Two well-known paths and a third
A Jetson module is a complete computer the size of a card: an Arm processor, an NVIDIA GPU and memory shared by both, soldered onto a board that is mounted on a carrier board.
An x86 PC with an RTX graphics card is a computer made of standard components: an Intel or AMD processor, expandable memory and one or more graphics cards in PCIe slots.
There is a third, more recent path: the x86 compact with unified memory. It shares memory between processor and GPU, like a Jetson, but runs the same Windows or Linux as a desktop computer.
Power and size: where Jetson wins
If the product runs on a battery, flies on a drone or works inside a sealed enclosure, power draw decides. A Jetson Orin Nano works at between 7 and 25 W. The Jetson Thor series is configured at between 40 and 130 W.
A single high-end RTX graphics card asks for several hundred watts, and with them a power supply, a chassis and cooling to match. When there is a socket and room, that is not a problem. When there is not, it is everything.
The software decides: Arm or x86
Jetson uses an Arm processor and its own system, JetPack, based on Linux. It does not run Windows. If your application, your camera drivers or a third-party library only exist for Windows or for x86, the choice is already made.
On x86, whatever runs on the computer where you develop runs here too, and that saves weeks of porting. Both worlds share CUDA and the NVIDIA libraries: a model trained on a workstation with RTX can be taken to a Jetson. What changes is everything around the model.
Memory and room to grow
The memory the AI model can use decides which model fits. And the way it is expanded decides how long the design lives.
| Platform | Memory for the model | Can it be expanded? |
|---|---|---|
| Jetson Orin Nano | 4 or 8 GB, shared | No |
| Jetson Orin NX | 8 or 16 GB, shared | No |
| Jetson AGX Orin | 32 or 64 GB, shared | No |
| Jetson Thor | Up to 128 GB, shared | No |
| x86 compact with unified memory | Up to 96 GB of the 128 GB | No: chosen at purchase |
| x86 PC with one RTX PRO | 16 to 96 GB on the card | Yes: a bigger card or one more |
On a Jetson and on a unified-memory compact, the memory is soldered. On a PC, the graphics card is changed or another is added when the model grows, without redesigning the product.
Lifecycle: which date each one has
NVIDIA publishes how long each module will be available: Jetson Orin modules until January 2032, and the Thor series until 2035 and 2036, with a minimum notice of 8 months before the last shipment. That is an advantage for planning. It also warns that the components of a module may change during its life, and in 2026 it brought forward the end of several modules from earlier generations.
An x86 PC has no single date: each part has its own. What sustains it is a platform frozen by the supplier, with a second source for critical parts and notice before any change.
Which one to choose
| If your product… | It fits |
|---|---|
| Runs on a battery or in a small sealed enclosure | Jetson |
| Takes many cameras over MIPI CSI | Jetson, with its carrier board |
| Uses Windows or software that only exists for x86 | x86 PC |
| Needs several graphics cards or cards of 48 GB or more | x86 PC with RTX PRO |
| Runs large language models on little power | x86 compact with unified memory |
| Has to grow without redesigning the electronics | x86 PC: the graphics card is changed |
We build the x86 side: compacts, workstations and servers. If your product is one of those that call for a Jetson, we will tell you before proposing anything.
The x86 compact, with prices
The integrated GPU in these compacts is from AMD: it runs language and vision models with the usual engines, but code written only for CUDA needs an NVIDIA card. That is why any of the three can be fitted with an RTX PRO, in a chassis of about 10 litres.
| Machine | Memory | Price |
|---|---|---|
| Compact AI 32 | 32 GB unified | €2,065 |
| Compact AI 64 | 64 GB unified | €2,940 |
| Compact AI 128 | 128 GB unified | €4,820 |
| With NVIDIA RTX PRO 2000 Blackwell | 16 GB more, with CUDA | + €2,005 |
Related hardware and guides
Frequently asked questions
Is a PC with RTX more powerful than a Jetson?
In raw power, yes: a desktop graphics card has many more watts available. Per watt consumed, Jetson is more efficient. The useful question is how much power your model needs and how much energy you have.
Can I develop on a PC and deploy on Jetson?
Yes, that is the usual way: you train on a workstation with RTX and optimise the model for the module. You have to allow for the work of porting the rest of the software to Arm and JetPack.
Does Jetson run Windows?
No. It uses JetPack, the NVIDIA system based on Linux. If your product needs Windows, the platform is x86.
What if I need low power and x86 as well?
That is the gap the unified-memory compact fills: a low-power x86 processor, up to 128 GB of shared memory and desktop Windows or Linux. If you also need CUDA, an RTX PRO is added.
Tell us about your product
Tell us what it does, where it will work and how many units a year. We will propose the machine, the lead time and the price in writing within 12 working hours.