GLM-5.3 Flash: how much VRAM it needs and which GPU runs it

GLM-5.3 Flash in UD-Q4_K_XL with an 8k context takes 187.3 GB of GPU memory. The cheapest option right now is 8× Tesla V100 32 GB, from $3.28/h.

Find a server Open in the calculator

VRAM by quantisation

The weight file and the total GPU memory with the context cache. One GPU, one request.

Quantisation	File	8K context	32K context	128K context	Fits on
FP8native	305.8	307.1	307.4	308.4	2+ GPUs
BF16	597.6	598.9	599.2	600.2	2+ GPUs
Q8_0	317.6	318.9	319.1	320.2	2+ GPUs
UD-Q6_K_XL	271.8	273.1	273.4	274.4	B300 288 GB
UD-Q4_K_XL	186.0	187.3	187.6	188.6	B300 288 GB
UD-Q2_K_XL	101.3	102.6	102.8	103.9	H200 NVL 141 GB

Context cache

How much memory the cache takes per request in FP16. A Q8 cache is about half. 11 KB per token.

8K context: 0.27 GB
32K context: 0.53 GB
128K context: 1.6 GB

Cheapest GPUs that run it

GPUs from the catalog the model fits on, by price per hour.

8× Tesla V100 32 GB78% usedfrom $3.28/hRent
4× RTX PRO 5000 72 GB68% usedfrom $4.50/hRent
4× A100 PCIe 80 GB61% usedfrom $4.76/hRent
4× A100 SXM4 80 GB61% usedfrom $6.04/hRent
8× RTX 4090 48 GB52% usedfrom $6.51/hRent

The price is the cheapest GPU of the model × the GPU count. A 2–8 GPU server can cost a little more per GPU.

How to run it on gpu.nz

For an API under load there is the vLLM template with an OpenAI-compatible API.

Questions

How much VRAM does GLM-5.3 Flash need?: In UD-Q4_K_XL with an 8k context, 187.3 GB: weights 186.0 GB, context cache 0.27 GB and about 1 GB of runtime memory. In BF16, 598.9 GB.
Does GLM-5.3 Flash run on an RTX 4090 with 24 GB?: It does not fit on one RTX 4090 even in UD-Q4_K_XL: it needs 187.3 GB. A 288 GB GB card or several RTX 4090s will do.
How much does it cost to run GLM-5.3 Flash?: The cheapest fitting server right now is 8× Tesla V100 32 GB, from $3.28/h. Billing is per minute: stop the server and only its disk is billed.
Which quantisation should I pick?: Q4_K_M is the chat default: close to the original quality at a third of BF16's size. Q5_K_M and Q6_K are more accurate, Q8_0 is almost the original. For vLLM use FP8 or BF16.

File sizes checked against Hugging Face on 2026-10-19. 1 GB here is 1024³ bytes, like GPU memory. Hugging Face lists decimal gigabytes, so its numbers are 7% larger.

VRAM by quantisation

The weight file and the total GPU memory with the context cache. One GPU, one request.

Quantisation	File	8K context	32K context	128K context	Fits on
FP8native	305.8	307.1	307.4	308.4	2+ GPUs
BF16	597.6	598.9	599.2	600.2	2+ GPUs
Q8_0	317.6	318.9	319.1	320.2	2+ GPUs
UD-Q6_K_XL	271.8	273.1	273.4	274.4	B300 288 GB
UD-Q4_K_XL	186.0	187.3	187.6	188.6	B300 288 GB
UD-Q2_K_XL	101.3	102.6	102.8	103.9	H200 NVL 141 GB

Cheapest GPUs that run it

GPUs from the catalog the model fits on, by price per hour.

8× Tesla V100 32 GB78% usedfrom $3.28/hRent

4× RTX PRO 5000 72 GB68% usedfrom $4.50/hRent

4× A100 PCIe 80 GB61% usedfrom $4.76/hRent

4× A100 SXM4 80 GB61% usedfrom $6.04/hRent

8× RTX 4090 48 GB52% usedfrom $6.51/hRent

The price is the cheapest GPU of the model × the GPU count. A 2–8 GPU server can cost a little more per GPU.

Questions

How much VRAM does GLM-5.3 Flash need?

In UD-Q4_K_XL with an 8k context, 187.3 GB: weights 186.0 GB, context cache 0.27 GB and about 1 GB of runtime memory. In BF16, 598.9 GB.

Does GLM-5.3 Flash run on an RTX 4090 with 24 GB?

It does not fit on one RTX 4090 even in UD-Q4_K_XL: it needs 187.3 GB. A 288 GB GB card or several RTX 4090s will do.

How much does it cost to run GLM-5.3 Flash?

The cheapest fitting server right now is 8× Tesla V100 32 GB, from $3.28/h. Billing is per minute: stop the server and only its disk is billed.

Which quantisation should I pick?

Q4_K_M is the chat default: close to the original quality at a third of BF16's size. Q5_K_M and Q6_K are more accurate, Q8_0 is almost the original. For vLLM use FP8 or BF16.