Skip to content

Phi 4 on RTX PRO 5000 Blackwell 48 GB: VRAM, speed and fit

Phi 4 at 8-bit runs on the RTX PRO 5000 Blackwell 48 GB with 16K of context at about 58 tokens a second.

Text generation · Chat · Ollama / LM Studio

Runs comfortably · on RTX PRO 5000 Blackwell 48 GB · 8-bit (Q8_0) · Ollama / LM Studio · Chat

GPU memory
18.4 GB
of 48.0 GB
Context
16K
auto
Speed
58 tok/s
48 when full
First reply
0.4 s
2K prompt

Where the memory goes

Model
14.5 GB
Context
3.1 GB
Overhead
817 MB
OS
0.0 GB
Free
29.6 GB
Memory
18.4 GB of 48.0 GB
Context
16K tokens · auto
Speed
58 tok/s · 48 when full
First reply
0.4 s for a 2K prompt
Running cost
$5.69/mo at home · $30/mo on a cloud RTX A4500

Validation: Validated

What Phi 4 on RTX PRO 5000 Blackwell 48 GB was checked against
CheckTargetResultStatus
Fits in GPU memory48.0 GB usable18.4 GB neededPass
Headroom5% free29.6 GB free (62%)Pass
Context for chat16K or more16KPass
Writing speed15 tok/s or more58 tok/sPass
First replyunder 3 s, 2K prompt0.4 sPass

Targets: a reply you can read along with, a first reply the use case tolerates, and 5% of memory left free. Calculated from published specs, not benchmarked.

The setup

Model
Phi 4
Parameters
14.7B
Precision
8-bit (Q8_0)
Context
16K, auto
Chats at once
1
Use
Chat
Runs in
Ollama / LM Studio
Hardware
RTX PRO 5000 Blackwell 48 GB
GPU can use
48.0 GB of 48.0 GB

Running cost

At home
$5.69/mo
260 W at the wall
Renting instead
$30/mo
RTX A4500 at $0.25/h

4 h a day, power at $0.18 per kWh, cloud at RunPod on-demand prices as of Nov 21, 2026.

How to run it

llama.cpp
llama-server -hf MaziyarPanahi/phi-4-GGUF
vLLM
vllm serve microsoft/phi-4
MLX (Mac)
mlx_lm.chat --model mlx-community/phi-4-4bit

Phi 4 on similar machines

All machines: Phi 4 VRAM requirements. Everything that runs here: what can the RTX PRO 5000 Blackwell 48 GB run?

Other models that run on RTX PRO 5000 Blackwell 48 GB

Questions

Can Phi 4 run on RTX PRO 5000 Blackwell 48 GB?

Phi 4 at 8-bit runs on the RTX PRO 5000 Blackwell 48 GB with 16K of context at about 58 tokens a second.

Report VR-1XEETFO, calculated Nov 22, 2026 from published specs, not benchmarked.