How much VRAM does Gemma 4 12B need?

23.9 GB for its weights at 16-bit, 12 GB at 8-bit and 6.73 GB at 4-bit, plus the KV cache: 16.4 KB for each token, 872 MB for a conversation of 32K tokens and 4.63 GB at its longest context of 262,144 tokens. The server also keeps about 8% of each GPU free.

Can Gemma 4 12B run on 1 GPU?

Yes, at 4-bit: the smallest single GPU here that holds it with a 32K conversation is the L4.

How many H100s does Gemma 4 12B need?

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

Can I run Gemma 4 12B on an RTX 5090?

The RTX 5090 has 32 GB. With a 32K conversation, Gemma 4 12B needs 1 GeForce RTX 5090 at 4-bit and 1 GeForce RTX 5090 at 16-bit.

How fast does Gemma 4 12B run?

At 4-bit on 1 H100, memory lets 1 conversation get at most 441 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 Gemma 4 12B need?

Gemma 4 12B by Google has 12B parameters. Its weights take 23.9 GB at 16-bit, 12 GB at 8-bit and 6.73 GB at 4-bit; released in BF16, its files are 23.9 GB. Each token of a conversation adds 16.4 KB of KV cache, so a 32K conversation holds 872 MB.

Size Gemma 4 12B for your setup Questions

Google · released in BF16

Gemma 4 12B at a glance

Parameters: 12B
As released: 23.9 GB; Its BF16 files
At 16-bit: 23.9 GB
At 4-bit: 6.73 GB; NVFP4
KV cache a token: 16.4 KB; At 16-bit
Longest context: 262,144; 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	2 GPUs24 tokens/s	1 GPU23 tokens/s	1 GPU39 tokens/s
GeForce RTX 5090	32 GB	1 GPU72 tokens/s	1 GPU140 tokens/s	1 GPU236 tokens/s
L40S	48 GB	1 GPU35 tokens/s	1 GPU67 tokens/s	1 GPU114 tokens/s
A100 80GB	80 GB	1 GPU82 tokens/s	1 GPU159 tokens/s	1 GPU268 tokens/s
H100	80 GB	1 GPU135 tokens/s	1 GPU261 tokens/s	1 GPU441 tokens/s
RTX PRO 6000 Blackwell	96 GB	1 GPU72 tokens/s	1 GPU140 tokens/s	1 GPU236 tokens/s
DGX Spark	128 GB	1 GPU11 tokens/s	1 GPU21 tokens/s	1 GPU36 tokens/s
H200	141 GB	1 GPU194 tokens/s	1 GPU374 tokens/s	1 GPU632 tokens/s
B200	180 GB	1 GPU323 tokens/s	1 GPU623 tokens/s	1 GPU1,053 tokens/s
Instinct MI300X	192 GB	1 GPU214 tokens/s	1 GPU413 tokens/s	1 GPU697 tokens/s
Instinct MI325X	256 GB	1 GPU242 tokens/s	1 GPU468 tokens/s	1 GPU789 tokens/s
Instinct MI355X	288 GB	1 GPU323 tokens/s	1 GPU623 tokens/s	1 GPU1,053 tokens/s

Models of a similar size

Every model

Questions

How much VRAM does Gemma 4 12B need?: 23.9 GB for its weights at 16-bit, 12 GB at 8-bit and 6.73 GB at 4-bit, plus the KV cache: 16.4 KB for each token, 872 MB for a conversation of 32K tokens and 4.63 GB at its longest context of 262,144 tokens. The server also keeps about 8% of each GPU free.
Can Gemma 4 12B run on 1 GPU?: Yes, at 4-bit: the smallest single GPU here that holds it with a 32K conversation is the L4.
How many H100s does Gemma 4 12B need?: With 1 conversation of 32K tokens, 1 H100 at 16-bit and 1 H100 at 4-bit. Each H100 has 80 GB, of which vLLM uses 92% by default.
Can I run Gemma 4 12B on an RTX 5090?: The RTX 5090 has 32 GB. With a 32K conversation, Gemma 4 12B needs 1 GeForce RTX 5090 at 4-bit and 1 GeForce RTX 5090 at 16-bit.
How fast does Gemma 4 12B run?: At 4-bit on 1 H100, memory lets 1 conversation get at most 441 tokens a second. Real servers reach part of that ceiling, and each conversation gets less when many run at once.

Sources

Gemma 4 12B at a glance

GPUs needed at each number format

Models of a similar size

Questions

Where each fact comes from