Platforms

Platform comparison

What each platform is on paper, what it was actually configured to for these runs, and how it performed. Specifications are the vendors' own; everything measured is from the runs in this repository.

Silicon and memory

Vendor specifications for the three platforms evaluated.

FeatureX100 / Strix HaloJetson ThorJetson AGX Orin
CPU16x Zen 5 (32T) @ 5.1 GHz14x Arm Neoverse V3AE @ 2.6 GHz12x Arm Cortex-A78AE @ 2.2 GHz
GPU architectureRDNA 3.5BlackwellAmpere
GPU cores2560 Shaders (40 CUs)2560 CUDA + 96 Tensor2048 CUDA + 64 Tensor
NPU / DLAXDNA 2 (50 TOPS)None2x NVDLA 2.0
RAMUp to 128 GB LPDDR5X128 GB LPDDR5X64 GB LPDDR5
Memory bandwidth256 GB/s273 GB/s204.8 GB/s
Power (TDP)45-120 W40-130 W15-60 W
VendorAMDNVIDIANVIDIA

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.

Full profile →

Jetson Thor

NVIDIA's next-generation embedded AI module built on the Blackwell GPU architecture. Thor is designed as the next platform for physical AI and advanced robotics applications and enables edge AI such as LLM/VLM and foundation models.

Full profile →

Jetson AGX Orin

NVIDIA's established embedded AI platform widely adopted in robotics which require edge AI such as detection, segmentation, and reinforcement learning. The Orin is the current workhorse of many production robotics deployments.

Full profile →

As tested

The tested TDP and other key configurations.

ConfigurationX100 / Strix HaloJetson ThorJetson AGX Orin
TDP, Max Power (W)12013065
TDP, Max Power (AI Workload Optimized) (W)120130n/a
TDP, Balanced Power (W)8590 / 7065
Software stackUbuntu, ROCmJetPack, CUDAJetPack 6.2.2, CUDA 12.6
ROS distributionROS 2 JazzyROS 2 JazzyROS 2 Jazzy
DDSCycloneDDSCycloneDDSCycloneDDS
Run length900 s at 1 Hz sampling900 s at 1 Hz sampling900 s at 1 Hz sampling
Nav2 configurationnav2_params.yamlnav2_params.yamlnav2_params.yaml