EDGE OSAI HARDWARE COMPARISON
Jetson Thor vs DGX SparkSpecs, memory and LLM performance
Both have 128 GB of LPDDR5X at 273 GB/s. Jetson AGX Thor is an embedded robotics module (40 to 130 W) with about 2 PFLOPS of sparse FP4; DGX Spark is a desktop for AI developers with a faster Grace CPU, 4 TB of storage and networking to link two units.
Memory
128 GB vs 128 GB
Bandwidth
273 GB/s vs 273 GB/s
Power
40 to 130 W vs 240 W power supply
Type
Robotics and edge module vs Desktop AI system
Which should you choose?
CHOOSE JETSON THOR IF
You build robots or edge devices that need real-time sensor processing.
CHOOSE DGX SPARK IF
You develop, fine-tune and run LLMs at your desk.
Still deciding between Jetson Thor and DGX Spark?
Tell us your models, users and data. We will recommend a setup and run it inside your own building with Vanik Edge OS. Free, no obligation.
Jetson Thor vs DGX Spark specs
Manufacturer specifications. Green marks the stronger figure where a higher number is better. Dense compute is shown without sparsity.
| Specification | NVIDIA Jetson AGX Thor | NVIDIA DGX Spark |
|---|---|---|
| Type | Robotics and edge module | Desktop AI system |
| Memory | 128 GB | 128 GB |
| Memory type | LPDDR5X unified memory | LPDDR5X unified memory |
| Memory bandwidth | 273 GB/s | 273 GB/s |
| FP16 / BF16 (dense) | Not published | Not published |
| FP8 (dense) | Not published | Not published |
| FP4 (dense) | 1,035 TFLOPS | 500 TFLOPS |
| Power | 40 to 130 W | 240 W power supply |
| Multi-GPU link | 4 x 25 GbE | ConnectX-7 200 Gb/s: link two units for 256 GB |
| Form factor | Embedded module and developer kit | Palm-sized desktop with 4 TB NVMe |
| Architecture | Blackwell, 2025 | GB10 Grace Blackwell, 2025 |
| Software | CUDA, JetPack, Isaac, Holoscan, vLLM | CUDA, NVIDIA AI software, vLLM, llama.cpp, Ollama |
| Price | US$3,499 developer kit | US$3,999 at launch (4 TB) |
What they mean for running LLMs
For large language models, memory decides which models fit and how many users you can serve, and memory bandwidth decides how fast each user gets tokens. These are planning figures for one GPU or system using a 4-bit model.
| For LLMs | Jetson Thor | DGX Spark |
|---|---|---|
| Usable for models | About 109 GB | About 109 GB |
| Largest dense model at 4-bit (1 user, short context) | About 170B parameters | About 170B parameters |
| Llama 3.3 70B at 4-bit, single-user speed ceiling | About 7 tokens/s | About 7 tokens/s |
Which open-weight models fit
Memory needed for one user with 8K context at the lowest precision available for each model. Open a model for team-size and fine-tuning figures.
| Model (1 user, 8K context, lowest available precision) | Jetson Thor | DGX Spark |
|---|---|---|
| gpt-oss-20b 21B · about 15 GB at MXFP4 | Yes | Yes |
| Gemma 4 26B A4B 25.2B · about 16 GB at 4-bit | Yes | Yes |
| Qwen3.6-27B 27.78B · about 20 GB at 4-bit | Yes | Yes |
| Qwen3.8-27B 27.8B · about 24 GB at NVFP4 | Yes | Yes |
| Gemma 4 31B 30.7B · about 20 GB at 4-bit | Yes | Yes |
| Qwen3.6-35B-A3B 35B · about 24 GB at NVFP4 | Yes | Yes |
| Llama 3.3 70B Instruct 70.6B · about 46 GB at INT4 | Yes | Yes |
| gpt-oss-120b 117B · about 70 GB at MXFP4 | Yes | Yes |
| MiniMax-M2.5 229B · about 135 GB at 4-bit | No | 2 linked units |
| GLM-5.3-Flash 320B · about 192 GB at NVFP4 | No | 2 linked units |
| DeepSeek-V4.1-Flash 552B · about 320 GB at 4-bit | No | No |
| DeepSeek-R1 671B · about 410 GB at 4-bit | No | No |
| DeepSeek-V3.2 685B · about 419 GB at NVFP4 | No | No |
| GLM-5.2 744B · about 455 GB at NVFP4 | No | No |
| GLM-5.3 753B · about 460 GB at NVFP4 | No | No |
| GLM-5.1 754B · about 461 GB at NVFP4 | No | No |
| Kimi K2 1T · about 579 GB at 4-bit | No | No |
| DeepSeek-V4-Pro 1.6T · about 976 GB at FP4 + FP8 | No | No |
| Kimi K3 2.8T · about 1.62 TB at MXFP4 | No | No |
Models in amber need several of these GPUs working as one. Buying the GPUs is the easy part. Vanik Edge OS splits the model across them, keeps it running if a server fails and manages it all from one console. See how Edge OS clusters GPU servers
FROM SIZING TO RUNNING
Sizing the hardware is the easy part.
Getting several GPUs to work as one system for a whole company is where most projects slow down. That is the part Vanik Edge OS does for you.
STEP 01 · THIS TOOL
Size it
Work out the GPU memory your models need and which GPUs or servers can hold them.
STEP 02 · VANIK EDGE OS
Cluster it
Edge OS joins your GPU servers into one cluster, splits a large model across them and keeps it running if a server fails. NVIDIA and AMD servers work side by side.
STEP 03 · VANIK EDGE OS
Run it
Your team gets AI apps with single sign-on and an audit trail, managed from one console, with your data staying inside your building.
Jetson Thor vs DGX Spark questions
What is the difference between the Jetson Thor and the DGX Spark?
Both have 128 GB of LPDDR5X at 273 GB/s. Jetson AGX Thor is an embedded robotics module (40 to 130 W) with about 2 PFLOPS of sparse FP4; DGX Spark is a desktop for AI developers with a faster Grace CPU, 4 TB of storage and networking to link two units.
Jetson Thor vs DGX Spark: which is better for LLM inference?
Both have the same 273 GB/s of memory bandwidth, so they generate tokens at similar speeds when a model fits on both. Choose the Jetson Thor if you build robots or edge devices that need real-time sensor processing. Choose the DGX Spark if you develop, fine-tune and run LLMs at your desk.
Can the Jetson Thor and DGX Spark run Llama 3.3 70B?
Llama 3.3 70B needs about 46 GB of memory at 4-bit for one user with 8K context. The Jetson Thor runs it on one unit at about 7 tokens/s at most for one user, and the DGX Spark runs it on one unit at about 7 tokens/s at most for one user. Speeds are theoretical ceilings from memory bandwidth; real speeds are usually 50 to 80% of that.
How much does the Jetson Thor cost compared with the DGX Spark?
Jetson Thor: US$3,499 developer kit. DGX Spark: US$3,999 at launch (4 TB). Data center GPUs are usually bought as part of 4- or 8-GPU servers, so compare full-system quotes.
How are these numbers calculated?
Memory, bandwidth, compute and power are manufacturer specifications. Model memory needs come from our VRAM calculator: weights plus KV cache plus about 5 to 10% runtime overhead, keeping 10% of each GPU free (15% on GB10 systems, 25% on Mac). The speed ceiling is memory bandwidth divided by the model’s weight size. All figures are planning estimates.
Related comparisons
Specifications from the manufacturer pages for the NVIDIA Jetson Thor and NVIDIA DGX Spark. Speed ceilings are bandwidth divided by model size, before batching; real throughput depends on the inference engine and settings. Last checked November 10, 2026.
Not sure which hardware fits your team?
Vanik Edge OS runs open-weight models on NVIDIA DGX, AMD Instinct and other GPU servers on your premises, and helps you size the hardware first.