@MiaAI_lab ↔ @ManpreetBola ↔ @Weeniehuahuaxd ↔ @jvr0x conversation

20 views Aug 08, 2026 ~1 min read

Mia@MiaAI_lab

I need 8 DGX Sparks total to optimize my workflow.

Manpreet Bola@ManpreetBola

asus's GX10 are pcie5 on 4TB, how would you optimize them make it work with nvidia's DGX pcie4 on 4TB.

Mia@MiaAI_lab

for head/worker weights?

Manpreet Bola@ManpreetBola

Yeah exactly- head/worker split across pcie generations bottlenecks the interconnect.

Mia@MiaAI_lab

you could network mount a folder in the 4tb spark and use that in the worker

Manpreet Bola@ManpreetBola

I was thinking of buying GX10 and DGX, then i noticed that difference. Thats why i asked you when i saw post. Lets try that out then

Weenie@Weeniehuahuaxd

There is essentially no difference for the most part. Typical worst case scenario, you're kinda capped at pcie4 speeds. Everything else runs fine.

Javier ⚛ priv/acc@jvr0x

And that only really matters when loading models, once they are loaded it's the same

Mia@MiaAI_lab

Yep

Includes 7-day Premium trial