Can I run it?
Can I run GLM 5.3 on RTX 6000 Ada?
Won't fit
Won't fit on RTX 6000 Ada. The smallest quant UD-IQ1_S is 201.8 GB before KV cache, and usable memory is 45.1 GB.
Best quantUD-IQ1_S
Working context—
Max that fits—
Speed—Estimate from memory bandwidth, not a benchmark. Method.
Weights201.8 GB
KV cache0.00 GB
Total0.00 GB
Usable45.1 GB
GLM 5.3 details · RTX 6000 Ada page · How this estimate is made
Run commands
llama.cpp
llama-server -hf unsloth/GLM-5.3-GGUF:UD-IQ1_S -c 2048 -ngl 0
Ollama
ollama run glm-5.3
LM Studio
lms get unsloth/GLM-5.3-GGUF # Then load quant UD-IQ1_S
Assumptions
- Estimate from memory bandwidth, not a benchmark. Decode speed uses 55–75% of peak bandwidth.
- KV cache is FP16. Engine overhead assumes llama.cpp (0.3 GB + 1% of the weights). Ollama is closer to 0.9 GB + 3%, LM Studio closer to 1 GB + 3%.
- 0.5 GB is reserved because the GPU also drives a display, then 5% is left free.
- Usable memory for this profile is 45.1 GB.
Models that fit this machine more comfortably
- Mistral Medium 3.5 128B — Heavily degraded, UD-IQ2_XXS, 14–19 tok/s (estimate)
- Qwen3.5 122B A10B MTP — Tight fit, UD-Q2_K_XL, 30–41 tok/s (estimate)
- Mistral Small 4 119B 2603 — Tight fit, UD-IQ3_S, 29–39 tok/s (estimate)
Updated 2026-10-24. Profile on this page is the device default plus 32 GB of DDR5-5600 system RAM.