Summary: The video compares a first-generation Snapdragon X Elite Dell XPS 13 against an Intel Lunar Lake Core Ultra 7 Dell XPS 13, both in the same chassis. The tests focus on developer-related tasks, including single-core performance (Speedometer 3), multi-core performance (Mandelbrot algorithm), compilation (C# project), AI inferencing (Olama and LM Studio), and real-world battery life.
Overall, the Snapdragon X Elite demonstrated superior single-core performance and initial battery efficiency. However, the Lunar Lake unexpectedly outperformed the X Elite in the multi-core Mandelbrot test and significantly won the comprehensive battery life test. The AI inferencing results were mixed, with the Intel machine showing better performance when GPU offload was enabled in LM Studio. The pricing difference heavily favored the X Elite at the time of the video.
Introduction & Setup:
- Devices: Two Dell XPS 13 laptops.
- One with Snapdragon X Elite (first generation, over one year old).
- One with Intel Lunar Lake Core Ultra 7.
- Chassis: Both laptops share the exact same physical body.
- Purpose: Perform developer-related tests, not physical review details.
- Physical Features:
- No function keys.
- Cramped key layout.
- Invisible trackpad.
- USB-C ports on both sides (a nice perk).
- Desire for more ports.
- System Information (CPU Details):
- Snapdragon X Elite: 12 cores, base speed 3.42 GHz.
- Lunar Lake: 8 cores, base speed 2.2 GHz.
- On paper disadvantage for Lunar Lake.
Performance Tests - Initial Battery & Pricing:
- Battery Drain (Idle Overnight):
- X Elite: 96% remaining.
- Lunar Lake: 86% remaining.
- X Elite showed better initial efficiency.
- Pricing (at time of video):
- X Elite: $749.
- Lunar Lake: $1199.
- Significant price difference attributed to gaming community backlash and x86 compatibility concerns for ARM-based Windows.
Performance Tests - Speedometer 3 (Single-Core):
- Test Purpose: Measures web application responsiveness, sensitive to single-core performance (JavaScript, TypeScript frameworks).
- Results:
- X Elite: 34.5 (consistent).
- Lunar Lake: 18.5 (second worst score ever seen by the presenter).
- Conclusion: X Elite significantly outperformed Lunar Lake in single-core performance, which was expected given the core speeds, but the magnitude of the difference was surprising.
Performance Tests - Mandelbrot Algorithm (Multi-Core, Interpreted):
- Test Purpose: Python algorithm utilizing all available CPU cores, measured by execution time.
- Expectation: X Elite, with more cores, should win.
- First Run Results:
- X Elite: 85 seconds (noisy fan).
- Lunar Lake: 70 seconds (quiet).
- Second Run Results:
- X Elite: 89 seconds.
- Lunar Lake: 59 seconds.
- Conclusion: Unexpectedly, the Lunar Lake machine finished significantly faster despite having fewer cores and running quieter.
Performance Tests - C# Project (Compiled):
- Test Purpose: Compiling a large C# project (100,000+ namespaces/classes), simulating typical developer build tasks.
- Initial Conditions: Both running on battery.
- Results on Battery:
- X Elite: 114 seconds.
- Lunar Lake: 136 seconds.
- Observation: Intel's machine performed well on battery, traditionally a weakness for Intel.
- Post-Compile Battery Remaining:
- X Elite: 82%.
- Lunar Lake: 74%.
- Results Plugged In:
- X Elite: Slightly slower (5-6 seconds slower).
- Lunar Lake: Consistent with battery performance (within margin of error).
- Conclusion: X Elite performed faster, but Lunar Lake's performance consistency whether plugged in or on battery was a notable improvement for Intel.
Performance Tests - AI Inferencing (Olama & LM Studio):
- Olama (CPU-only):
- Test: Generate a 1000-word story with Gemma 3B model.
- Observation: X Elite was significantly faster in filling the screen (noisy fan).
- Prompt eval rate: X Elite: 59 tokens/sec; Lunar Lake: 49 tokens/sec.
- Generation rate: X Elite: 19.42 tokens/sec; Lunar Lake: 17 tokens/sec.
- Llama CPP CPU engine.
- Neither utilized GPUs effectively (CPU processing only).
- LM Studio (GPU Offload):
- Intel: Detects Intel ARC GPU, allows 34/34 GPU offload via Vulcan API (VRAM 9GB).
- X Elite: Does not detect GPU, no offload option (VRAM 0GB shown).
- Test: Generate story with Gemma 3B model.
- Results:
- Intel (with GPU Offload): 25 tokens/sec.
- X Elite (CPU only): 19 tokens/sec.
- Conclusion: GPU offload on Lunar Lake provided a modest but noticeable improvement over CPU-only processing, showcasing Intel's advantage in this specific AI workload due to better software integration.
Performance Tests - Dev Loop Battery Test:
- Test Purpose: Comprehensive real-world developer simulation (writing/building code, Mandelbrot, Notion, to-do list, music, video, full screen brightness).
- Conditions: Both charged to 100%, drained to complete shutdown.
- Observations: X Elite died first, to the presenter's surprise. Lunar Lake lasted significantly longer, even after low-battery warnings.
- Conclusion: Lunar Lake won the overall battery life test by a significant margin. "Intel does it with efficiency here."