Snapdragon X Elite-powered Microsoft Surface laptop tests show impressive results, with a caveat

midian182

midian182

Posts: 9,857   +125
Staff member
In brief: Microsoft made a lot of bold claims about the Snapdragon X Elite-powered latest Surface laptop, including that it's up to 58% faster than Apple's M3 MacBook Air while offering excellent battery life. Now, a third party has put those claims to the test, and for the most part, they appear accurate. The caveat is that the review was paid for by Microsoft.

Signal 65's Ryan Shrout put the Snapdragon X Elite SoC through a series of tests and benchmarks to see if it lives up to the promises Microsoft made during its recent event introducing the Copilot+ PCs, including the latest Surface line.

The new Surface Laptop 7 with the Snapdragon X Elite SoC was compared against the older Surface Laptop 5 featuring an Alder Lake Core i7-1255U, a Surface Pro 9 with a Microsoft SQ3 chip, MSI's Prestige 16 AI EVO (Meteor Lake Core Ultra 7 155H), and an M3-powered MacBook Air 15.

Starting with thermals under load, the latest Surface laptop did run hotter than the Intel-powered Surface and the MacBook Air, but was cooler than the Prestige. In standard workloads, however, the Snapdragon X Elite laptop fared better. The MacBook Air was the best performer in all the thermal tests.

Microsoft says the new Surface laptops are incredibly quiet. That appears to be the case in the standard workload (Cinebench 2024 single-thread) test, where the X Elite machine was the quietest of all the laptops at 26.3 dB. The multithread test showed it was louder than the MacBook Air and previous-gen Surface.

Battery life, one of the elements of the new Copilot+ PCs that Microsoft says is outstanding, proved to be excellent in the two tests – video playback and Procyon. The new Surface Laptop powered by the Snapdragon X Elite had the longest battery life of any of the systems, though the MacBook Air wasn't tested in the Procyon battery test.

Moving on to performance, the Snapdragon X Elite came out top in multi-threaded performance in Cinebench and Geekbench, while the Apple laptop was slightly ahead in both benchmarks' single-threaded tests.

It should come as little surprise to see the Snapdragon X Elite decimating the other machines in the AI tests. The chip's 45 TOPS of NPU performance is a first, making it twice as fast as the SQ3 laptop and MacBook Air in the Procyon AI Computer Vision test.

Plenty of other tests were carried out, including graphics performance. The X Elite was around twice as fast as the two other Surface laptops, but Meteor Lake with its integrated Arc Graphics proved faster in a couple of 3Dmark tests. The MacBook Air was also faster in the Solar Bay Unlimited and Steel Nomad benchmarks.

Overall, the X Elite performed well, or even excellently, in many tests, though some are arguably not as earth-shattering as Microsoft suggested. It's also worth remembering that this review was commissioned by the Redmond company, so independent reviews will likely paint a more accurate picture.

Permalink to story:

Snapdragon X Elite-powered Microsoft Surface laptop tests show impressive results, with a caveat

 
The noise test results are a bit shady. Since the MBA has no active cooling, where does the sound come from? How is the laptop quieter in the single thread test than the multi threaded test? it only means that the noise baseline was different. And that means that they did not use a controlled environment so the baseline could be different from test to test and thus invalidating the comparison.
 
  • Like
Reactions: Fastturtle, Theinsanegamer and Dreadcthulhu
Taking the paid review into account, we can still safely say it is a *competitive* chip trading blows with top chips. Which is a huge leap from where Snapdragon was even two years ago.

Had they been able to launch this 6 months ago it would be even more impressive but that’s true of all the chips these days. Competition is grand!
 
  • Like
Reactions: kiwigraeme, Theinsanegamer and midian182
Not too bad so far. But I want a thorough test suite of apps like techpowerup use for productivity including scientific workloads and how good emulation is for things like Lightroom, Photoshop, blender. Also what is the timeline for getting major apps ported to ARM. Adobe have not even bothered with Linux binaries for it's software, will they care about ARM?

The good news is AMD is entering the ARM market and it seems like Nvidia too, leaving Intel claiming ARM is in it's rearview mirror along with AMD. AMD's soundwave APU sounds like they are taking this segment very seriously. Hopefully Qualcomm's dominance which has hindered ARM will soon come to an end.
 
  • Like
Reactions: Charles Olson
Any tests using geekbench shows they aren't serious about showing off what the computer can do. Geekbench scores are meaningless as no real world software scales the same as geekbench across different hardware and software. A 20% increase in performance in geek bench might be only a 5-10% difference or nothing at all using software people actually use.

NPU performance is dumb as most discrete video cards will outperform a CPU by a factor of 10 or more. According to NVIDIA their video cards can do "up to" 1300 TOPS compared to 40-50 for a CPU. A 4070 can do nearly 900 TOPS. If you need AI performance, get a device with a discrete video card.

To really give people an idea of the actual performance of this SoC we need to see software people actually use. If you are going to show off gaming performance a selection of games would have been much better to show than a synthetic benchmark.

I'm still unconvinced, but still hopeful, these new SoC's will be much better than Microsoft's previous attempts at a non-X86 based device.
 
  • Like
Reactions: Dreadcthulhu
cristianm said:
The noise test results are a bit shady. Since the MBA has no active cooling, where does the sound come from? How is the laptop quieter in the single thread test than the multi threaded test? it only means that the noise baseline was different. And that means that they did not use a controlled environment so the baseline could be different from test to test and thus invalidating the comparison.
Click to expand...

that 26 decibels is the background noise in a room, silent
 
  • Like
Reactions: thedaimonrings and Charles Olson
Quite nice overall SoC. If smartphones are any indication, this will perform very well over time and across many kinds of workloads.

It is strange they did not put this against the M4, as it is quite a jump over the M3.
 
Bobbydpue said:
Any tests using geekbench shows they aren't serious about showing off what the computer can do. Geekbench scores are meaningless as no real world software scales the same as geekbench across different hardware and software. A 20% increase in performance in geek bench might be only a 5-10% difference or nothing at all using software people actually use.

NPU performance is dumb as most discrete video cards will outperform a CPU by a factor of 10 or more. According to NVIDIA their video cards can do "up to" 1300 TOPS compared to 40-50 for a CPU. A 4070 can do nearly 900 TOPS. If you need AI performance, get a device with a discrete video card.

To really give people an idea of the actual performance of this SoC we need to see software people actually use. If you are going to show off gaming performance a selection of games would have been much better to show than a synthetic benchmark.

I'm still unconvinced, but still hopeful, these new SoC's will be much better than Microsoft's previous attempts at a non-X86 based device.
Click to expand...
I agree for the most part, but at least certain AI-type workloads aren't as efficient on a GPU as the numbers would suggest - that's why my home server has a Google Coral dual-TPU M.2 card in addition to an A2000 12GB GPU.

The two TPUs (Tensor Processing Units) together only do 8 TOPS, but do so at 4W total, and were designed specifically for TensorFlow Lite - they kick butt doing real time local analysis of my security camera footage (facial recognition, vehicle recognition, animal identification, license plate reading, etc).

But with AI models constantly improving, and their computational requirements with them, I still think it makes zero sense to have the NPU be part of the CPU outside of non-upgradeable systems like laptops. They should be optional PCIe add-in cards, so you can upgrade when you feel the need to, without having to replace your CPU. The Coral dual-TPU M.2 only needs a single PCIe 2.0 lane to run, so you should be able to easily run an NPU on a modern x4 slot.

Let laptops have their integrated NPUs, just like they have their integrated graphics - desktops and up should have a dedicated unit.
 
This is second competitor into an ARMs race ( Apple first, with limited access to invade other domains has can't be sold to other Nations eg Linuxia and Winlandia for their ARMies )
QC will sell these to who and whoever.

Thinking 5 years, this pretty amazing chip like the M4 are just going to be meh. Not just performance, but other features and abilities
 
KingAlistair said:
that 26 decibels is the background noise in a room, silent
Click to expand...
For one test it is 26, for the other it is 24. Also, how can a fan less computer with no moving parts (MBA) generate more noise than other computer unless the base line is not controlled and thus not equal between runs
 
You must log in or register to reply here.

Similar threads

D
Here are all the new Copilot+ AI laptops powered by Snapdragon X processors
Replies
0
Views
137
DragonSlayer101
D
Daniel Sims
Qualcomm is offering a NUC-like Snapdragon X mini desktop for Windows developers
Replies
1
Views
101
Bobbydpue
Bobbydpue
D
HP introduces Omni and Elite PCs, retires Pavilion, Spectre, Envy, and Dragonfly brands
Replies
4
Views
109
Theinsanegamer
T

Latest posts

Back