AMD Ryzen AI 300 CPU beats Intel Core Ultra 200V CPU in Linux showdown — Strix Point was up to 1.6X faster than Lunar Lake
Linux and Lunar Lake don't play nice.
According to in-depth testing from Phoronix, Intel’s latest Lunar Lake mobile CPUs are not performing well in Linux benchmarks.
The Linux-focused publication tested the Core Ultra 7 256V found inside Asus’s Zebook S14, which doesn’t support Linux. The lack of official support for Linux on Lunar Lake laptops might raise some eyebrows regarding testing. Still, Intel has released Linux updates crafted explicitly for Lunar Lake throughout 2024 and even 2023. Phoronix also tested two Zenbook S16 laptops equipped with Ryzen AI 300 APUs, the AI 9 HX 370 and AI 9 365, and like the Zenbook S 14, they do not officially support Linux either.
In over 300 benchmarks, the 265V only scored a few victories over the Ryzen AI 300 chips, doing well almost solely in encryption, random read, video encoding tests, and Python, Perl, and JSON benchmarks. But when Lunar Lake lost, it lost hard, sometimes to the point where the Ryzen AI 300 APUs racked up double the performance. The average performance for these chips reflects that, as the AI 9 365 was 57% faster.
However, Lunar Lake even struggled against its predecessors; not only did the Meteor Lake-based Core Ultra 7 155H outperform the 265V by 24% on average, but the Alder Lake-powered Core i7-1280P from 2022 barely eked out a win, too. The 265V did at least beat the Core i7-1185G7 from 2020, a Tiger Lake processor.
Phoronix did note that the 265V was at least power efficient and did well in single-threaded benchmarks, but it lost primarily when workloads could utilize more threads. That’s not entirely surprising since Lunar Lake caps out at four E-cores and four P-cores, while both the 155H and 1280P have more E- and P-cores. Intel’s latest CPU is more geared towards lighter and less performance-focused laptops, similar to the M3 in the MacBook Air.
But clearly, something other than Lunar Lake’s architectural design must be at play. In our review of the Zenbook S14 with the Core Ultra 7 258V (which is just a tad faster than the 256V), we found that Lunar Lake was nipping on the heels of the 155H and even the AI HX 370 in multi-threaded benchmarks like Cinebench 2024. Phoronix speculates these performance problems could either be down to the laptop precisely (perhaps a firmware problem) or something to do with how Lunar Lake interacts with Linux.
Some strange things are happening with Lunar Lake in the Zenbook S14. Somehow, the higher-end Core Ultra 9 288V model of the S14 performs worse than the 258V-equipped model, which is incorrect. Intel claims this is because of an issue with the laptop’s current BIOS. However, it’s unclear if this problem is also responsible for the poor performance Phoronix demonstrated in Linux.
Stay On the Cutting Edge: Get the Tom's Hardware Newsletter
Get Tom's Hardware's best news and in-depth reviews, straight to your inbox.
Matthew Connatser is a freelancing writer for Tom's Hardware US. He writes articles about CPUs, GPUs, SSDs, and computers in general.
-
TheSecondPower Phoronix's benchmark suite doesn't seem representative of ordinary use, especially for the lowest-power laptop market. In most of the web benchmarks—which do represent ordinary use—Phoronix's testing placed Lunar Lake as a stellar performer both in overall performance and power consumption. Much of the testing is multithreaded, which is certainly not where the 4/4+4/4 processor will win against the 4/8+8/16 processor. (big cores/big threads+little cores/little threads)Reply -
redgarl
Do I need to refer to this for you? REAL WORLD PERFORMANCE?TheSecondPower said:Phoronix's benchmark suite doesn't seem representative of ordinary use, especially for the lowest-power laptop market. In most of the web benchmarks—which do represent ordinary use—Phoronix's testing placed Lunar Lake as a stellar performer both in overall performance and power consumption. Much of the testing is multithreaded, which is certainly not where the 4/4+4/4 processor will win against the 4/8+8/16 processor. (big cores/big threads+little cores/little threads)
Stop Gaslighting...
Gj5_uc4ScNEView: https://youtu.be/Gj5_uc4ScNE?si=saqUN466-5_FwbQE -
Giroro Ryzen AI 9 HX 370 Is a 38 Watt (Up to 54W) chip designed for mid-to-large High-Performance / Desktop Replacement / Gaming laptops.Reply
Core Ultra 7 256V is a 17W (which includes the RAM) high-efficiency chip designed for Ultra thin and Ultra portable laptops and meant to directly compete with Apple M3 and other ARM/RISC processors.
This is why I complain about how badly Intel the marketing of the Ultra X 2xxV series, or however we are supposed to genericize the absurd number of SKUs they made out of this single chip. Intel should have been selling this product in a way where people are impressed that they were able to achieve reasonably close performance at half the power. But instead we are still talking about how it loses in absolute performance against a totally different market segment where they should not be trying to compete.
Intel is still using the marketing tactics of a stable, growing monopoly in an era where they are bleeding market share and just had to lay off 15% of the company. This is what happens when you fire low level staff instead of the overpriced C-suite execs who got your company into this mess. The results continue to speak for themselves. -
rtoaht
So, if my AI 300 laptop is slower than my friend's Core Ultra 200V laptops and doesn't last as long on battery should I just watch that video and be happy?redgarl said:Do I need to refer to this for you? REAL WORLD PERFORMANCE?
Stop Gaslighting...
Gj5_uc4ScNEView: https://youtu.be/Gj5_uc4ScNE?si=saqUN466-5_FwbQE -
bit_user
Obviously, the multithreaded performance was going to suck. And, about that single-threaded performance...The article said:Phoronix did note that the 265V was at least power efficient and did well in single-threaded benchmarks, but it lost primarily when workloads could utilize more threads.
You're not looking at them critically enough. If you pay attention, those benchmarks not only had the Ryzen AI machines performing abysmally, they also had the Ryzen AI using significantly less power than the Lunar Lake. So, something was obviously wrong.TheSecondPower said:In most of the web benchmarks—which do represent ordinary use—Phoronix's testing placed Lunar Lake as a stellar performer both in overall performance and power consumption.
It turns out that the Ryzen AI CPUs were running the lightly-threaded jobs on the Zen 5C cores, due to a known Linux scheduling bug affecting them. So, they were at a substantial disadvantage and much worse than what we've seen from comparable benchmarks on Windows.
In fact Notebook Check found Ryzen AI beat the Ultra 7 258V on all of the the web benchmarks they ran:
https://www.notebookcheck.net/Intel-Lunar-Lake-CPU-analysis-The-Core-Ultra-7-258V-s-multi-core-performance-is-disappointing-but-its-everyday-efficiency-is-good.893405.0.html
The 288V managed to take some wins on the WebXPRT benchmarks, but that's not the model Phoronix tested (and the 288V has worse power efficiency, not surprisingly). -
bit_user
Developers, developers, developers, developers. Developers! We want a thin & light laptop, but we want fast compiles when we do them.blargh4 said:What is the % of people buying thin+light laptops for heavy multithreaded workloads?
Everyone seems to think you're either doing long-running renders (in which case something like a workstation definitely makes sense), or else you don't care about MT performance. This is a false dichotomy.
Apple seems to think there's a market for this. -
rluker5
So 1%?bit_user said:Developers, developers, developers, developers. Developers! We want a thin & light laptop, but we want fast compiles when we do them.