You could be forgiven for being a touch confused about the performance Intel’s Lakefield SoC offers, if you’ve kept an eye on headlines the past few days. After the first review data popped up online, different publications have either praised the results or indicated they should be cause for alarm at Microsoft and Samsung HQ. So which is it?
Kind of both, depending on how you think about the chip. Let’s look at Notebookcheck’s data.
What’s Lakefield’s Competition?
The above question isn’t rhetorical. Lakefield — a 7W CPU with four Tremont CPU cores (no HT) and a single Sunny Cove CPU core — is no competition for a full-size Core processor, which is why it doesn’t shine in comparisons against those devices. Its single-thread CB15 performance lags the Core m3-8100y, 88.3 points to 96.2. It does better in multi-core, with a score of 255 versus 162. The m3-8100Y is arguably the best x86 comparison point for the chip because the Core M CPU is a 2C/4T product with a 5W TDP and a 1.1GHz base / 3.4GHz boost clock speed.
Data by Notebookcheck
This pattern — weak single-thread, combined with better multi-thread relative to Core M — repeats in both Cinebench (R10, R15, R20) and in Geekbench 5.1. The m3-8100Y wins every single-threaded benchmark test, sometimes by a larger margin than others, but it loses every multi-threaded test, typically by a larger gap than it wins the single-thread.
This pattern continues in the browser-based benchmarks. The Galaxy Book S NotebookCheck is benchmarking is 1.2x faster than the Core m3-8100Y in JetStream 1.1. It’s 6 percent slower in WebXPRT 3, but the m3-8100Y takes 36 percent longer to complete Mozilla Kraken 1.1 than the Lakefield CPU does.
Data by Notebookcheck. Smaller is better.
Techradar argues that it can hear Qualcomm grinning “from all the way over here,” but it’s not clear what the company would be grinning about. Currently, Windows on ARM is limited to 32-bit emulation for x86 compatibility, and emulation performance is hit-or-miss. Geekbench suggests that Lakefield offers better ST performance than the 8cx (797 points to 700) but worse multi-threading (1441 versus 2760). Qualcomm will likely surpass these targets with its next-generation CPUs, but Intel has an edge for now. THG’s writeup is more positive.
The 1.8x scaling in Geekbench 5.2 is so poor for Lakefield, it’s worth asking if there’s a breakdown somewhere in how the application is treating the CPU. I’m not saying that’s the case, but it’s weaker than what we’d expect. The 8cx shows a scaling factor of 3.46x, which might be explained by TDP differences. Either way, it’s a large gap — easily the largest in our tests.
Lakefield is faster than the 8cx in every browser-based benchmark that NotebookCheck ran on both platforms.
The reason why Lakefield might not look so great, especially when compared with other systems in this manner, is because low-power devices really don’t benefit when benchmarked directly against their bigger brothers. The best-case scenario is that the ultra-low-power product delivers some really killer battery life and efficiency metrics and, in exchange, gets nuked from orbit in every single benchmark. Most reviewers run a lot more benchmarks than they do battery life tests.
Because manufacturers tend to put a high premium on cutting-edge hardware with good battery life, devices equipped with power-sipping processors aren’t necessarily less expensive than their meatier alternatives. There’s a natural, understood difference between an iPhone and an iPad that makes it easy for people to grasp why one is slower than the other: The iPhone is smaller. Ultra-thin and light laptops are still lighter and thinner than regular systems, but the difference is less dramatic.
Notebookcheck hasn’t released its full data set, and without battery life, it’s impossible to evaluate how Lakefield performs in the first place. What seems likely, based on the data we’ve got in the review already, is that the chip is designed to improve multi-threading performance over Core M, without compromising too badly on single-thread. It beats the 8cx in browser tests and Geekbench single-thread and loses in Geekbench multi-threaded.
Battery life should be at least as good as Core M, though not as good as the 8cx. Whether the tradeoff will be worth it to go x86 versus ARM on the Galaxy Book S isn’t clear right now. The performance data points towards x86, but the battery life will likely point back towards ARM. We’ll just have to see how big each gap is once a larger suite of tests can be considered.
I’ve focused on CPUs because that’s the more interesting discussion. Graphics-wise, Lakefield leads against its Core M and Snapdragon competitors, but not by a large-enough gap to make this 7W chip any kind of gaming device. So while gaming and GPU performance are up, and up markedly, you won’t be using Lakefield to handle AAA gaming.
Lakefield looks faster than Core M in multi-threading, at least for now. Will it be sold at price points that help these devices actually move in volume? Probably not. But we’ll reserve judgment until we see more details on battery life and price, as more devices come to market. Performance-wise, Lakefield seems to slip ahead of Intel’s previous ultra-low-power platform, while being generally faster than the ARM Snapdragon competitor outside of Geekbench. But until we know more about battery life, we really can’t judge whether the performance/power tradeoff between Lakefield and big-core Intel CPUs is worth it.
Now Read:
- Intel Unveils Lakefield CPU Specifications: Up to 3GHz, 64 EUs, and 7W TDP
- Intel Shares New Data on Lakefield’s Low-Power Tremont Microarchitecture
- Intel Uses New Foveros 3D Chip-Stacking to Build Core, Atom on Same Silicon
No comments:
Post a Comment