Models / GLM-5.3 Flash / VRAM
How much VRAM do you need for GLM-5.3 Flash?
Last updated:
Weight math from the model card, plus every measured RigBase result for GLM-5.3 Flash. Leave extra room for context (KV cache).
Weight size by quant
| Quant | Bits/weight | Weights only |
|---|---|---|
| Q8_0 | 8.50 | 340.0 GB |
| Q6_K | 6.56 | 262.5 GB |
| Q5_K_M | 5.70 | 228.1 GB |
| Q4_K_M | 4.89 | 195.8 GB |
| Q3_K_M | 4.00 | 159.8 GB |
| Q2_K | 3.16 | 126.4 GB |
Bits-per-weight for GGUF rows: llama.cpp quantize README. Parameter count: Z.ai model card: 320B total, 18B active (Hugging Face lists 321B). We advise staying at Q4 or above for chat quality — see Is Q4 good enough?.
Measured speeds for GLM-5.3 Flash
| Hardware | Quant | Engine | tok/s |
|---|---|---|---|
| 4x RTX Pro 6000 (Blackwell) · 384 GB | EXL3 4-bit + DFlash2 | TensorFold, 1 stream | 265.9 ⚡ |
| 2x RTX Pro 6000 (Blackwell) · 192 GB | EXL3 4-bit + DFlash2 | TensorFold, 1 stream | 238.3 ⚡ |
| 2x RTX PRO 6000 Blackwell Max-Q 96 GB · 192 GB | modelopt_mixed (ModelOpt mixed precision), FP8 KV | vLLM (karmic-kraken image), TP 2, MTP 3 | 177.7 ⚡ |
| 4x Nvidia GB10 (unified, DGX Spark) · 512 GB | EXL3 + DFlash2 | TensorFold TP4, 4 nodes, 1 stream | 127 ⚡ |
| 4x Nvidia GB10 (unified, DGX Spark) · 512 GB | NVFP4 | vLLM TP4 (RiNGSiDE), single-stream code | 105.3 |
| 4x Nvidia GB10 (unified, DGX Spark) · 512 GB | EXL3 4-bit + DFlash2 | TensorFold, 4 nodes, 1 stream | 95.2 ⚡ |
| 3x Nvidia GB10 (unified, DGX Spark) · 384 GB | EXL3 4-bit + DFlash2 | TensorFold, 3 nodes, 1 stream | 77.6 ⚡ |
| 2x Nvidia GB10 (unified, DGX Spark) · 256 GB | NVFP4 + DFlash2 | vLLM TP2, 2 nodes, 1 stream | 68.9 ⚡ |
| 4x Nvidia GB10 (unified, DGX Spark) · 512 GB | EXL3 + DFlash2 | TensorFold TP4, 4 nodes, 1 stream | 65 ⚡ |
| 4x Nvidia GB10 (unified, DGX Spark) · 512 GB | NVFP4 | vLLM TP4 (RiNGSiDE), single-stream prose | 61.7 |
| 2x Nvidia GB10 (unified, DGX Spark) · 256 GB | EXL3 4-bit + DFlash2 | TensorFold, 2 nodes, 1 stream | 60.4 ⚡ |
| 2x Nvidia GB10 (unified, DGX Spark) · 256 GB | NVFP4 | vLLM TP2 (RiNGSiDE), single-stream code | 56.5 |
⚡ = speculative decoding (MTP, DFlash…). Full list: best home rig for GLM-5.3 Flash.
Related
All GLM-5.3 Flash builds VRAM + KV calculator By VRAM size GGUF vs AWQ vs FP8 vs NVFP4 How much system RAM?
FAQ
How much VRAM do you need for GLM-5.3 Flash?
Weights alone are about 195.8 GB at Q4_K_M (320B × 4.89 bits ÷ 8). Add memory for the KV cache (context) and the runtime. The smallest-memory measured result in RigBase is 32 GB (RTX 5090) at 30.1 tok/s. Going below Q4 usually hurts quality for chat; see the Q4 guide.
What is the fastest measured speed for GLM-5.3 Flash?
4x Nvidia GB10 (unified, DGX Spark): 105.3 tok/s plain decode (NVFP4, vLLM TP4 (RiNGSiDE), single-stream code).
Does system RAM replace VRAM?
Only for overflow. When the whole model fits in GPU or unified memory, system RAM barely matters for decode speed. See the system-RAM guide.
Referral link: if you sign up with Google and add $10, Runpod gives you and RigBase bonus credit (terms). It never affects our data.