Results

Max Power (AI Workload Optimized)

Maximum rated TDP with the AI inference stack tuned for the platform according to OEM best practices.

Runs in this category. Every chart below is measured from these runs.
Run	TDP (W)	Duration (s)	Samples	Raw logs
AMD Strix Halo (optimized)	120	900	900	`max_power_optimized/amd_strix_halo_optimized`
NVIDIA Jetson Thor (optimized)	130	900	900	`max_power_optimized/jetson_thor_optimized`

Overview

Where each platform sits across the whole workload before drilling into any one resource.

#Platform balance

Six dimensions, each min-max normalized against the strongest platform in this category.

Data table (2 rows)

Normalized 0–1 against the best platform on each axis; higher is better.
Platform	CPU Headroom	CPU Capability	GPU Capability	Memory Capability	Memory Headroom	Clock Speed
AMD Strix Halo (optimized)	1.00	1.00	1.00	1.00	1.00	1.00
NVIDIA Jetson Thor (optimized)	0.62	0.26	0.97	0.99	0.23	0.59

Source: max_power_optimized/amd_strix_halo_optimized, max_power_optimized/jetson_thor_optimized · CSV CSV

#Mean utilization by resource

Mean utilization of each resource across the run, sampled once a second.

Data table (2 rows)

Platform	CPU (%)	GPU (%)	RAM (%)
AMD Strix Halo (optimized)	14.4	94.6	66.7
NVIDIA Jetson Thor (optimized)	47.2	91.6	81.6

Source: max_power_optimized/amd_strix_halo_optimized, max_power_optimized/jetson_thor_optimized · CSV CSV

#CPU headroom, used against available

The faded portion is what is left for the application developer once the navigation and AI workload is running.

Data table (2 rows)

Platform	Used (%)	Available (%)	Total cores	Free core equiv.
AMD Strix Halo (optimized)	14.4	85.6	32	27.4
NVIDIA Jetson Thor (optimized)	47.2	52.8	14	7.4

Source: max_power_optimized/amd_strix_halo_optimized, max_power_optimized/jetson_thor_optimized · CSV CSV

#Mean board power

Measured at the board, averaged across the run.

Data table (2 rows)

Platform	Configured TDP (W)	Mean (W)	p95 (W)	Max (W)
AMD Strix Halo (optimized)	120	116.1	126.1	134.0
NVIDIA Jetson Thor (optimized)	130	83.8	100.5	129.5

Source: max_power_optimized/amd_strix_halo_optimized, max_power_optimized/jetson_thor_optimized · CSV CSV

#Mean CPU and GPU temperature

Steady-state package temperatures under the composed workload.

Data table (2 rows)

Platform	CPU mean (°C)	CPU max (°C)	GPU mean (°C)	GPU max (°C)
AMD Strix Halo (optimized)	83.3	93.8	71.6	84.0
NVIDIA Jetson Thor (optimized)	61.5	67.7	62.4	72.8

Source: max_power_optimized/amd_strix_halo_optimized, max_power_optimized/jetson_thor_optimized · CSV CSV

Application performance

What the robot actually achieved. A platform that looks healthy on utilization counters but misses its control deadlines is not a platform you can ship.

#Completed missions

Navigation goals that reached "Goal succeeded" within the run, counted from the ROS logs.

Data table (2 rows)

Counted from [bt_navigator]: Goal succeeded lines in the ROS logs.
Platform	Missions completed	Run duration (s)
AMD Strix Halo (optimized)	10	900
NVIDIA Jetson Thor (optimized)	10	900

Source: max_power_optimized/amd_strix_halo_optimized, max_power_optimized/jetson_thor_optimized · CSV CSV

#Control loop misses per second

The Nav2 controller server targets 30 Hz; each miss is a control cycle that did not complete on time.

Data table (2 rows)

Platform	Misses per second	Total misses	Target rate (Hz)
AMD Strix Halo (optimized)	0.49	440	30
NVIDIA Jetson Thor (optimized)	1.85	1665	30

Source: max_power_optimized/amd_strix_halo_optimized, max_power_optimized/jetson_thor_optimized · CSV CSV

#Planner cycle time

Time to produce one Hybrid-A* plan with full-footprint SE2 collision checking, derived from the planner server loop-rate warnings.

Data table (2 rows)

Platform	Samples	Mean (ms)	p50 (ms)	p95 (ms)	p99 (ms)
AMD Strix Halo (optimized)	610	129	30	430	1,801
NVIDIA Jetson Thor (optimized)	646	151	40	490	1,633

Source: max_power_optimized/amd_strix_halo_optimized, max_power_optimized/jetson_thor_optimized · CSV CSV

#VLM query outcomes

Scene-understanding queries issued into the navigation behavior tree, and how many of them returned an answer the tree could use.

Data table (2 rows)

Platform	Total	Success	Error	Cancelled
AMD Strix Halo (optimized)	80	76	0	1
NVIDIA Jetson Thor (optimized)	71	67	0	3

Source: max_power_optimized/amd_strix_halo_optimized, max_power_optimized/jetson_thor_optimized · CSV CSV

Supporting charts

VLM query latency

#VLM query latency

Wall-clock time for a successful query, from request to a parsed answer. Only platforms that completed at least one query appear.

Data table (2 rows)

Platform	Successful queries	Mean (s)	p95 (s)
AMD Strix Halo (optimized)	76	9.56	23.50
NVIDIA Jetson Thor (optimized)	67	10.07	24.56

Source: max_power_optimized/amd_strix_halo_optimized, max_power_optimized/jetson_thor_optimized · CSV CSV

CPU

Total load, how it is distributed across cores, and how much clock is behind it. A workload spread evenly across many cores behaves very differently from the same total load piled onto a few.

#CPU utilization over the run

Total CPU utilization across all cores, sampled once a second for the length of the run.

Data table (2 rows)

Platform	Mean (%)	p50 (%)	p95 (%)	Max (%)
AMD Strix Halo (optimized)	14.39	13.80	18.70	21.30
NVIDIA Jetson Thor (optimized)	47.17	47.40	50.20	53.50

Source: max_power_optimized/amd_strix_halo_optimized, max_power_optimized/jetson_thor_optimized · CSV CSV

Supporting charts

Per-core mean utilization

#Per-core mean utilization

One point per physical core. Flat lines mean the workload spread; spikes mean individual cores are pinned and are the ones that will miss deadlines first.

Data table (2 rows)

Platform	Cores	Mean CPU (%)	Busiest core (%)	Cores above 60%
AMD Strix Halo (optimized)	32	14.4	50.3	none
NVIDIA Jetson Thor (optimized)	14	47.2	61.5	3

Source: max_power_optimized/amd_strix_halo_optimized, max_power_optimized/jetson_thor_optimized · CSV CSV

Per-core utilization percentiles

#Per-core utilization percentiles

The distribution of individual core loads. A high p99 with a low p50 is the signature of a few saturated cores inside an otherwise idle package.

Data table (2 rows)

Platform	p50 (%)	p95 (%)	p99 (%)
AMD Strix Halo (optimized)	13.2	30.5	37.2
NVIDIA Jetson Thor (optimized)	42.6	85.9	89.8

Source: max_power_optimized/amd_strix_halo_optimized, max_power_optimized/jetson_thor_optimized · CSV CSV

Per-core utilization distribution

#Per-core utilization distribution

Every per-core sample in the run, bucketed. Mass on the left is idle silicon available to the application; mass on the right is silicon already spoken for.

Data table (2 rows)

Platform	Samples below 5%	Samples above 60%	Total samples
AMD Strix Halo (optimized)	34.4%	2.7%	28800
NVIDIA Jetson Thor (optimized)	0.3%	23.9%	12600

Source: max_power_optimized/amd_strix_halo_optimized, max_power_optimized/jetson_thor_optimized · CSV CSV

CPU clock frequency

#CPU clock frequency

Sustained clock under load. Clock and core count together set how much work a single real-time thread can finish inside its deadline.

Data table (2 rows)

Platform	Mean (MHz)	p50 (MHz)	p95 (MHz)	Max (MHz)
AMD Strix Halo (optimized)	3,192.94	3,410.85	4,193.19	4,714.80
NVIDIA Jetson Thor (optimized)	1,880.34	1,865.55	2,225.66	2,561.10

Source: max_power_optimized/amd_strix_halo_optimized, max_power_optimized/jetson_thor_optimized · CSV CSV

Headroom

What is left over. This is the number that decides whether your perception stack, your fleet client, and your safety monitor also fit on the same computer.

#Absolute compute headroom

Free cores multiplied by sustained clock, so core count and clock speed are counted together.

Data table (2 rows)

Platform	Total (GHz-cores)	Used	Available	Cores	Clock (GHz)
AMD Strix Halo (optimized)	102.2	14.7	87.5	32	3.19
NVIDIA Jetson Thor (optimized)	26.3	12.4	13.9	14	1.88

Source: max_power_optimized/amd_strix_halo_optimized, max_power_optimized/jetson_thor_optimized · CSV CSV

Supporting charts

Free core equivalents

#Free core equivalents

Available CPU expressed as whole cores, which is how you actually budget a new node onto the machine.

Data table (2 rows)

Platform	Total cores	Free core equiv.	Available CPU (%)
AMD Strix Halo (optimized)	32	27.4	85.6
NVIDIA Jetson Thor (optimized)	14	7.4	52.8

Source: max_power_optimized/amd_strix_halo_optimized, max_power_optimized/jetson_thor_optimized · CSV CSV

Single-thread headroom

#Single-thread headroom

Clock available on the least-loaded core, which is the budget for one new real-time control thread that cannot be parallelized.

Data table (2 rows)

Platform	Least-loaded core (%)	Clock (GHz)	Headroom (GHz)
AMD Strix Halo (optimized)	1.0	3.19	3.16
NVIDIA Jetson Thor (optimized)	20.4	1.88	1.50

Source: max_power_optimized/amd_strix_halo_optimized, max_power_optimized/jetson_thor_optimized · CSV CSV

GPU and memory

The VLM is sized to saturate the GPU, so GPU utilization here reads as "kept fed", not as "under strain". Memory headroom decides how large a model you can host.

#GPU utilization over the run

A platform pinned near 100% is fully absorbing the AI workload; a low figure means the GPU is starved because the CPU cannot feed it.

Data table (2 rows)

Platform	Mean (%)	p50 (%)	p95 (%)	Max (%)
AMD Strix Halo (optimized)	94.55	100.00	100.00	100.00
NVIDIA Jetson Thor (optimized)	91.57	98.00	98.00	98.00

Source: max_power_optimized/amd_strix_halo_optimized, max_power_optimized/jetson_thor_optimized · CSV CSV

Supporting charts

GPU effective throughput

#GPU effective throughput

Utilization multiplied by clock, which separates a GPU that is busy at a high clock from one that is busy at a throttled one.

Data table (2 rows)

Platform	Mean (GHz)	p50 (GHz)	p95 (GHz)	Max (GHz)
AMD Strix Halo (optimized)	265.53	285.40	288.90	289.40
NVIDIA Jetson Thor (optimized)	144.20	154.35	154.35	154.35

Source: max_power_optimized/amd_strix_halo_optimized, max_power_optimized/jetson_thor_optimized · CSV CSV

GPU clock frequency

#GPU clock frequency

Sustained GPU clock, the second half of effective throughput.

Data table (2 rows)

Platform	Mean (MHz)	p50 (MHz)	p95 (MHz)	Max (MHz)
AMD Strix Halo (optimized)	2,707.54	2,855.50	2,889.00	2,894.00
NVIDIA Jetson Thor (optimized)	1,519.61	1,575.00	1,575.00	1,575.00

Source: max_power_optimized/amd_strix_halo_optimized, max_power_optimized/jetson_thor_optimized · CSV CSV

Memory headroom

#Memory headroom

System memory used against free. On unified-memory parts the free portion is directly available to the GPU, which is what sets the largest model you can serve.

Data table (2 rows)

Platform	Total (GB)	Used (GB)	Free (GB)	VRAM used (MB)	VRAM free (MB)
AMD Strix Halo (optimized)	123.5	79.8	43.7	301	1,747
NVIDIA Jetson Thor (optimized)	122.8	99.0	23.8	n/a	n/a

Source: max_power_optimized/amd_strix_halo_optimized, max_power_optimized/jetson_thor_optimized · CSV CSV

System memory utilization over the run

#System memory utilization over the run

Memory pressure across the run, including any swap the platform resorted to.

Data table (2 rows)

Platform	Mean (%)	p50 (%)	p95 (%)	Max (%)
AMD Strix Halo (optimized)	66.71	66.70	66.80	66.90
NVIDIA Jetson Thor (optimized)	81.59	81.60	81.70	81.80

Source: max_power_optimized/amd_strix_halo_optimized, max_power_optimized/jetson_thor_optimized · CSV CSV

Power and thermal

Efficiency at the board level, and how much thermal margin is left before the platform starts throttling in a sealed robot enclosure.

#Available compute per watt

Free GHz-cores divided by mean board power.

Data table (2 rows)

Platform	Available (GHz-cores)	Mean power (W)	GHz-cores/W
AMD Strix Halo (optimized)	87.5	116.1	0.753
NVIDIA Jetson Thor (optimized)	13.9	83.8	0.166

Source: max_power_optimized/amd_strix_halo_optimized, max_power_optimized/jetson_thor_optimized · CSV CSV

Supporting charts

GPU utilization per watt

#GPU utilization per watt

GPU-only efficiency, isolating the AI workload from the CPU side of the comparison.

Data table (2 rows)

Platform	Mean GPU (%)	Mean power (W)	GPU %/W
AMD Strix Halo (optimized)	94.6	116.1	0.814
NVIDIA Jetson Thor (optimized)	91.6	83.8	1.093

Source: max_power_optimized/amd_strix_halo_optimized, max_power_optimized/jetson_thor_optimized · CSV CSV

Power against available compute

#Power against available compute

Up and to the left is better: more headroom for fewer watts. Each point is one platform in this power category.

Data table (2 rows)

Platform	Mean power (W)	Available (GHz-cores)
AMD Strix Halo (optimized)	116.1	87.5
NVIDIA Jetson Thor (optimized)	83.8	13.9

Source: max_power_optimized/amd_strix_halo_optimized, max_power_optimized/jetson_thor_optimized · CSV CSV

Board power over the run

#Board power over the run

Instantaneous board draw, showing how each platform tracks its configured TDP.

Data table (2 rows)

Platform	Configured TDP (W)	Mean (W)	p95 (W)	Max (W)
AMD Strix Halo (optimized)	120	116.1	126.1	134.0
NVIDIA Jetson Thor (optimized)	130	83.8	100.5	129.5

Source: max_power_optimized/amd_strix_halo_optimized, max_power_optimized/jetson_thor_optimized · CSV CSV

CPU temperature over the run

#CPU temperature over the run

Package temperature under sustained load.

Data table (2 rows)

Platform	Mean (°C)	p50 (°C)	p95 (°C)	Max (°C)
AMD Strix Halo (optimized)	83.27	83.25	89.75	93.75
NVIDIA Jetson Thor (optimized)	61.51	63.00	66.40	67.70

Source: max_power_optimized/amd_strix_halo_optimized, max_power_optimized/jetson_thor_optimized · CSV CSV

GPU temperature over the run

#GPU temperature over the run

GPU die temperature with the VLM saturating the accelerator.

Data table (2 rows)

Platform	Mean (°C)	p50 (°C)	p95 (°C)	Max (°C)
AMD Strix Halo (optimized)	71.56	72.00	81.00	84.00
NVIDIA Jetson Thor (optimized)	62.41	63.90	67.40	72.80

Source: max_power_optimized/amd_strix_halo_optimized, max_power_optimized/jetson_thor_optimized · CSV CSV

Thermal runway

#Thermal runway

Margin to a 100 °C throttle point. Small margins on a bench become no margin inside a sealed robot enclosure in a warm warehouse.

Data table (2 rows)

Platform	CPU margin (°C)	GPU margin (°C)
AMD Strix Halo (optimized)	16.7	28.4
NVIDIA Jetson Thor (optimized)	38.5	37.6

Source: max_power_optimized/amd_strix_halo_optimized, max_power_optimized/jetson_thor_optimized · CSV CSV

Workload efficiency summary

The same figures the analysis pipeline emits in its comparison report, for every run in this category.

Run	Mean CPU (%)	Available CPU (%)	Cores	Free core equiv.	Mean GPU (%)	Available GPU (%)	Mean RAM (%)	Available RAM (%)
AMD Strix Halo (optimized)	14.4	85.6	32	27.4	94.6	5.5	66.7	33.3
NVIDIA Jetson Thor (optimized)	47.2	52.8	14	7.4	91.6	8.4	81.6	18.4

Scroll the table sideways for every column.

Detailed metrics comparison

Mean / p95 / max for every captured metric, across every run in this category.
Metric	AMD Strix Halo (optimized)			NVIDIA Jetson Thor (optimized)
Metric	Mean	p95	Max	Mean	p95	Max
CPU Total (%)	14.39	18.70	21.30	47.17	50.20	53.50
CPU Frequency (MHz)	3,192.94	4,193.19	4,714.80	1,880.34	2,225.66	2,561.10
Peak Core Utilization (%)	77.69	100.00	100.00	89.65	93.30	100.00
Mean Core Utilization (%)	14.38	18.70	21.50	47.76	50.90	54.20
Active Cores (%)	61.80	90.60	100.00	99.65	100.00	100.00
RAM Usage (%)	66.71	66.80	66.90	81.59	81.70	81.80
RAM Used (MB)	81,724.55	81,825.33	81,975.40	101,367.66	101,546.81	101,685.80
RAM Total (MB)	126,424.40	126,424.40	126,424.40	125,748.60	125,748.60	125,748.60
Swap Usage (%)	0.00	0.00	0.00	0.10	0.10	0.10
GPU Utilization (%)	94.55	100.00	100.00	91.57	98.00	98.00
GPU Clock (MHz)	2,707.54	2,889.00	2,894.00	1,519.61	1,575.00	1,575.00
GPU Memory Clock (MHz)	1,000.00	1,000.00	1,000.00	n/a	n/a	n/a
GPU Effective Throughput (GHz)	265.53	288.90	289.40	144.20	154.35	154.35
GPU Memory Used (MB)	300.94	302.40	302.40	n/a	n/a	n/a
GPU Memory Total (MB)	2,048.00	2,048.00	2,048.00	n/a	n/a	n/a
VCN Utilization (%)	0.00	0.00	0.00	n/a	n/a	n/a
Board Power (W)	n/a	n/a	n/a	83.76	100.48	129.49
Board Power (W)	116.14	126.06	134.04	n/a	n/a	n/a
CPU Temperature (°C)	83.27	89.75	93.75	61.51	66.40	67.70
GPU Temperature (°C)	71.56	81.00	84.00	62.41	67.40	72.80
Disk Read (MB/s)	0.00	0.00	2.70	0.00	0.00	0.70
Disk Write (MB/s)	0.05	0.10	5.00	0.04	0.10	5.70
Disk Usage (%)	32.80	32.80	32.80	27.00	27.00	27.00
Net Recv (MB/s)	66.89	67.20	68.10	66.25	66.91	71.00
Net Sent (MB/s)	5.71	7.90	8.00	5.67	7.90	8.00
Network Errors	0.00	0.00	0.00	0.00	0.00	0.00
Load Average (1m)	4.65	6.49	7.17	10.75	12.38	13.76
Load Average (5m)	3.47	4.87	5.06	9.38	10.95	11.09
Load Average (15m)	1.85	3.11	3.14	7.34	8.82	8.87
Process Count	512.83	534.00	535.00	465.33	470.00	471.00

Download:AMD Strix Halo (optimized) (CSV) · NVIDIA Jetson Thor (optimized) (CSV) · Full dataset (JSON)