Platforms · AMD

X100 / Strix Halo

AMD's flagship targeting AI and robotics edge workloads. Strix Halo represents an x86-based alternative to the Jetson ecosystem, offering strong CPU performance with unified memory for LLM/VLM workloads.

16x Zen 5 (32T) @ 5.1 GHz RDNA 3.5 Up to 128 GB LPDDR5X 45-120 W

Max Power at 120 W

CPU available: 81.9%; 26.2 of 32 cores free
Missions completed: 10; in 900 s
Control loop misses: 0.46/s; against a 30 Hz target
VLM queries completed: 34 of 47; scene-understanding queries

CPU

Total load, its distribution across cores, and the clock behind it.

#CPU utilization

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

Data table (4 rows)

Metric	Mean	p50	p95	Max
CPU total (%)	18.13	18.00	21.40	23.30
Peak core (%)	89.35	99.00	100.00	100.00
Mean core (%)	18.13	18.00	21.40	23.40
Active cores (%)	73.57	75.00	90.60	100.00

Source: max_power/amd_strix_halo · CSV

Supporting charts

Per-core utilization over time

#Per-core utilization over time

One row per core. Persistent dark bands are cores that stay pinned for the whole run.

Data table (32 rows)

Core	Mean (%)
0	14.7
1	21.5
2	28.4
3	34.4
4	35.2
5	29.9
6	40.5
7	18.6
8	6.9
9	19.4
10	5.4
11	4.5
12	6.5
13	16.1
14	9.2
15	11.7
16	10.0
17	14.3
18	18.0
19	55.9
20	23.3
21	63.1
22	30.8
23	11.8
24	5.5
25	10.3
26	5.0
27	4.4
28	5.1
29	7.1
30	6.3
31	6.7

Source: max_power/amd_strix_halo · CSV

Per-core utilization distribution

#Per-core utilization distribution

Every per-core sample in the run, bucketed into 5% bands.

Data table (20 rows)

Utilization band (%)	Samples
0–5	5700
5–10	8734
10–15	3390
15–20	2732
20–25	1858
25–30	1179
30–35	902
35–40	684
40–45	561
45–50	510
50–55	778
55–60	300
60–65	247
65–70	146
70–75	138
75–80	97
80–85	78
85–90	71
90–95	73
95–100	622

Source: max_power/amd_strix_halo · CSV

CPU clock frequency

#CPU clock frequency

Sustained CPU clock frequency under load.

Data table (1 rows)

Metric	Mean	p50	p95	Max
CPU frequency (MHz)	2,644.94	2,754.05	3,512.76	4,230.60

Source: max_power/amd_strix_halo · CSV

GPU and memory

Accelerator load from the VLM workload, and the memory it leaves behind.

#GPU utilization

GPU utilization while the VLM issues queries continuously.

Data table (3 rows)

Metric	Mean	p50	p95	Max
GPU utilization (%)	95.13	99.00	100.00	100.00
GPU clock (MHz)	2,756.60	2,874.00	2,900.00	2,900.00
Effective throughput (GHz)	265.66	284.23	289.90	290.00

Source: max_power/amd_strix_halo · CSV

Supporting charts

System memory usage

#System memory usage

System memory in use across the run, as a share of the total fitted.

Data table (3 rows)

Metric	Mean	p50	p95	Max
RAM (%)	47.30	47.90	48.60	49.40
RAM used (MB)	58,632.88	59,358.55	60,230.12	61,304.30
Swap (%)	0.00	0.00	0.00	0.00

Source: max_power/amd_strix_halo · CSV

GPU memory usage

#GPU memory usage

GPU memory in use, as reported by the platform driver.

Data table (2 rows)

Metric	Mean	p50	p95	Max
GPU memory used (MB)	638.58	640.20	645.40	645.40
GPU memory total (MB)	2,048.00	2,048.00	2,048.00	2,048.00

Source: max_power/amd_strix_halo · CSV

Power and thermal

Board draw and the thermal margin left underneath it.

#Board power

Board power draw across the run, against the configured TDP.

Data table (1 rows)

Metric	Mean	p50	p95	Max
Board power (W)	118.24	120.00	124.03	139.00

Source: max_power/amd_strix_halo · CSV

Supporting charts

CPU and GPU temperature

#CPU and GPU temperature

CPU and GPU package temperature under sustained load, against a 100 °C throttle point.

Data table (2 rows)

Metric	Mean	p50	p95	Max
CPU temperature (°C)	77.00	76.50	87.12	96.38
GPU temperature (°C)	72.44	74.00	79.00	83.00

Source: max_power/amd_strix_halo · CSV

Clock against temperature

#Clock against temperature

A clock that falls as temperature climbs is thermal throttling; a flat clock means the platform held its performance for the whole run.

Data table (2 rows)

Metric	Mean	p50	p95	Max
CPU temperature (°C)	77.00	76.50	87.12	96.38
CPU clock (MHz)	2,644.94	2,754.05	3,512.76	4,230.60

Source: max_power/amd_strix_halo · CSV

Compute delivered per watt

#Compute delivered per watt

CPU utilization divided by instantaneous board power, showing how efficiency moves as the workload shifts between planning and perception.

Data table (3 rows)

Metric	Value
Mean available compute per watt (GHz-cores/W)	0.586
Mean GPU utilization per watt (%/W)	0.805
Mean board power (W)	118.2

Source: max_power/amd_strix_halo · CSV

System behaviour

Stability, I/O, and process load, the context around the headline counters.

#CPU against GPU utilization

Each point is one second of the run. A tight cluster means the two resources move together; a wide spread means one is waiting on the other.

Data table (2 rows)

Metric	Mean	p50	p95	Max
CPU total (%)	18.13	18.00	21.40	23.30
GPU utilization (%)	95.13	99.00	100.00	100.00

Source: max_power/amd_strix_halo · CSV

Supporting charts

Rolling stability

#Rolling stability

Standard deviation over a 30-sample window. Spikes are moments the load became bursty rather than steady.

Data table (2 rows)

Series	Window (samples)
CPU total	30
GPU utilization	30

Source: max_power/amd_strix_halo · CSV

Network throughput

#Network throughput

Sensor data arriving over the wired DDS link from the simulation machine. Every platform is offered the same load.

Data table (3 rows)

Metric	Mean	p50	p95	Max
Network received (MB/s)	66.94	67.00	67.40	68.50
Network sent (MB/s)	0.28	0.30	0.40	0.40
Network errors	1.00	1.00	1.00	1.00

Source: max_power/amd_strix_halo · CSV

Disk throughput

#Disk throughput

Disk read and write throughput across the run.

Data table (2 rows)

Metric	Mean	p50	p95	Max
Disk read (MB/s)	0.00	0.00	0.00	0.00
Disk write (MB/s)	0.05	0.00	0.10	4.70

Source: max_power/amd_strix_halo · CSV

Load average

#Load average

Run-queue depth. A load average well above the core count means threads are waiting for a core rather than running on one.

Data table (3 rows)

Metric	Mean	p50	p95	Max
Load average (1m)	5.02	5.15	6.54	7.23
Load average (5m)	3.74	4.41	5.18	5.31
Load average (15m)	2.02	2.19	3.34	3.45

Source: max_power/amd_strix_halo · CSV

Process count

#Process count

Total processes on the machine while the benchmark runs.

Data table (1 rows)

Metric	Mean	p50	p95	Max
Process count	566.85	564.00	582.00	583.00

Source: max_power/amd_strix_halo · CSV

All metrics for Max Power

Mean / p95 / max for every captured metric, across every run in this category.
Metric	AMD Strix Halo
Metric	Mean	p95	Max
CPU Total (%)	18.13	21.40	23.30
CPU Frequency (MHz)	2,644.94	3,512.76	4,230.60
Peak Core Utilization (%)	89.35	100.00	100.00
Mean Core Utilization (%)	18.13	21.40	23.40
Active Cores (%)	73.57	90.60	100.00
RAM Usage (%)	47.30	48.60	49.40
RAM Used (MB)	58,632.88	60,230.12	61,304.30
RAM Total (MB)	126,428.70	126,428.70	126,428.70
Swap Usage (%)	0.00	0.00	0.00
GPU Utilization (%)	95.13	100.00	100.00
GPU Clock (MHz)	2,756.60	2,900.00	2,900.00
GPU Memory Clock (MHz)	1,000.00	1,000.00	1,000.00
GPU Effective Throughput (GHz)	265.66	289.90	290.00
GPU Memory Used (MB)	638.58	645.40	645.40
GPU Memory Total (MB)	2,048.00	2,048.00	2,048.00
VCN Utilization (%)	0.00	0.00	0.00
Board Power (W)	118.24	124.03	139.00
CPU Temperature (°C)	77.00	87.12	96.38
GPU Temperature (°C)	72.44	79.00	83.00
Disk Read (MB/s)	0.00	0.00	0.00
Disk Write (MB/s)	0.05	0.10	4.70
Disk Usage (%)	9.40	9.40	9.40
Net Recv (MB/s)	66.94	67.40	68.50
Net Sent (MB/s)	0.28	0.40	0.40
Network Errors	1.00	1.00	1.00
Load Average (1m)	5.02	6.54	7.23
Load Average (5m)	3.74	5.18	5.31
Load Average (15m)	2.02	3.34	3.45
Process Count	566.85	582.00	583.00

Raw samples: CSV · max_power/amd_strix_halo

Max Power (AI Workload Optimized) at 120 W

CPU available: 85.6%; 27.4 of 32 cores free
Missions completed: 10; in 900 s
Control loop misses: 0.49/s; against a 30 Hz target
VLM queries completed: 76 of 80; scene-understanding queries

CPU

Total load, its distribution across cores, and the clock behind it.

#CPU utilization

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

Data table (4 rows)

Metric	Mean	p50	p95	Max
CPU total (%)	14.39	13.80	18.70	21.30
Peak core (%)	77.69	76.80	100.00	100.00
Mean core (%)	14.38	13.80	18.70	21.50
Active cores (%)	61.80	59.40	90.60	100.00

Source: max_power_optimized/amd_strix_halo_optimized · CSV

Supporting charts

Per-core utilization over time

#Per-core utilization over time

One row per core. Persistent dark bands are cores that stay pinned for the whole run.

Data table (32 rows)

Core	Mean (%)
0	12.7
1	14.8
2	22.2
3	33.7
4	31.9
5	28.1
6	32.1
7	14.8
8	4.6
9	13.7
10	3.6
11	2.9
12	4.6
13	9.8
14	5.8
15	7.4
16	7.8
17	12.7
18	15.0
19	45.0
20	18.0
21	50.3
22	26.8
23	9.3
24	3.6
25	6.3
26	3.2
27	2.9
28	3.4
29	4.7
30	4.1
31	4.7

Source: max_power_optimized/amd_strix_halo_optimized · CSV

Per-core utilization distribution

#Per-core utilization distribution

Every per-core sample in the run, bucketed into 5% bands.

Data table (20 rows)

Utilization band (%)	Samples
0–5	9893
5–10	6903
10–15	3154
15–20	2375
20–25	1511
25–30	969
30–35	703
35–40	610
40–45	488
45–50	450
50–55	747
55–60	214
60–65	135
65–70	87
70–75	77
75–80	58
80–85	61
85–90	30
90–95	31
95–100	304

Source: max_power_optimized/amd_strix_halo_optimized · CSV

CPU clock frequency

#CPU clock frequency

Sustained CPU clock frequency under load.

Data table (1 rows)

Metric	Mean	p50	p95	Max
CPU frequency (MHz)	3,192.94	3,410.85	4,193.19	4,714.80

Source: max_power_optimized/amd_strix_halo_optimized · CSV

GPU and memory

Accelerator load from the VLM workload, and the memory it leaves behind.

#GPU utilization

GPU utilization while the VLM issues queries continuously.

Data table (3 rows)

Metric	Mean	p50	p95	Max
GPU utilization (%)	94.55	100.00	100.00	100.00
GPU clock (MHz)	2,707.54	2,855.50	2,889.00	2,894.00
Effective throughput (GHz)	265.53	285.40	288.90	289.40

Source: max_power_optimized/amd_strix_halo_optimized · CSV

Supporting charts

System memory usage

#System memory usage

System memory in use across the run, as a share of the total fitted.

Data table (3 rows)

Metric	Mean	p50	p95	Max
RAM (%)	66.71	66.70	66.80	66.90
RAM used (MB)	81,724.55	81,715.20	81,825.33	81,975.40
Swap (%)	0.00	0.00	0.00	0.00

Source: max_power_optimized/amd_strix_halo_optimized · CSV

GPU memory usage

#GPU memory usage

GPU memory in use, as reported by the platform driver.

Data table (2 rows)

Metric	Mean	p50	p95	Max
GPU memory used (MB)	300.94	300.40	302.40	302.40
GPU memory total (MB)	2,048.00	2,048.00	2,048.00	2,048.00

Source: max_power_optimized/amd_strix_halo_optimized · CSV

Power and thermal

Board draw and the thermal margin left underneath it.

#Board power

Board power draw across the run, against the configured TDP.

Data table (1 rows)

Metric	Mean	p50	p95	Max
Board power (W)	116.14	119.10	126.06	134.04

Source: max_power_optimized/amd_strix_halo_optimized · CSV

Supporting charts

CPU and GPU temperature

#CPU and GPU temperature

CPU and GPU package temperature under sustained load, against a 100 °C throttle point.

Data table (2 rows)

Metric	Mean	p50	p95	Max
CPU temperature (°C)	83.27	83.25	89.75	93.75
GPU temperature (°C)	71.56	72.00	81.00	84.00

Source: max_power_optimized/amd_strix_halo_optimized · CSV

Clock against temperature

#Clock against temperature

A clock that falls as temperature climbs is thermal throttling; a flat clock means the platform held its performance for the whole run.

Data table (2 rows)

Metric	Mean	p50	p95	Max
CPU temperature (°C)	83.27	83.25	89.75	93.75
CPU clock (MHz)	3,192.94	3,410.85	4,193.19	4,714.80

Source: max_power_optimized/amd_strix_halo_optimized · CSV

Compute delivered per watt

#Compute delivered per watt

CPU utilization divided by instantaneous board power, showing how efficiency moves as the workload shifts between planning and perception.

Data table (3 rows)

Metric	Value
Mean available compute per watt (GHz-cores/W)	0.753
Mean GPU utilization per watt (%/W)	0.814
Mean board power (W)	116.1

Source: max_power_optimized/amd_strix_halo_optimized · CSV

System behaviour

Stability, I/O, and process load, the context around the headline counters.

#CPU against GPU utilization

Each point is one second of the run. A tight cluster means the two resources move together; a wide spread means one is waiting on the other.

Data table (2 rows)

Metric	Mean	p50	p95	Max
CPU total (%)	14.39	13.80	18.70	21.30
GPU utilization (%)	94.55	100.00	100.00	100.00

Source: max_power_optimized/amd_strix_halo_optimized · CSV

Supporting charts

Rolling stability

#Rolling stability

Standard deviation over a 30-sample window. Spikes are moments the load became bursty rather than steady.

Data table (2 rows)

Series	Window (samples)
CPU total	30
GPU utilization	30

Source: max_power_optimized/amd_strix_halo_optimized · CSV

Network throughput

#Network throughput

Sensor data arriving over the wired DDS link from the simulation machine. Every platform is offered the same load.

Data table (3 rows)

Metric	Mean	p50	p95	Max
Network received (MB/s)	66.89	67.00	67.20	68.10
Network sent (MB/s)	5.71	7.80	7.90	8.00
Network errors	0.00	0.00	0.00	0.00

Source: max_power_optimized/amd_strix_halo_optimized · CSV

Disk throughput

#Disk throughput

Disk read and write throughput across the run.

Data table (2 rows)

Metric	Mean	p50	p95	Max
Disk read (MB/s)	0.00	0.00	0.00	2.70
Disk write (MB/s)	0.05	0.00	0.10	5.00

Source: max_power_optimized/amd_strix_halo_optimized · CSV

Load average

#Load average

Run-queue depth. A load average well above the core count means threads are waiting for a core rather than running on one.

Data table (3 rows)

Metric	Mean	p50	p95	Max
Load average (1m)	4.65	4.83	6.49	7.17
Load average (5m)	3.47	4.21	4.87	5.06
Load average (15m)	1.85	2.03	3.11	3.14

Source: max_power_optimized/amd_strix_halo_optimized · CSV

Process count

#Process count

Total processes on the machine while the benchmark runs.

Data table (1 rows)

Metric	Mean	p50	p95	Max
Process count	512.83	506.00	534.00	535.00

Source: max_power_optimized/amd_strix_halo_optimized · CSV

All metrics for Max Power (AI Workload Optimized)

Mean / p95 / max for every captured metric, across every run in this category.
Metric	AMD Strix Halo (optimized)
Metric	Mean	p95	Max
CPU Total (%)	14.39	18.70	21.30
CPU Frequency (MHz)	3,192.94	4,193.19	4,714.80
Peak Core Utilization (%)	77.69	100.00	100.00
Mean Core Utilization (%)	14.38	18.70	21.50
Active Cores (%)	61.80	90.60	100.00
RAM Usage (%)	66.71	66.80	66.90
RAM Used (MB)	81,724.55	81,825.33	81,975.40
RAM Total (MB)	126,424.40	126,424.40	126,424.40
Swap Usage (%)	0.00	0.00	0.00
GPU Utilization (%)	94.55	100.00	100.00
GPU Clock (MHz)	2,707.54	2,889.00	2,894.00
GPU Memory Clock (MHz)	1,000.00	1,000.00	1,000.00
GPU Effective Throughput (GHz)	265.53	288.90	289.40
GPU Memory Used (MB)	300.94	302.40	302.40
GPU Memory Total (MB)	2,048.00	2,048.00	2,048.00
VCN Utilization (%)	0.00	0.00	0.00
Board Power (W)	116.14	126.06	134.04
CPU Temperature (°C)	83.27	89.75	93.75
GPU Temperature (°C)	71.56	81.00	84.00
Disk Read (MB/s)	0.00	0.00	2.70
Disk Write (MB/s)	0.05	0.10	5.00
Disk Usage (%)	32.80	32.80	32.80
Net Recv (MB/s)	66.89	67.20	68.10
Net Sent (MB/s)	5.71	7.90	8.00
Network Errors	0.00	0.00	0.00
Load Average (1m)	4.65	6.49	7.17
Load Average (5m)	3.47	4.87	5.06
Load Average (15m)	1.85	3.11	3.14
Process Count	512.83	534.00	535.00

Raw samples: CSV · max_power_optimized/amd_strix_halo_optimized

Verdict

Production-capable for this workload and the platform with the most CPU left over with 82% of the CPU bandwidth available to application developers. It can absorb work that today is split between a Jetson Orin and a separate x86 computer.

Strong

Highest available CPU capability per watt. Can absorb at least an additional eight cores of load with no change in application performance.
Completed every mission while holding the lowest control-loop miss rate.
Best in class GPU performance inline with NVIDIA Jetson Thor.
Unified memory: free system RAM is directly available to the GPU for larger models.
x86, so existing x86-only robotics software runs without porting.

Weakness

In the balanced-power category it shows more control-loop misses than at max power, because the generic BIOS balanced profile allocates TDP toward the saturated GPU rather than the CPU.
No equivalent of the Jetson software ecosystem (yet).
Thermal design in a sealed enclosure may need closer evaluation.

Specification

The vendor's numbers on the left; what this platform's sensor drivers actually cost it, measured on real hardware, on the right.

Vendor specification
CPU	16x Zen 5 (32T) @ 5.1 GHz
GPU architecture	RDNA 3.5
GPU cores	2560 Shaders (40 CUs)
Memory bandwidth	256 GB/s
NPU / DLA	XDNA 2 (50 TOPS)
RAM	Up to 128 GB LPDDR5X
Power (TDP)	45-120 W
Product page	AMD

Measured sensor-driver load (fraction of one core, per instance)
3D LiDAR	0.160
2D safety LiDAR	0.014
RGBD camera	0.520
Total for the robot's sensor suite	2.07 cores

Setup instructions for this platform are in the platform setup guide.

Results by TDP or Configuration

Each category fixes a TDP power budget or optimized configuration and runs the workload on every platform. Open one to see how this platform behaved under those conditions, and how it compared against the others.

Max Power

Each platform configured to its maximum rated TDP, so the comparison reflects the most compute each board can deliver.

See the results

Max Power (AI Workload Optimized)

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

See the results

Balanced Power

Each platform set to a balanced power profile (except Orin AGX, max TDP 65W).

See the results