😍Excited to see NADDOD join the SONiC Foundation as a General Member! Looking forward to what’s ahead and bringing open, AI-optimized networking solutions to more customers. #SONiC#AIDC
@MrSage@TechMDAI If you’re stopping at 6 Sparks, I’d still stick with the CRS812. I don’t see a better value option so far, even from our own brand.
The whole switch market is facing chip shortages. Stock is tight, and I don’t expect a quick improvement. I’ll keep an eye on it and share any news.
🧐Recently, a European customer asked for my view: Are more companies building local AI clusters to integrate AI into their businesses?
I shared what I’m seeing—from rising DGX Spark demand and chip supply constraints to ROI.
The signals speak for themselves.
@TechMDAI Yes, this switch is indeed a high-performance one! I also agree with Gene that if you're building a DGX Spark cluster, the CRS series is a great fit. However, when it comes to high-performance, cost-effective connectivity, our cables are definitely worth considering!
Choosing a GPU server for AI inference isn’t just about peak TFLOPS.
Memory capacity, bandwidth, latency, concurrency, and GPU interconnects all matter.
The right server starts with your workload. 🚀
Read the full guide: https://t.co/a0qF22l2cj
DGX Spark cluster throwing errors? 3 quick fixes:
Dual Direct: Link 2 nodes via 1 DAC cable.
3-Node Ring: Loop them (1→2→3→1).
Switch Setup: Use breakout cables into a 200G/400G switch for big LLM training.
https://t.co/rSjBNwrf2Q