Qwen3 Coder 480B-A35B: how much VRAM it needs and which GPU runs it

Qwen3 Coder 480B-A35B in Q4_K_M with an 8k context takes 273.1 GB of GPU memory. The cheapest option right now is 4× RTX PRO 5000 72 GB, from $4.50/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
BF16	894.5	897.5	903.3	926.5	2+ GPUs
FP8	449.0	452.0	457.8	481.1	2+ GPUs
Q8_0	475.3	478.3	484.1	507.3	2+ GPUs
Q6_K	367.0	370.0	375.8	399.1	2+ GPUs
Q5_K_M	317.1	320.1	325.9	349.1	2+ GPUs
Q4_K_M	270.1	273.1	278.9	302.2	B300 288 GB
Q3_K_M	213.5	216.5	222.3	245.5	B300 288 GB
Q2_K	162.7	165.6	171.5	194.7	B200 180 GB

Context cache

How much memory the cache takes per request in FP16. A Q8 cache is about half. 248 KB per token.

8K context: 1.9 GB
32K context: 7.8 GB
128K context: 31.0 GB

Cheapest GPUs that run it

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

4× RTX PRO 5000 72 GB98% usedfrom $4.50/hRent
4× A100 PCIe 80 GB89% usedfrom $4.76/hRent
4× A100 SXM4 80 GB89% usedfrom $6.04/hRent
8× RTX 4090 48 GB75% usedfrom $6.51/hRent
8× A100 PCIe 40 GB90% usedfrom $7.53/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 Qwen3 Coder 480B-A35B need?: In Q4_K_M with an 8k context, 273.1 GB: weights 270.1 GB, context cache 1.9 GB and about 1 GB of runtime memory. In BF16, 897.5 GB.
Does Qwen3 Coder 480B-A35B run on an RTX 4090 with 24 GB?: It does not fit on one RTX 4090 even in Q4_K_M: it needs 273.1 GB. A 288 GB GB card or several RTX 4090s will do.
How much does it cost to run Qwen3 Coder 480B-A35B?: The cheapest fitting server right now is 4× RTX PRO 5000 72 GB, from $4.50/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
BF16	894.5	897.5	903.3	926.5	2+ GPUs
FP8	449.0	452.0	457.8	481.1	2+ GPUs
Q8_0	475.3	478.3	484.1	507.3	2+ GPUs
Q6_K	367.0	370.0	375.8	399.1	2+ GPUs
Q5_K_M	317.1	320.1	325.9	349.1	2+ GPUs
Q4_K_M	270.1	273.1	278.9	302.2	B300 288 GB
Q3_K_M	213.5	216.5	222.3	245.5	B300 288 GB
Q2_K	162.7	165.6	171.5	194.7	B200 180 GB

Cheapest GPUs that run it

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

4× RTX PRO 5000 72 GB98% usedfrom $4.50/hRent

4× A100 PCIe 80 GB89% usedfrom $4.76/hRent

4× A100 SXM4 80 GB89% usedfrom $6.04/hRent

8× RTX 4090 48 GB75% usedfrom $6.51/hRent

8× A100 PCIe 40 GB90% usedfrom $7.53/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 Qwen3 Coder 480B-A35B need?

In Q4_K_M with an 8k context, 273.1 GB: weights 270.1 GB, context cache 1.9 GB and about 1 GB of runtime memory. In BF16, 897.5 GB.

Does Qwen3 Coder 480B-A35B run on an RTX 4090 with 24 GB?

It does not fit on one RTX 4090 even in Q4_K_M: it needs 273.1 GB. A 288 GB GB card or several RTX 4090s will do.

How much does it cost to run Qwen3 Coder 480B-A35B?

The cheapest fitting server right now is 4× RTX PRO 5000 72 GB, from $4.50/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.