Models now include a proper filtering per device type, and consists 16 different recipes! Includes DGX Sparks, RTX 5090, RTX 6000 Pro, and 16-24 GB GPUs. Let me know what you think!

🔥🔥🔥

This is VERY cool. Nicely done!

Thanks!

Added "Last updated"

Thanks looks amazing. Will you ever do an RTX 3090, or better yet, 2x RTX 3090?

Having in the card the latest update date would be useful!

Great idea, will add

Great site, Mia. Well done and very helpful.

Maybe include each model strengths and weakness per usage case.

Yes, planned

👏👏👏

do you have any tips on tweaking these for dual gpu setups?

My setup runs Qwen3.8 27B NVFP4 at 200 tok/s+ on 1x5090 btw.

Great!

need to have up to 8 sparks

Artificial Intelligence score would be another QOL to add

no metal 🥲

How do you feel about these two extra options regarding the licenses? - For personal use only, but offers the fastest performance. - Available for commercial use, but slightly slower.

this is great, setting up second spark today and will be immediately useful

I’ve only been following you for a short while, but I think that’s actually a plus. I’d add a link to the original model’s page and a quick comparison showing what’s different in your version. Personally, I’d also love to see your models available in GGUF or as Docker images.

Almost certainly out of scope, but I'd like to see some benchmark scores. Whilst playing with Qwen-flash-next variants, I realised that some are stronger than others, even if a little slower. At the very least, some indication of capability for the model as a whole would be nice

this is the right way to do local model guides. matching the recipe to the gpu matters more than another leaderboard.

Very useful. If/when i get the second/third and fourth spark, it's because of you and @spark_arena . @NVIDIAAI @NVIDIARTXSpark should know that

we need a comparison feature! and to filter if you want it for coding streams¡, how many, etc to recommend you which one is best :D

What do you recommend for an arc pro B70?

Great design! Maybe some small GPUs in the future? Bonsai 2 in a RTX 3060 is doing stuff.

What would you run with 5 3090s? And also 6?

Any local AI for video/image workflow?

What about Mac Studio?👍🏻

Glad you separate single-stream speed from total throughput. When you’re waiting for one answer, those are very different numbers.

Filter by box. That’s the page I want.