(Corrected we are not sure about the model size) Hi everyone, Architectural Foundations & Design MiniMax M3 is a native multimodal model pre-trained on over 100 trillion tokens of interleaved multimedia data, allowing it to excel at complex tasks like agentic coding (scoring 59.0% on SWE-bench Pr…

awq will not fit in dual spark [image]

It’s a pity that the new model has grown so much in size… I need to save up and purchase two more devices)

Apologies, but this is a topic on MiniMax M3 on Quad Spark no ? So AWQ on 4 x DGX should fit.

This thread was originally created without knowing the model’s size. Since the Minimax m2.5-m2.7 is one of the best models for dual-spark, it makes sense to discuss it here without creating a separate thread.

I understand, I was just surprised that my comment was miss-cast. No intention of harm or otherwise.

experimental MiniMax M3 out now for DGX Spark! (4x Sparks for NVFP4, 2x sparks for NVFP4 REAP50) [image] MiniMax M3 NVFP4 and NVFP4 REAP 50 for 4x & 2x DGX Sparks DGX Spark / GB10 Projects Weights are out and it’s time to code! Experimental MiniMax M3 for DG…

@tenari this might be an interesting one for PrismQuant. NVFP4 is a little too heavy for 2 Sparks but a quant between 3.5 and 4.0 bits/param should be able to make things work. Would you consider giving this a stab? I believe it would be more desirable than a REAP. @dbsci congrats on the quant and …

Yeah worth exploring more. I “ran” to NVFP4 at the moment for compatibility, but given the trouble to run this model on especially 2x sparks, it would be nice to gain a little more breathing room.

I spent a couple days on REAP a month or two ago. The problem is that these models have almost no redundant experts. They’ve already been incredibly optimized, so it becomes a question of what to get rid of using sets of calibration data. I was incredibly unimpressed with REAP on M2.6 to the point I…

I noticed there is unsloth gguf and q3_sx would run fine on 2x sparks. Do we have a good way to cluster Llamacpp?

MiniMax M3 : NVFP4 for Quad DGX Spark

Accelerated Computing DGX Spark / GB10 User Forum DGX Spark / GB10

dbsci June 12, 2026, 4:56pm 51

experimental MiniMax M3 out now for DGX Spark!
(4x Sparks for NVFP4, 2x sparks for NVFP4 REAP50)

Topic		Replies	Views
MiniMax M3 NVFP4 and NVFP4 REAP 50 for 4x & 2x DGX Sparks DGX Spark / GB10 Projects	28	1110	June 14, 2026
MiniMax M2.5 released (not available on HuggingFace as of now) -- is DGX Spark ready? DGX Spark / GB10	92	6518	April 12, 2026
MiniMax M2.7 NFVP4 Recipe & Benchmarks DGX Spark / GB10 llama	123	11370	May 19, 2026
DGX Spark performance DGX Spark / GB10	49	5783	February 13, 2026
MiniMax 2.5 REAP - NVFP4 on single DGX Spark DGX Spark / GB10	25	3153	April 1, 2026
MiniMax-2.5 on DGX Spark (thanks to Unsloth https://unsloth.ai/docs/models/minimax-2.5) DGX Spark / GB10 llama	12	3971	February 20, 2026
Qwen3.5-122B-A10B NVFP4 Quantized for DGX Spark — 234GB → 75GB, Runs on 128GB DGX Spark / GB10 Projects	44	11088	April 9, 2026
What's the best speed we can get with Qwen 3.6 27B without quantizing? DGX Spark / GB10	31	15271	June 13, 2026
Qwen/Qwen3.6-35B-A3B (and FP8) has landed DGX Spark / GB10 agentic-ai	308	26328	June 9, 2026
MiMo-V2.5-NVFP4 on 2x Spark Cluster - Recipe, findings, fixes, benchmarks DGX Spark / GB10	44	2940	June 15, 2026

MiniMax M3 : NVFP4 for Quad DGX Spark

Related topics