Can I Run DeepSeek-OCR on 96 GB system RAM?

Written by Jakub Rusinowski · Last updated September 11, 2026

Yes — comfortably

Yes, comfortably — DeepSeek-OCR 3B at Q8_0 needs about 4.8 GB of the 76.8 GB usable on 96 GB system RAM, leaving ~72 GB spare and running at ~57.1 tok/s (estimated), with room for about 8,192 tokens of context.

Confidence: medium · Recommended quantization: Q8_0 · Estimated speed: ~57.1 tok/s

See what else this hardware can run →

Buy this hardware Intel Arc B570 10GB — 10 GB VRAM · 150 W board power Deploy in the cloud now RTX 4090 on RunPod — from $0.34/hr · rate checked 2026-07

or compare on Vast.ai from $0.35/hr (typical low · varies)

As an Amazon Associate we earn from qualifying purchases. Cloud GPU links are referral links — we may earn a commission at no extra cost to you.

96 GB system RAM — what it gives a model

Usable memory for models	76.8 GB
Memory bandwidth	90 GB/s

DeepSeek-OCR on 96 GB system RAM: memory by quantization

Quant	Memory needed	Fits 76.8 GB?	Max context	Est. speed	Download
F16	7.6 GB	✓ Yes	8K	~39.3 tok/s	6 GB
Q8_0	4.8 GB	✓ Yes	8K	~57.1 tok/s	3.2 GB
Q6_K	4.1 GB	✓ Yes	8K	~64.8 tok/s	2.5 GB
Q5_K_M	3.8 GB	✓ Yes	8K	~69 tok/s	2.1 GB
Q4_K_M	3.5 GB	✓ Yes	8K	~73.5 tok/s	1.8 GB
Q3_K_M	2.9 GB	✓ Yes	8K	~82.6 tok/s	1.3 GB
Q2_K	2.6 GB	✓ Yes	8K	~88.7 tok/s	1 GB

What to watch out for

This model does not publish its full attention configuration, so the KV-cache share of these figures is inferred from its parameter count rather than computed exactly.
These figures assume CPU-only inference. Any discrete GPU, even an 8 GB one, will be several times faster for models that fit in its VRAM.

Recommended setup

llama.cpp (CPU build) or Ollama — both run without a GPU

How these numbers are calculated

Memory need = quantized weights + KV cache at 8,192 tokens (f16) + 0.8 GB runtime overhead.
76.8 GB of the 96 GB is treated as usable for model weights (80% — the rest is the OS and running applications).
DDR5-5600 dual channel at 89.6 GB/s peak. CPU decode is assumed to sustain 35% of that peak, because CPU inference is not purely bandwidth-bound — it also spends real time in compute and thread synchronisation. This figure is an assumption, not a fitted constant: no CPU measurement is in the calibration set.
CPU-only inference: no GPU is assumed. A GPU of any size will beat these figures substantially.
Throughput is a memory-bandwidth roofline estimate, not a measurement. It is labelled "estimated" everywhere it appears.
KV cache is inferred from the parameter count because this model does not publish its full attention configuration.

FAQ