Can I Run OLMo 2 on 96 GB system RAM?

作者： Jakub Rusinowski · 最后更新： 2024年11月26日

Yes — comfortably

Yes, comfortably — OLMo 2 13B Instruct at Q5_K_M needs about 17.2 GB of the 76.8 GB usable on 96 GB system RAM, leaving ~59.6 GB spare and running at ~5.1 tok/s (estimated), with room for about 4,096 tokens of context.

Confidence: high · Recommended quantization: Q5_K_M · Estimated speed: ~5.1 tok/s

Buy This HardwareAMD Radeon RX 7900 XT 20GB — 20 GB VRAM · 315 W board power Deploy in the Cloud NowRTX 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

OLMo 2 on 96 GB system RAM: memory by quantization

Quant	Memory needed	Fits 76.8 GB?	Max context	Est. speed	Download
F16	34.9 GB	✓ Yes	4K	~2.2 tok/s	27.4 GB
Q8_0	22.1 GB	✓ Yes	4K	~3.7 tok/s	14.6 GB
Q6_K	18.7 GB	✓ Yes	4K	~4.6 tok/s	11.2 GB
Q5_K_M	17.2 GB	✓ Yes	4K	~5.1 tok/s	9.7 GB
Q4_K_M	15.8 GB	✓ Yes	4K	~5.7 tok/s	8.3 GB
Q3_K_M	13.4 GB	✓ Yes	4K	~7.2 tok/s	5.8 GB
Q2_K	12 GB	✓ Yes	4K	~8.4 tok/s	4.5 GB

Which OLMo 2 sizes fit

Variant	Needs at Q4_K_M	Fits?	Est. speed
OLMo 2 13B Instruct	15.8 GB	✓ Fits	~5.7 tok/s
OLMo 2 7B Instruct	9.5 GB	✓ Fits	~10 tok/s

What to watch out for

Context is capped at about 4,096 tokens before memory runs out, which is short for document or agent work.
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 computed from this model's published attention configuration.

FAQ