were benchmarking it against! A 15" Air!

The chip they had is not the very top chip, but it has the same core configuration as the top chip (6P, 8e, 2lp - ultra-low-power cores - my understanding is that they are smaller than a conventional e-core). It's 200 MHz slower than the top chip, and it's 28W (turboing to 115W), while the top chip will be 45W, and it might turbo a bit higher more of the time. This is what Intel's calling a Core Ultra 7 155H, while the Core Ultra 9 185H is the top chip.

Their 28W chip one model from the top of the line is barely beating our 8W (?, but it's fanless) chip from the last generation. Just going to the latest generation of our fanless chip would pull within 5% in Geekbench, and get within 20% in Cinebench. Since the M3 MacBook Airs aren't out, the base M3 results are from the iMac (I found iMac benchmarks before I found base M3 MacBook Pro benchmarks, and M-series chips traditionally barely vary by chassis). Coming closer to a 28W chip with an M3 Pro, the brand-new Meteor Lake falls behind by 25% in Geekbench and around 40% in Cinebench (odd result).

If Intel dares call that a top-of the line laptop chip, they are calling the M3 Max 16/40 into the arena... Not even close - go home and try again tomorrow. The M3 Max creamed the Core Ultra 7 by darned close to 2 to 1.

To be fair to Intel, the Acer Swift Ultra GO that has been supplied for testing is priced competitively with an Air ($1000 base - doesn't sound like it's built anywhere NEAR as well as an Air), although the same CPU will show up in premium laptops - Lenovo is teasing it in $3000 Thinkpads. Intel may well have made a 28W MacBook Air (with a fan) that can turbo up to 115W. It lasts 8 hours on battery when playing back video (and probably much less doing literally anything else! - they seem to be doing a similar hardware decode trick to Apple). I can't imagine why anyone would want one unless they REALLY need Windows.

 

[Flash1420]

Just remember competition is a good thing. We got Apple Silicon because of Intel. If competition didn't exist, innovation would be lackluster and companies would just get lazy.

 

[Pakaku]

I can't imagine why anyone would want one unless they REALLY need Windows.

People like different operating systems, there's no problem with that. And if I don't like Windows, Linux is a great alternative.

 

[Drifter759]

The reviews seem to be all over the place. I'm guessing Meteor Lake is having teething problems because of their new tiled design; hopefully most of them are fixable with BIOS updates. I expect Arrow Lake (Meteor Lake's successor) to be pretty good, though.

For what it's worth, OEMs seem to have wide latitude in determining MTL's power limits; I have no idea what turbo limits the units out for review today use. More detailed analyses should come out once more people get their hands on them.

Looks like a big step in the right direction for Intel, but they're still pretty far from Apple Silicon efficiency. And in terms of absolute CPU performance, looks like it might be a slight regression from their previous generation chips. Their integrated GPUs are much better, though.

 

[leman]

The awkward thing about Meteor Lake is that single-core CPU scores seem to be slower compared to previous generation, and that's despite the move to the new Intel 4 process. And of course, you still get the "15W" CPUs that actually draw 60 watts or more in benchmarks. The Intel 4 CPU with 6+8 cores barely manage to outperform an M3 with 4+4 cores in multi-core while using 4-5x more power. That must hurt.

I am curious about future investigations from Chips and Cheese and other sources who will inevitably dissect these chips. Is it the new multi-chip architecture that limits the performance and efficiency, does Intel 4 have issue with higher clocks, or maybe something else?

 

[TheRealAlex]

#1. Those new Intel CPU's are the biggest Scam They "Use" Influencers to peddle false hoods and Compare Apple M2 chips to Intels Newest Flagship when they are totally different 53 Watts Vs 75 Watt TDP SKUs They don't even compare them with M2 Pro or M2 Max.
#2. Intel essentially caught up to AMD's 6800 SKU that's all.
#3. No Mention of Unified Memory or Memory speed.
#4. The New Naming Scheme is Confusing on Purpose. Period.
#5. Is it"better" sure but Its no where Near What AMD has on Tap next year or its 3 Generations behind Apple.

 

[leman]

#3. No Mention of Unified Memory or Memory speed.

Intel used unified memory on their SoCs since 2012. Memory is up to 96GB LPDDR5X-7467 (119.5GB/s for the usual 128-bit memory bus). I'd expect most of these SoCs to ship with the usual LPDDR5-6400 for the same bandwidth as M2/M3.

 

[Basic75]

6P, 8e, 2lp - ultra-low-power cores - my understanding is that they are smaller than a conventional e-core.

Intel's regular E cores are optimized for small area. They exist to give maximum multithreaded bang-for-the-buck within a limited die size. The new LP cores are optimized for low energy consumption. The interesting thing is that they are not located on the CPU tile together with the P and E cores, they are on the SoC tile. This allows the entire CPU tile to be powered down and still be able to run some background tasks on the two LP cores. We'll have to wait and see how much near idle and connected standby power consumption benefit from this.

 

[Basic75]

Intel used unified memory on their SoCs since 2012.

Here's a question I wanted to ask for a long time, how "dynamic" are current UMA implementations?

On an Amiga (assuming an A500 or A1200 with only chip RAM) there is no concept of partitioning the memory between CPU and GPU. Every chunk of memory that is allocated by an application can be used by the CPU and/or the GPU (if I may call the Amiga chipset a GPU.)

On Intel MacBooks with UMA the system information would usually say something like 1536MB of RAM for the iGPU, which tells me that some kind of partitioning is going on.

I haven't done a technical deep dive on this subject, but perhaps somebody can confirm that Apple's UMA is as flexible as the Amiga's?

And Intel's isn't? What about AMD? I believe that they have gained the ability to dynamically resize the portion of the RAM that can be used by the iGPU, but that's not the same thing as being able to freely place GPU resources anywhere in main memory.

 

[NEPOBABY]

Intel used unified memory on their SoCs since 2012. Memory is up to 96GB LPDDR5X-7467 (119.5GB/s for the usual 128-bit memory bus). I'd expect most of these SoCs to ship with the usual LPDDR5-6400 for the same bandwidth as M2/M3.

Their UMA is obviously a lot more limited in terms of what the iGPU can do and all the system wide software optimizations.

The closest UMA implementation to Apple was SGI’s circa 2000.

 

[NEPOBABY]

Here's a question I wanted to ask for a long time, how "dynamic" are current UMA implementations?

On an Amiga (assuming an A500 or A1200 with only chip RAM) there is no concept of partitioning the memory between CPU and GPU. Every chunk of memory that is allocated by an application can be used by the CPU and/or the GPU (if I may call the Amiga chipset a GPU.)

The first GeForce card was officially the first graphics card called a ‘GPU’ because it could do certain compute tasks on the processor. Before that we used the terms graphics accelerators or video accelerators.

 

[leman]

I haven't done a technical deep dive on this subject, but perhaps somebody can confirm that Apple's UMA is as flexible as the Amiga's?

And Intel's isn't? What about AMD? I believe that they have gained the ability to dynamically resize the portion of the RAM that can be used by the iGPU, but that's not the same thing as being able to freely place GPU resources anywhere in main memory.

Both Intel and Apple iGPUs can access any memory pages that the CPU can access too. I am not aware of any partitioning on either platform. The "reserved RAM" bit on Intel Macs you mention was just the way the driver worked, not any hardware restriction. And I do not know whether that RAM represented the limit of possible GPU memory allocation for the driver or merely was a pool reserved for the graphics operation.

At any rate, Intel has OpenCL extensions that can map any CPU memory allocation to be GPU-visible, just like Apple does.

Their UMA is obviously a lot more limited in terms of what the iGPU can do and all the system wide software optimizations.

I see no evidence of that.

The main difference between Apple's UMA and Intel's UMA is that Apple ships large amount of fast cache and their memory controller is optimised for GPUs with massive queues and very high memory-level parallelism. And I am sure that Apple has plenty of its own magic sauce (like QoS on memory packets, reserved bandwidth etc).

 

[Basic75]

The main difference between Apple's UMA and Intel's UMA is that Apple ships large amount of fast cache

Right, when did Intel stop shipping the 128MB eDRAM L4 cache? Though the current Intel chips aren't half bad, they have up to 36MB of L3, and 24MB for mobile chips. (I don't count Intel's HX line of rebranded desktop chips as "mobile".)

 

[Basic75]

Both Intel and Apple iGPUs can access any memory pages that the CPU can access too. I am not aware of any partitioning on either platform.

Do the GPUs have an MMU to work with the same virtual addresses as the applications do on the CPU? If not, how do data structures with pointers work?

 

[NEPOBABY]

I see no evidence of that.

The main difference between Apple's UMA and Intel's UMA is that Apple ships large amount of fast cache and their memory controller is optimised for GPUs with massive queues and very high memory-level parallelism. And I am sure that Apple has plenty of its own magic sauce (like QoS on memory packets, reserved bandwidth etc).

Thanks for posting the evidence of that

 

[leman]

Right, when did Intel stop shipping the 128MB eDRAM L4 cache? Though the current Intel chips aren't half bad, they have up to 36MB of L3, and 24MB for mobile chips. (I don't count Intel's HX line of rebranded desktop chips as "mobile".)

Well, Intel's eDRAM only offered about 50GB/s, so it became obsolete with DDR4 and later tech. I don't remember when they stopped shipping it.

Do the GPUs have an MMU to work with the same virtual addresses as the applications do on the CPU? If not, how do data structures with pointers work?

GPUs certainly have MMUs, whether they can work with the same virtual addresses depends on the vendor. Nvidia and AMD support unified virtual memory space (same pointers on CPU and GPU), although AMD seems to warn that it can have a performance overhead. Nvidia doesn't seem to have any such restrictions (and their GPUs support page faulting at hardware level to ensure memory coherency with the CPU even across PCIe). Apple does not support unified virtual memory — CPU and GPU pointers to the same data have different addresses. If you want to have structures with pointers you either have to use offsets or fixup the addresses when writing them into memory. I do not know whether it is a hardware limitation or just something the current driver stack does not support. I also do not know whether Intel supports unified virtual memory (didn't really look at their docs lately).

Thanks for posting the evidence of that

You want the evidence for Apple having more cache? Go look at a spec sheet maybe? As to memory parallelism, Anandtech did a nice review of M1 back in the day, check it out.

 

[california_kid]

Let's say Intel's new CPU proved superior to a comparable M3 in every single benchmark. Does that mean Apple will consider going back to Intel? Zero percent chance for that. Does that mean a decent number of current Apple silicon owners will dump their Macs for an Intel laptop? That's the question. If the performance was percentages better, I doubt Intel will make a dent for Mac owners. If the performance was in multiples better, then we would "worry".

 

[danwells]

What I find interesting is that Apple's performance IS close to "multiples better". The M3 Max is darned close to twice as fast as Meteor Lake in a lot of tests. Apple is also coming close to Intel with the fanless (8 watt?) base M3 chip (the Intel can draw 28 watts standard and 115 watts peak, in the same range as the M3 Max). The benchmarks I've seen are about 20% faster than a base M2 (which is what it's typically reviewed against), and less than 10% faster than a base M3. We haven't seen the fanless M3 machines yet (both the base MacBook Pro and the iMac have fans, though neither uses them much), but I'd be very surprised NOT to see the M3 in both the MacBook Air and even the iPad Pro (from M-series history). . Apple either gets 90% of the performance for a fraction of the power OR nearly twice the performance for similar power.

The new Qualcomm Snapdragon X is a similar situation - they're performing between a base M3 and a Pro, but drawing up to 80 watts to do that (the 23 watt profile can't quite catch a base M3). The Snapdragon X is, of course, unreleased, so they may get their performance per watt up prior to release.

AMD is closer in performance, but drawing more power. The big AMD laptop chips that perform about 70-75% as well as a M3 Max are only going in gaming laptops and a few mobile workstations, both of which have abysmal battery life.

Im surprised how much of a sustained edge Apple has had - it's just over three years since the first M1 came out, and nobody has come close. No non-Apple laptop has ever caught up to a same-year Max chip in raw CPU performance, no matter how much power-hungrier the other machine is. Other makers HAVE had more GPU power, but only with high-power "barely laptop" GPUs. The M3 Max closes the GPU gap, although it doesn't eliminate it - a mobile RTX 4090 is faster (for 175 watts, it had better be). Nothing fanless has come CLOSE to the base M-series chips in CPU OR GPU power.

 

[leman]

What I find interesting is that Apple's performance IS close to "multiples better". The M3 Max is darned close to twice as fast as Meteor Lake in a lot of tests. Apple is also coming close to Intel with the fanless (8 watt?) base M3 chip (the Intel can draw 28 watts standard and 115 watts peak, in the same range as the M3 Max). The benchmarks I've seen are about 20% faster than a base M2 (which is what it's typically reviewed against), and less than 10% faster than a base M3. We haven't seen the fanless M3 machines yet (both the base MacBook Pro and the iMac have fans, though neither uses them much), but I'd be very surprised NOT to see the M3 in both the MacBook Air and even the iPad Pro (from M-series history). . Apple either gets 90% of the performance for a fraction of the power OR nearly twice the performance for similar power.

The new Qualcomm Snapdragon X is a similar situation - they're performing between a base M3 and a Pro, but drawing up to 80 watts to do that (the 23 watt profile can't quite catch a base M3). The Snapdragon X is, of course, unreleased, so they may get their performance per watt up prior to release.

I think you might be mischaracterizing some of the aspects of these chips. It’s important to look at power consumption in a specific context. M3 CPU draws about 6 watts at peak single-core (that’s 4.05ghz operation) and up to 65-70 watts (the full 12+4 Max). In contrast to Intel Apples peak power consumption is also the sustained power consumption (Intel can go way over the declared sustained). I don’t know the peak multicore consumption of base M3, but I’d put it somewhere around 20 watts.
I haven’t seen precise measures of Meteor Lake power draw, so I can’t comment on that. Qualcomm’s Oryon 23W config is roughly 20% faster than base M3 in multicore GB6, actual power consumption is again unknown.

 

[HawkTheHusky1902]

Just remember competition is a good thing. We got Apple Silicon because of Intel. If competition didn't exist, innovation would be lackluster and companies would just get lazy.

yeah, seems like OP doesn't get that in the original post.

 

[Zest28]

Doesn't looks so bad as it outperforms M2 by 20% in geekbench. And this chip is for thin and light laptops, like the M2 MacBook Air.

And laptops with this new Intel chip seem to last as long as the M2 MacBook Air too. The AMD Ryzen 7 also keeps up with the M2 MacBook Air no problem.

So with the new Intel chip, you can get the same battery life and performance as the M2 MacBook air.

 

[Pressure]

View attachment 2326930

So with the new Intel chip, you can get the same battery life and performance as the M2 MacBook air.

The Zenbook comes with a battery that has over 41% more capacity.

Normalise the scores and account for one of them being completely fanless while the other is not.

The new M3 based 14-inch MacBook Pro comes with a fan and a 72.4WH battery, so that would be the better option for a comparison.

Link to Geekbench comparison.

The Zenbook has a 16-core Intel Core Ultra 7 155H CPU (22-threads) compared to the 8-core Apple M3 SoC.

 

[Zest28]

Man, that Acer laptop is $999 and it has:
- Touch screen
- 16GB RAM
- 512 SSD.

The Asus Zenbook 14 is $1299 and has:
- OLED display
- 32GB RAM
- 1 TB SSD

At the same time, the 15" M2 MBA with 16GB RAM and 512 SSD costs $1700 in comparison. At a 15" M2 MacBook Air (inferior display), 24GB RAM (inferior amounts of RAM) and 1 TB SSD is $2100.

Looks like M2 MacBook Air just got beat hard. The Asus Zenbook is an absolute steal.

 

[Zest28]

The Zenbook comes with a battery that has over 41% more capacity.

Normalise the scores and account for one of them being completely fanless while the other is not.

Who cares. The Asus Zenbook destroys the M2 MacBook Air. It has an OLED display, 32GB RAM and 1TB SSD for only $1299. And you get the same battery life and same performance.

You have to spend $2100 on a M2 MacBook Air that only comes with 24GB RAM and an outdated IPS panel to even try to come close.

 

[Zest28]

I know people on MacRumors have got infinite pockets and don't care about price, but for most people price is a consideration. And the prices on these machines are just insane for what you get.