Korean
Lenovo Yoga 9n 2-in-1 16 Review
The Lenovo Yoga 9n 2-in-1 16 packs NVIDIA RTX Spark N1X into a convertible: 16-inch OLED touchscreen with tablet/tent stand modes, pen-compatible haptic touchpad, and a 0.69-inch aluminum chassis. Our review unit features the full 20-core N1X, 6,144-core Blackwell GPU, 64GB LPDDR5X and 1TB Gen4 SSD, priced $3,749.99 at Best Buy. An 18-core, 32GB, 1TB variant sells for $3,099.99.
We benchmarked it against three reference devices: the 128GB HP OmniBook Ultra 16, a 16-inch clamshell and our fastest-tested RTX Spark laptop for thermal-performance tradeoff comparison; Lenovo Pro 9n 15, its 18-core sibling; and Razer Blade 16 with RTX 5090 representing high-end x64 laptops. This review covers build, local AI, CPU, memory, gaming, thermals and storage; battery data will arrive once our new test rig is ready.
Temperature defines this Yoga. Under full all-core workloads that push the OmniBook to 98°C, the Yoga’s CPU peaks 71–86°C and sustains 72°C over a 30-minute loop with zero performance drop. Lenovo redesigned the thermal stack with enlarged vents and a raised thermal bump; the convertible hinge creates extra rear airflow space. The tradeoff is lower all-core clock speeds: multi-threaded scores run 4–10% behind the OmniBook, while single-thread performance matches.
Key Takeaways
- ---Its 32GB GPU memory runs Tier-S local AI models at full speed: gpt-oss:20b coding agent hits 58.0 tokens/s, identical to the 128GB OmniBook, and MLPerf results sit within 5%. Tier-M Qwen3.6-35B-A3B requires ~40GB GPU memory and cannot run here.
- ---Coolest RTX Spark laptop we have tested: steady 72°C in Cinebench and sustained looping versus the OmniBook’s 98°C; steady-state performance remains 100% of initial run.
- ---Minor CPU trade for much cooler operation: Cinebench 2026 multi-core score 6,292, 10% below OmniBook and 22% ahead of Razer Blade 16; single-core score 557.0 matches HP.
- ---Creator-focused convertible: 2.8K OLED with 1,100-nit peak brightness, haptic touchpad supporting bundled Yoga Pen Gen 2, rotating soundbar tweeters and a 99.9Wh battery inside a 4.28 lb chassis.
- ---QLC SSD is the main weakness: Gen4 QLC sequential read hits 6,874MB/s but only 68.9K 4K random write IOPS, the lowest of the group, adding ~2 seconds to cold AI model loads.
Build and Design
The Yoga 9n 2-in-1 uses Lenovo’s convertible hinge for tablet, tent and stand modes. Cosmic Blue aluminum covers top and bottom panels; dimensions 14.17 × 9.57 × 0.69 inches, starting weight 4.28 lb. It passes MIL-STD-810H durability testing, and the bottom cover uses 50% recycled aluminum.
It shares rounded edges and port layout with the Pro 9n, with a magnet-secured lid. The full-size keyboard includes a numpad and solid key feel; the large glass Force Pad supports multi-touch and pen input. Wider than the Pro 9n, its folding hinge creates greater clearance between heat vents and the hinge than clamshell designs.
Cooling differentiates it from other RTX Spark laptops. Enlarged bottom vents plus an elevated thermal bump leverage extra rear chassis airflow from the convertible hinge. Under heavy loads, Stockfish ran at 71°C versus 98°C on the OmniBook, at the cost of reduced peak clock speeds.
A large perforated intake grille occupies most of the rear bottom half, with exhaust along the rear edge under the hinge. Two rear-corner fans feed a wide central heatsink assembly. Though visually similar to the Pro 9n cooler, the Yoga uses un-ducted fans while the Pro 9n has partial fan intake covers. A central M.2 2242 SSD sits under the heatsink, with a second empty M.2 2242 slot. The 99.9Wh battery fills the front chassis half, flanked by woofer chambers venting from side front slits.
Unlike competing NVIDIA mobile platforms using one 2280 slot, Lenovo uses dual 2242 M.2 bays for expandability, yet 2242 SSDs typically have smaller capacities and weaker performance than standard 2280 drives.
The 16-inch PureSight Pro OLED touchscreen runs 2880 × 1800 with 30–120Hz variable refresh. HDR peak brightness reaches 1,100 nits, with full sRGB and P3 coverage, Delta E＜1, Dolby Vision and VESA DisplayHDR True Black 1000 certification. Audio features four 2W woofers plus two 2W tweeters mounted on the rotating hinge, directing sound toward the user across all usage modes.
Keyboard keys have 1.5mm travel, backlighting and a Copilot key. The haptic Force Pad doubles as drawing surface for Yoga Pen Gen 2, which supports AES 3.0, 8,192 pressure levels and tilt, charges over USB‑C and magnetically attaches to the lid. A 9.2MP IR camera has electronic shutter paired with three microphones.
Left side: HDMI 2.1, 10Gbps USB‑A, 40Gbps USB4 (PD + DP2.1), 3.5mm audio jack. Right side: power button, UHS-I SD reader, second USB4 port, second USB‑A (always-on). Wireless includes Wi‑Fi 7 and Bluetooth 5.4. The 99.9Wh battery supports Rapid Charge Express, delivering 3 hours usage from a 15-minute charge.
RTX Spark N1X Platform
RTX Spark is the Windows PC family built on the N1X superchip, laptop counterpart to the GB10 data center chip. This Yoga uses the top-bin variant: 20-core Grace CPU (10P+10E, no SMT) paired with 6,144-core Blackwell RTX GPU.
Laptop N1X power ranges 45–80W; OEM cooling tuning impacts performance as much as silicon. Lenovo rates Yoga and HP OmniBook at up to 70W; Pro 9n hits up to 80W, yet all three run at differing clocks and temperatures under identical workloads.
N1X’s LPDDR5X memory is shared by CPU and GPU. Memory partitioning is adjusted via NVIDIA software (not BIOS) directly from Windows desktop. Our 64GB Yoga uses a 32GB GPU / 32GB CPU split, enough for Tier-S local AI models but insufficient for Tier-M requiring ~40GB. Memory capacity remains the primary buying constraint for RTX Spark systems.
This static memory split is a Windows limitation, not NVIDIA’s. Windows lacks full unified memory support like Linux on DGX Spark, forcing pre-set allocation. Microsoft and NVIDIA are actively developing native unified memory support.
It runs Windows 11 26H1 Arm64. x64 software runs through Prism emulation; all open-source benchmarks use native Arm64 builds for fair CPU comparison.
Lenovo Yoga 9n 2-in-1 16 Specifications
| Specification | Yoga 9n 2-in-1 16 (83N7000UUS) |
|---|---|
| Platform |
NVIDIA RTX Spark N1XTested:20C Grace CPU +6144-core Blackwell GPUOptional:18C CPU +5120-core GPU |
| Memory | 64GB soldered LPDDR5X-9400;32GB GPU /32GB system splitAlternative:32GB |
| Storage | 1TB PCIe Gen4 QLC M.2 2242 Sandisk SN5100S |
| OS | Windows11 Home Arm64 (tested) |
| Display |
16-inch 2.8K PureSight Pro OLED touch,30–120Hz VRR1100 nits HDR, 100% sRGB/P3,DeltaE<1Dolby Vision, DisplayHDR True Black 1000 |
| Ports | 2×USB4 40Gbps,2×10Gbps USB-A(1 always-on)HDMI2.1,UHS-I SD,3.5mm audio |
| Wireless | Wi-Fi7,BT5.4 |
| Camera | 9.2MP IR with electronic shutter |
| Audio | 6-speaker system:4×2W woofers,2×2W rotating tweeters,3 mics |
| Input |
Backlit 1.5mm travel keyboard, Copilot keyHaptic Force Pad; Yoga Pen Gen2 included (AES3.0,8192 pressure) |
| Battery | 99.9Wh; Rapid Charge Express |
| Thermals | Enlarged vents + thermal bump;70W TDP |
| Dimensions | 14.17 ×9.57 ×0.69 in |
| Weight | 4.28 lb |
| Build | Cosmic Blue aluminum; MIL-STD-810H certified |
| Pricing | $3,099.99 (18C,32GB,1TB);$3,749.99 (20C,64GB,1TB tested) |
Test Setup & Methodology
Our Yoga test unit: RTX Spark N1X 20-core, 64GB LPDDR5X split 32/32, 1TB Sandisk SN5100S Gen4 SSD, Windows11 Home 26H1 Arm64, Lenovo Performance mode.
All systems ran sr-client 1.2 benchmark suite with factory Windows images and OEM drivers. Unified lab settings: Windows Best performance, plugged in, 50% screen brightness, sleep timers disabled, firewall/UAC prompts off. Results are median values of repeated runs; tests with result spread >2% are re-run.
| System | CPU | Memory | GPU | SSD | OS & Mode |
|---|---|---|---|---|---|
| Yoga 9n 2-in-1 | N1X 20C | 64GB LPDDR5X (32/32) | 6144-core Blackwell,32GB usable | Sandisk SN5100S 1TB Gen4 | Win11 Home 26H1 Arm64, Lenovo Performance |
| OmniBook Ultra16 | N1X 20C | 128GB LPDDR5X (64/64) | 6144-core Blackwell,62.5GB usable | Sandisk SN8050S 2TB Gen5 | Win11 Pro 26H1 Arm64, HP Performance |
| Pro9n 15 | N1X18C | 32GB LPDDR5X (12/20) | 5120-core Blackwell,12GB usable | Samsung 1TB Gen4 | Win11 Pro 26H1 Arm64, Lenovo Performance |
| Razer Blade16 | Ryzen AI9 HX370 | 32GB LPDDR5X-8000 | RTX5090 Laptop,24GB VRAM | Samsung 2TB Gen4 | Win11 Home25H2 x64, Razer Performance |
Two table notations: “not run” = insufficient GPU memory to launch test. N/A = benchmark irrelevant for given hardware.
Local AI Performance
Local AI performance hinges on model fit and inference speed. The Yoga’s 32GB usable GPU memory fully runs Tier-S gpt-oss:20b matching 128GB OmniBook generation speed, yet lacks capacity for Tier-M models the OmniBook can handle.
MLPerf Client v2.0
MLPerf Client tests Llama3.1 8B and Phi-4-mini, measuring first-token latency and token throughput, plus software and data analyst agent workflows. All systems use llama.cpp with CUDA for consistent comparison.
| MLPerf v2.0 | Yoga 9n | OmniBook128GB | Pro9n | Razer Blade16 |
|---|---|---|---|---|
| Llama3.1 8B first token(s) | 0.353 | 0.274 | 0.303 | 0.149 |
| Llama3.1 8B tok/s | 54.4 | 56.5 | 55.1 | 128.8 |
| Phi-4-mini tok/s | 91.8 | 96.8 | 95.2 | 222.7 |
| Data Analyst Agent(s) | 59.7 | 53.0 | 56.2 | 27.8 |
| SWE Agent(s) | 51.6 | 45.3 | 48.0 | 24.6 |
Yoga hits 54.4 tok/s on Llama3.1 8B and 91.8 tok/s on Phi-4-mini, trailing OmniBook by 4–5%. First-token latency gap is larger, pointing to lower CPU/GPU clocks on Yoga. Agent workflows take 13–14% longer than OmniBook; Blade16 completes agent tasks in less than half the Yoga’s time.
Model Loading
Ollama tests cold (cache flushed) and warm model load time. Tier-S: gpt-oss:20b; Tier-M: Qwen3.6-35B-A3B requiring ~40GB GPU memory.
| Model Loading | Yoga 9n | OmniBook128GB | Pro9n | Razer Blade16 |
|---|---|---|---|---|
| gpt-oss:20b cold load(s) | 8.06 | 6.23 | Not run | 8.71 |
| gpt-oss:20b warm load(s) | 3.60 | 3.59 | Not run | 6.93 |
| Cold load throughput(MB/s) | 2480 | 3044 | Not run | 3874 |
| Qwen3.6-35B-A3B | Not run | 9.04 | Not run | Not run |
Warm load speed nearly matches OmniBook. Cold load is slower due to Yoga’s QLC SSD lower read throughput. Blade16 is slower in both modes, as model weights transfer across PCIe to discrete VRAM.
Coding Agent
Codex CLI runs fixed coding tasks; success requires all repository tests to pass. We record runtime, pass count and token generation rate. Concurrent browser + AI testing quantifies workload slowdown.
| Codex CLI | Yoga 9n | OmniBook128GB | Pro9n | Razer Blade16 |
|---|---|---|---|---|
| gpt-oss:20b runtime(s) | 132.5 | 115.3 | Not run | 64.4 |
| Success runs /3 | 2 | 3 | Not run | 2 |
| Generation tok/s | 58.0 | 58.0 | Not run | 137.9 |
| Speedometer score w/agent | 14.1 | 21.2 | Not run | 36.3 |
| Browser slowdown(x) | 2.44 | 1.73 | Not run | 1.06 |
Yoga delivers identical 58 tok/s generation speed for gpt-oss:20b but longer runtime. It passed two out of three test attempts, same as Blade16. Concurrent workloads hit Yoga hardest: browser performance drops sharply because unified memory competes between the AI agent and browser, while discrete GPU on Blade isolates the workloads.
Retrieval-Augmented Generation
RAG embeds 20,963 SQuAD passages and answers 400 queries via Phi-4-mini, measuring ingestion, retrieval speed and answer quality. All devices can run this small model, so comparison focuses purely on speed.
| RAG | Yoga 9n | OmniBook128GB | Pro9n | Razer Blade16 |
|---|---|---|---|---|
| Ingest time(s) | 188.8 | 187.0 | 191.8 | 131.0 |
| Answer generation tok/s | 91.1 | 104.4 | 96.6 | 209.2 |
| F1 Accuracy % | 73.97 | 73.97 | 74.27 | 74.47 |
Yoga nearly matches OmniBook ingestion speed but falls behind on prompt processing and generation. Answer accuracy remains almost identical across RTX Spark laptops.
CPU Performance
Yoga’s CPU matches OmniBook single-thread performance but lags in multi-core tests. Single-thread results sit within ~2% of HP; Yoga achieves top RTX Spark single-thread scores for 7-Zip, PyPerformance go and regex_compile. Multi-threaded workloads trail 4–10% generally, larger gaps on Stockfish and OpenBLAS. It still beats Blade16 on rendering, compression and compilation.
Cinebench 2026 & Sustained Load
| Cinebench 2026 | Yoga 9n | OmniBook128GB | Pro9n | Razer Blade16 |
|---|---|---|---|---|
| Single core | 557.0 | 558.6 | 518.2 | 480.0 |
| Multi core | 6292 | 7004 | 6394 | 5179 |
| Sustained steady/first pass | 1.001 | 0.998 | 1.011 | 1.009 |
Yoga multi-core score is 10% below OmniBook, slightly behind 18-core Pro9n, and 22% above Blade16. Over 30-minute sustained run, Yoga maintains flat performance at 72°C.
Other CPU Benchmarks
7-Zip multi-thread compression is 4% slower than OmniBook. Blender rendering lags HP by ~5%. LLVM compile takes 217.2s, still 34% faster than Blade16. HandBrake x265 encoding is 6% slower than OmniBook. SVT-AV1 encoding stays slow across all RTX Spark Arm64 systems, a limitation of immature Arm code.
Single-thread benchmarks like Stockfish, primesieve and Python favor Blade16’s Zen5 architecture. Prism x64 emulation carries modest overhead; x64 7-Zip reaches 86% of native Arm speed.
GROMACS, OpenBLAS & Memory
GROMACS molecular simulation results are close among RTX Spark laptops. OpenBLAS floating-point workloads show the biggest gaps: Yoga’s N=20000 LU benchmark hits 298.6 GFLOPS, 9% under OmniBook. Yoga completes all OpenBLAS problem sizes including N=40000 without thermal issues and keeps temperatures under 85°C.
STREAM Triad bandwidth reaches 112,067 MB/s, 8% below OmniBook and 2.3× Blade16. Yoga posts the lowest memory latency of tested RTX Spark devices at 141.8ns; inter-core latency is typical for N1X.
Gaming and GPU Encoding
SuperTuxKart runs at 1080p Ultra preset, testing native Arm64 and Prism-emulated x64 builds. At this resolution performance is CPU-bound.
| SuperTuxKart 1080p Ultra | Yoga 9n | OmniBook128GB | Pro9n | Razer Blade16 |
|---|---|---|---|---|
| Native avg FPS | 216.4 | 225.9 | 218.4 | 200.0 |
| Emulated x64 avg FPS | 186.8 | 266.5 | 248.6 | N/A |
| FFmpeg NVENC HEVC 4K60 FPS | 139 | 138 | 139 | 132 |
Yoga native frame rate sits 4% below OmniBook and slightly ahead of Blade16. Uniquely, its emulated x64 FPS is slower than native build, opposite of other RTX Spark laptops and will be re-verified. NVIDIA NVENC hardware encoding performance is consistent across the NVIDIA systems.
Thermals
Yoga is our coolest RTX Spark laptop tested. CPU peaks 71–86°C across benchmarks versus constant 98°C on OmniBook; sustained loop average is 70°C. Windows only reports CPU temperature for RTX Spark chips.
The cooler design trades down all-core clock speeds. During sustained load Yoga holds 3015MHz against OmniBook’s 3239MHz. Yoga and OmniBook share 70W TDP rating; the temperature/clock difference is Lenovo’s tuning choice prioritizing low heat and noise, not power limit. Yoga never throttles, sustaining steady clocks indefinitely at much lower temperatures. SSD peaks at 67°C.
Storage Performance
Fio storage benchmarks compare raw-device and filesystem performance at 25GB working set. Yoga’s BitLocker is disabled. Its Sandisk SN5100S Gen4 QLC SSD delivers sequential read 6,874MB/s and write 5,905MB/s. Random write performance is weak: only 68.9K 4K random write IOPS with high tail latency. QLC storage works fine for daily tasks and AI model loading but represents the biggest hardware limitation of this configuration.
Conclusion
Lenovo Yoga 9n 2-in-1 16 is our coolest-running RTX Spark laptop and the sole convertible model in this group. Its redesigned thermal hardware keeps N1X at 71–86°C under heavy all-core workloads while maintaining stable clocks for 30 minutes. The 2.8K OLED, pen-enabled Force Pad, rotating soundbar and 99.9Wh battery make it the most well-rounded creator-focused RTX Spark laptop.
Cool operation costs modest performance: all-core clocks are ~220MHz lower than OmniBook, multi-thread benchmarks trail 4–10%, GPU workloads 4–18%. Its 32GB GPU memory fully supports Tier-S local AI models but cannot run Tier-M; QLC SSD random write performance is the weakest among our test pool.
All standard RTX Spark platform limitations apply: Windows-on-Arm I/O overhead, slow Arm SVT-AV1 encoding, and shared memory contention that degrades browser performance under concurrent AI workloads. We expect continued software ecosystem improvements as these laptops ship.
For creators needing tablet mode, stylus, OLED and local AI with 20-core CPU, Yoga sacrifices minimal RTX Spark performance while running far cooler. Buyers chasing maximum multi-core throughput or Tier-M AI models should opt for a 128GB RTX Spark configuration.
Leaderboard: Lenovo Yoga 9n 2-in-1 16 claims the Coolest-Running RTX Spark Laptop spot on our local AI laptop leaderboard.
Beijing Qianxing Jietong Technology Co., Ltd.
Sandy Yang/Global Strategy Director
WhatsApp / WeChat: +86 13426366826
Email: yangyd@qianxingdata.com
Website: www.qianxingdata.com/www.storagesserver.com
Business Focus:
ICT Product Distribution/System Integration & Services/Infrastructure Solutions
With 20+ years of IT distribution experience, we partner with leading global brands to deliver reliable products and professional services.
“Using Technology to Build an Intelligent World”Your Trusted ICT Product Service Provider!