Discussion Qualcomm deliberately nerfed X Elite efficiency for Geekbench points.

[damola3]

[Moderator Note: Moving post from CPUs to Opinions and Experiences.]

So far, benchmarks have shows that the Snapdragon X Plus is the most efficient SKU of SXE and that performance doesn't reduce much unplugged, however the level of efficiency of the silicon is understated.

Just Josh (unbiased reviewer) was very critical of the chips, however his review conclusions weren't at all representative of the actual silicon. Using his own data (good quality data), he criticises the efficiency of the X Elite chips (compared to Apple/Intel/AMD).

Allow me to explain.

SKU - Single Core/Multi Core - TDP Sustained (Watts)

Omnibook (X-78-100) - 100/796 - 30

Yoga Slim 7x (X-78-100) - 107/998 - 54

Yoga Slim 7 (Ryzen 7840S) - 97/792 - 41

Yoga Slim 7i (Core 7 155H) - 103/741 - 40

M3 MacBook Pro (11 Core) - 139/846 - 23

M3 MacBook Air (8 Core) - 142/572 - 8

I have 5 initial observations:

1. Those single core results aren't sustained. Some machines can consume double the sustained TDP in a short burst to give you the illusion of greater real-world performance. For example, the Omnibook gets that single core score at 56W.
2. The M3 Air is ridiculously low, though one must remember that it has a 60 Hz refresh rate. The MacBook pro and the Windows machines all have 120 Hz adaptive, making a large portion of the deficit in 'efficiency' between the Air and Pro.
3. The Air is 8 core and Pro is 11/12 core (depending on SKU). The X Elite and X Plus are either 12 or 10 core.
4. The Air is passively cooled, while the Pro and Windows machines have fans.
5. You gain only 25.4% of sustained performance for an 80% increase in sustained TDP with the 78-100 variant.

Quite clearly, the X-Elite is a much cheaper MacBook Pro equivalent at half the cost.

According to his data, the M3 MacBook Pro is 25.4% more efficient in Performance Per Watt (PPW), compared to the 78-100 Omnibook, which isn't nearly as big as many other synthetic benchmarks show (especially when they use the Air).

As you can see, the Yoga slim XE consumes 13-14 Watts more power at peak load, making the higher score. It's at 18 points per Watt compared to the 19 of the AMD and Intel offerings. This basically means that it's as efficient at the highest TDPs.

What gets more interesting is how the Omnibook beats the Intel and AMD machines at 3/4 of the battery consumption. Essentially, the difference between 30w and 54w is an 80% gain and we lose 25.4% of performance. This means that scaling up, we gain 1% of performance per 3.18% increase in wattage.

Remember that efficiency is logistic. Say we project this downwards to the MacBook Pro's 23w, we get ~31 PPW compared to the MacBook's 37 PPW. That's ~84% of M3 efficiency. Not half bad!

These are essentially mobile cores, which still outperform x86 when scaled up. I would argue that Ryzen wasn't as efficient as made out (not calling it inefficient), but that Intel were complacent during the Bulldozer era.

In fact Josh criticised the Omnibook performance after this review by saying that this was the most efficienct only because of a firmware error, which limited the TDP.

For my conclusion and TLDR​

1. The X Elite silicon is incredible albeit not Apple level and the actual silicon itself was not overhyped 1 bit.
2. The X Elite chips have had their efficiency deliberately nerfed with less efficient and higher TDPs, in order to beat the M3 in multicore (which nobody needed or expected).
3. Snapdragon has best use case on lower end systems because the chipsets are half the cost of Intel's according to laptop manufacturers (Dell) . Intel only sell low power systems as their most expensive SKUs. AMD don't produce enough mobile chipsets and I'd assume that it's much more expensive.
4. Most Intel chipsets are not low power. That's limited to their more expensive SKUs. The implications being that all of the cheaper and lower core SKUs are going to have the same type of performance for users running non-taxing workloads.

This basically means that we could get fanless Windows machines with MacBook Air esque battery life for half the price. Intel IMO are about to be wiped out for the lower SKUs. For higher end workloads, the X Elite doesn't scale so well.

5) The dual core boost is entirely useless on X Elite outside of gaming. Boost clocks are really for gaming and burst tasks, where you want a small but taxing task done quickly, like preventing a framerate drop. The X Elite with the best sustained performance that I've seen in most benchmarks is the Vivobook, which is the lowest SKU X Elite but with the highest TDP.

6) We willy wave with too many synthetic benchmarks. Nobody cares about your 100 Geekbench points in real world performance. My battery life is nerfed by 50%, so that manufacturers can boast the highest Geekbench score on day 1. I have a 14 inch Galaxy Book Edge (14 inch doesn't have the dreadfully misaligned trackpad) and I simply use it on Quiet mode, meaning that the machine isn't pushed to the extent to which the fans kick in. Hence, I actually preserve my battery life.

7) If I wanted a machine with high performance (including graphics), battery life and compatibility altogether, then I'd wait for Lunar Lake or Strix Point. The Adreno on these machines is quite good, but only has 6 graphics cores and its drivers ATM are a mess. There's also lack of legacy x86 support. It would be more worth it if I had double the memory bandwidth and a beefier GPU.