Jetson MAXN vs MAXN Super: Power, FPS Gains, and When to Use Each

Quick Answer

What Is MAXN Mode?

Every Jetson Orin module ships with a default power mode that caps the SOM's power draw to a fixed wattage. The GPU, DLA, and CPU clocks are limited to stay within that budget. MAXN is the highest-performance nvpmodel preset on supported Jetson platforms and is designed to lift standard power constraints in favor of higher sustained clocks, subject to platform and thermal limits.

With higher-performance power modes enabled, the module can sustain higher clocks, subject to workload, thermal conditions, and platform support. This typically translates to higher inference throughput, but at the cost of higher power draw, more heat, and stricter cooling requirements.

NVIDIA introduced MAXN in JetPack 5.x for the AGX Orin. With JetPack 6.1 and later, MAXN and the newer MAXN Super preset became available on the Orin Nano Super and Orin NX families.

MAXN vs MAXN Super

The naming is easy to overread. Here is the practical difference:

On the Orin Nano Super, NVIDIA's documented nvpmodel presets go from 25W Super directly to MAXN Super — there is no separate MAXN preset on this module (based on JetPack 6.1 documentation; verify with nvpmodel -p --verbose on your specific build). On the Orin NX, MAXN targets roughly 25W uncapped, while MAXN Super targets roughly 40W+ with higher boost clocks.

Which Modules Support What

Based on NVIDIA JetPack documentation and EdgeAIStack engine data. Verify availability with nvpmodel -p --verbose on your specific JetPack version.

Power Specifications by Module

The values below are SOM-only power estimates derived from NVIDIA platform design guides and EdgeAIStack engine measurements. Carrier board, peripherals, NVMe storage, and cameras typically add 5-15W on top. Actual power draw varies by workload intensity, ambient temperature, and carrier board design.

Source: Documented platform behavior (base/peak from NVIDIA design guides) combined with EdgeAIStack measured ranges (typical operating range under inference workloads).

Orin Nano Super

Orin NX 16GB

AGX Orin 64GB

Thor T5000

Estimated FPS Gains

Switching to MAXN does not double throughput. Higher clocks face diminishing returns from memory bandwidth limits, thermal management overhead, and pipeline bottlenecks. The multipliers below are conservative EdgeAIStack planning estimates relative to each platform's default power mode, derived from clock-ratio scaling with a conservative derating factor. Actual gains vary by model architecture, precision, batch size, memory pressure, and thermal headroom.

Source: EdgeAIStack planning estimates based on GPU/DLA clock ratio analysis from JetPack 6.2 release notes and spec sheets. Not direct benchmark measurements.

Illustrative example: a YOLO11n model running at ~180 fps on an Orin NX at 15W might reach roughly 225 fps at MAXN (~1.25x) and ~279 fps at MAXN Super (~1.55x) under favorable thermal conditions. Whether those extra frames matter depends on your stream count and target FPS per camera.

Use the Stream Calculator to model the estimated impact of power mode changes on your camera count and throughput targets.

Thermal and Cooling Requirements

MAXN modes are generally a poor fit for passive or sealed fanless deployments unless the thermal design has been explicitly validated under sustained inference load. The higher sustained power draw requires active heat removal that a heatsink alone typically cannot provide.

In enclosed deployments (IP67 boxes, DIN-rail cabinets), the enclosure needs ventilation or active exhaust. Sealed enclosures running MAXN modes will typically thermal-throttle within minutes under sustained inference load. See the Thermal Design Guide for airflow sizing.

When to Use MAXN / MAXN Super

• You are close to stream or latency limits. The extra throughput can add 1-3 additional camera streams before hitting compute constraints.
• You are already thermally equipped. Active-cooled devkits and open-air rack deployments can typically handle the additional heat.
• You are benchmarking or prototyping. MAXN helps expose the module's practical upper performance range so you can size your production deployment around realistic bounds.
• You are running heavy models (segmentation, multi-model pipelines). The throughput gain is most valuable when the default mode is the bottleneck.

When Not to Use Them

• Fanless or passively cooled deployments. The module will likely thermal-throttle and may perform worse than a capped mode under sustained load.
• PoE-powered compute nodes. 802.3af delivers 12.95W, 802.3at delivers 25.5W at the device. Neither can supply the 30-50W+ that MAXN modes typically draw at the SOM plus carrier board level. (This does not affect PoE camera power budgeting — the concern is whether the edge compute node itself is being powered through a constrained PoE delivery path.)
• Battery-powered or solar setups. The variable power draw under MAXN makes capacity planning unreliable.
• When the default mode already has headroom. If your 4-camera deployment uses 60% of stream capacity at 15W, switching to MAXN adds power draw for no practical benefit.

Common Mistakes

• Enabling MAXN without validating the cooling solution. The module may boost briefly, then throttle to below capped-mode performance once thermal limits are reached. Always run a sustained-load test (10+ minutes) and monitor tegrastats temperatures before committing to MAXN in production.
• Assuming TOPS scales linearly with power. A 2x power increase does not yield 2x throughput. Memory bandwidth, decode pipelines, and framework overhead impose diminishing returns.
• Forgetting carrier board and peripheral overhead. PSU sizing based on SOM-only power specs will undersize the supply. Always account for NVMe, USB devices, and carrier board quiescent draw.
• Using uncapped mode when the current deployment already has headroom. If your stream utilization is below 70%, the capped mode is a better fit — it gives you predictable power and thermal behavior.

How to Enable MAXN Modes

Power modes are managed through nvpmodel. The mode IDs vary by module and JetPack version.

# Check current mode
sudo nvpmodel -q

# List available modes
sudo nvpmodel -p --verbose

# Switch to MAXN (mode ID varies — check output above)
sudo nvpmodel -m <mode_id>

# Apply max clock frequencies
sudo jetson_clocks

# Verify clocks are at maximum
sudo jetson_clocks --show

MAXN Super requires JetPack 6.1 or later. If you are on JetPack 5.x, the mode will not appear in nvpmodel -p. Flash or OTA-update to JetPack 6.1+ first.

The selected power mode is typically persistent across reboots (this is the default nvpmodel behavior, though it may depend on carrier board firmware). To revert, run sudo nvpmodel -m <default_mode_id>.

PSU and Power Delivery Sizing

The power numbers above are SOM-only. A complete system needs headroom for the carrier board, NVMe, USB peripherals, and cameras. A practical planning rule is to size the PSU at roughly 1.5x the expected peak module draw after adding carrier-board and peripheral overhead:

Source: EdgeAIStack planning estimates. System draw includes typical carrier board, NVMe, and peripheral overhead. Verify against your specific carrier board specs before procurement.

Use the Module Power Calculator for a more precise PSU recommendation for your specific configuration, and the Deployment Cost Calculator for full BOM pricing.

Frequently Asked Questions

Bottom Line

MAXN and MAXN Super are the right choice when you have active cooling, mains power, and need to push inference throughput closer to the module's ceiling. In practice, gains are meaningful but workload-dependent; conservative planning assumptions are typically in the ~1.2x to ~1.5x range on well-cooled systems. Size your PSU at roughly 1.5x the expected peak SOM draw (including peripheral overhead), verify your thermal solution under sustained load, and confirm you are on JetPack 6.1+ for MAXN Super support.

If your deployment is fanless, PoE-powered, or battery-constrained, stay on a capped mode. The capped modes exist precisely so you can predict power draw and thermal behavior in constrained environments.