xAI
Grok 4
Frontier reasoning model from xAI
Model Summary
Family
Grok
Version
4.0
Parameters
270B (est.)
Parameter counts for closed models are estimates; vendors rarely publish exact sizes.
VRAM Requirements by Quantization
Memory needed to serve Grok 4 for inference, including a 20% overhead for activations and KV cache.
| Precision | VRAM needed | Smallest single GPU that fits |
|---|---|---|
| INT4 (4-bit) | 150.87 GB | NVIDIA B200 SXM (192 GB) |
| INT8 (8-bit) | 301.75 GB | Multi-GPU required |
| FP16 (16-bit) | 603.50 GB | Multi-GPU required |
| FP32 (32-bit) | 1206.99 GB | Multi-GPU required |
Recommended GPU Configurations
Cheapest on-demand configurations to serve Grok 4 at 8-bit (302 GB VRAM).
2x AMD Instinct MI300X
384 GB total VRAM · CDNA 3
~$12.00/h
2x AMD Instinct MI325X
512 GB total VRAM · CDNA 3
~$16.00/h
4x NVIDIA A100 80GB
320 GB total VRAM · Ampere
~$18.00/h
Quick GPU Planning
Use the calculator pre-filled with this exact version to estimate memory, speed, and compute requirements in a few clicks.
Access Pre-filled CalculatorFrequently Asked Questions
How much VRAM do you need to run Grok 4?
With an estimated 270B parameters, Grok 4 needs roughly 302 GB of VRAM in 8-bit (INT8), 151 GB in 4-bit, and 603 GB in FP16, including a 20% overhead for activations and KV cache.
Which GPUs can run Grok 4?
At 8-bit quantization, the most cost-effective option is 2x AMD Instinct MI300X (384 GB combined VRAM, around $12.00/hour on-demand). Higher-end cards like the NVIDIA B200 or AMD MI355X reduce the GPU count needed.
Can Grok 4 run on a single GPU?
Only with aggressive quantization: in 4-bit, a single NVIDIA B200 SXM (192 GB) can fit it. At 8-bit or higher, you need a multi-GPU setup.
How much does it cost to serve Grok 4 in the cloud?
Renting 2x Instinct MI300X costs on the order of $12.00/hour, i.e. about $8,760/month running 24/7. Actual prices vary by provider and commitment; spot and reserved capacity can be significantly cheaper.
Deploy on a GPU cloud
Rent 2x Instinct MI300X by the hour instead of buying hardware.
