forum threadphoinix | Staff posted Nov 26, 2025 08:10 AM
Item 1 of 1
forum threadphoinix | Staff posted Nov 26, 2025 08:10 AM
$2799: NVIDIA Jetson AGX Thor Developer Kit at Amazon
$2,799
$3,499
20% offAmazon
Get Deal at AmazonGood Deal
Bad Deal
Save
Share
Leave a Comment
5 Comments
Sign up for a Slickdeals account to remove this ad.
1. Spark is newer and a full blown desktop replacement. Lack of jetpack means it has a great many more libraries that don't work on the spark. There are workarounds and code compiles possible, but you're treading new ground and unfortunately you will have to do customization for a variety of use cases. My case I was working primarily with Audio LLM's and Audio processing libraries (Torch, Piper, Kokoro, espeak, etc). All of this stuff just works on the Jetson Thor. Spark it does not.
2. LLM performance is a mixed bag on the thor, it does do fairly well but there is a speed difference I find due to the tensor core difference I believe. As previously noted by DeathSpiel, the Thor is intended for dev kits processing images and robotics use cases. The majority of the task I threw at it were LLM, VLM and Audio based. I can say it had no problems so far that I could find truly pulling it off. I only did Ollama testing though.
3. It doesn't support TensorRT that I could find (someone correct me if I'm wrong here), Ollama and PyTorch were my only 2 sets that worked.
4. The hardware quality of the Thor is a mixed bag. I found a brand new unit had not had the fan attached and my teammate who initially worked with it didn't realize. I discovered fan problems are apparently quite common on the Jetson devices, which is kinda awful for a $3k product... but otherwise the hardware looked solid.
5. You can and will find that at the moment the Jetpack capabilities make this a more useable platform for experiments in my experience. Despite the differences in the processing speeds, the platform does have immediately better compatibility on this day... that likely will change over time, but I was surprised that the DGX Spark's were so new that they suffered from the old "Linux libraries need to catch up" scenario.
Hopefully that helped. Would I buy one? Given my experience, I 'd probably do it as it is a fairly versatile platform, but the memory footprint (unified memory) is not as usable as some would believe. In my mind I'd probably take this at the price point over a Spark right now based on my experience. YMMV, so take this for what its worth.
Leave a Comment