How much VRAM does gpt-oss-120b need?

234 GB for its weights at 16-bit, 117 GB at 8-bit and 65.7 GB at 4-bit, plus the KV cache: 36.9 KB for each token, 1.21 GB for a conversation of 32K tokens and 4.84 GB at its longest context of 131,072 tokens. The server also keeps about 8% of each GPU free.

Can gpt-oss-120b run on 1 GPU?

Yes, at 4-bit: the smallest single GPU here that holds it with a 32K conversation is the A100 80GB.

How many H100s does gpt-oss-120b need?

With 1 conversation of 32K tokens, 4 H100s at 16-bit and 1 H100 at 4-bit. Each H100 has 80 GB, of which vLLM uses 92% by default.

Can I run gpt-oss-120b on an RTX 5090?

The RTX 5090 has 32 GB. With a 32K conversation, gpt-oss-120b needs 4 GeForce RTX 5090s at 4-bit and 8 GeForce RTX 5090s at 16-bit.

How fast does gpt-oss-120b run?

At 4-bit on 1 H100, memory lets 1 conversation get at most 821 tokens a second. Real servers reach part of that ceiling, and each conversation gets less when many run at once.

How much GPU memory does gpt-oss-120b need?

gpt-oss-120b by OpenAI has 117B parameters, 5.1B of them active for each token. Its weights take 234 GB at 16-bit, 117 GB at 8-bit and 65.7 GB at 4-bit; released in MXFP4, its files are 65.2 GB. Each token of a conversation adds 36.9 KB of KV cache, so a 32K conversation holds 1.21 GB.

Size gpt-oss-120b for your setup Questions

OpenAI · released in MXFP4

gpt-oss-120b at a glance

Parameters: 117B; 5.1B active for each token
As released: 65.2 GB; Its MXFP4 files
At 16-bit: 234 GB
At 4-bit: 65.7 GB; NVFP4
KV cache a token: 36.9 KB; At 16-bit
Longest context: 131,072; Tokens

GPUs needed at each number format

For 1 conversation of 32,768 tokens with a 16-bit KV cache, counting 92% of each GPU's memory as vLLM does. Under each count, the most tokens a second memory allows.

GPU	Memory	16-bit	8-bit	4-bit
L4	24 GB	16 GPUs210 tokens/s	8 GPUs380 tokens/s	4 GPUs294 tokens/s
GeForce RTX 5090	32 GB	8 GPUs1,256 tokens/s	8 GPUs2,271 tokens/s	4 GPUs1,756 tokens/s
L40S	48 GB	8 GPUs606 tokens/s	4 GPUs547 tokens/s	2 GPUs423 tokens/s
A100 80GB	80 GB	4 GPUs715 tokens/s	2 GPUs646 tokens/s	1 GPU500 tokens/s
H100	80 GB	4 GPUs1,174 tokens/s	2 GPUs1,061 tokens/s	1 GPU821 tokens/s
RTX PRO 6000 Blackwell	96 GB	4 GPUs628 tokens/s	2 GPUs568 tokens/s	1 GPU439 tokens/s
DGX Spark	128 GB	2 GPUs48 tokens/s	2 GPUs86 tokens/s	1 GPU67 tokens/s
H200	141 GB	2 GPUs841 tokens/s	1 GPU760 tokens/s	1 GPU1,176 tokens/s
B200	180 GB	2 GPUs1,402 tokens/s	1 GPU1,267 tokens/s	1 GPU1,960 tokens/s
Instinct MI300X	192 GB	2 GPUs929 tokens/s	1 GPU840 tokens/s	1 GPU1,299 tokens/s
Instinct MI325X	256 GB	1 GPU526 tokens/s	1 GPU950 tokens/s	1 GPU1,470 tokens/s
Instinct MI355X	288 GB	1 GPU701 tokens/s	1 GPU1,267 tokens/s	1 GPU1,960 tokens/s

Models of a similar size

Every model

Questions

How much VRAM does gpt-oss-120b need?: 234 GB for its weights at 16-bit, 117 GB at 8-bit and 65.7 GB at 4-bit, plus the KV cache: 36.9 KB for each token, 1.21 GB for a conversation of 32K tokens and 4.84 GB at its longest context of 131,072 tokens. The server also keeps about 8% of each GPU free.
Can gpt-oss-120b run on 1 GPU?: Yes, at 4-bit: the smallest single GPU here that holds it with a 32K conversation is the A100 80GB.
How many H100s does gpt-oss-120b need?: With 1 conversation of 32K tokens, 4 H100s at 16-bit and 1 H100 at 4-bit. Each H100 has 80 GB, of which vLLM uses 92% by default.
Can I run gpt-oss-120b on an RTX 5090?: The RTX 5090 has 32 GB. With a 32K conversation, gpt-oss-120b needs 4 GeForce RTX 5090s at 4-bit and 8 GeForce RTX 5090s at 16-bit.
How fast does gpt-oss-120b run?: At 4-bit on 1 H100, memory lets 1 conversation get at most 821 tokens a second. Real servers reach part of that ceiling, and each conversation gets less when many run at once.

Sources

gpt-oss-120b at a glance

GPUs needed at each number format

Models of a similar size

Questions

Where each fact comes from