Which local LLMs can the DDR5-5600 dual-channel CPU run?

No GPU — the whole model runs from system RAM

CPU-only inference with 4 GB of RAM kept for the OS, 8k context and an f16 KV cache. Verdicts are shown for each RAM size.

Specs

Memory options: 16 GB · 32 GB · 64 GB · 96 GB
Memory bandwidth: 89.6 GB/s
Launch year: 2021

Verdicts at a glance

At Q4_K_M (or the closest available quant) with 8k context.

16 GBRuns great0 modelsRuns well0 modelsRuns slowly13 modelsWon't run16 models3 more models run at a lower quant.
32 GBRuns great0 modelsRuns well0 modelsRuns slowly21 modelsWon't run8 models1 more model runs at a lower quant.
64 GBRuns great0 modelsRuns well0 modelsRuns slowly21 modelsWon't run8 models3 more models run at a lower quant.
96 GBRuns great0 modelsRuns well0 modelsRuns slowly24 modelsWon't run5 models1 more model runs at a lower quant.

How to read the verdicts

Runs great: Fully on the GPU at 20 tok/s or more
Runs well: Fully on the GPU at 8–20 tok/s; MoE experts in system RAM at 20 tok/s or more; or at least 90% on the GPU at 8 tok/s or more
Runs slowly: 2–8 tok/s; CPU-only; less than 90% on the GPU; or MoE experts in system RAM below 20 tok/s
Won't run: Does not fit, or under 2 tok/s

Every model on the DDR5-5600 dual-channel CPU

Scroll sideways to see every column.

29 models sorted by verdict and speed, with 96 GB of memory
Model	Verdict	Speed	Quant	Memory	Runs as	Context	Notes
EXAONE 4.0 1.2B	Runs slowly	est. 34.1 tok/s27.3–40.9calibrated estimate ±20%	Q4_K_M	1.3 GB RAM	CPU only	up to 64k	CPU inference — capped at Runs slowly
HyperCLOVA X SEED 1.5B	Runs slowly	est. 24.7 tok/s19.7–29.6calibrated estimate ±20%	Q4_K_M	1.8 GB RAM	CPU only	up to 16k	CPU inference — capped at Runs slowly
Qwen3 30B-A3B (2507)	Runs slowly	19.0 tok/s8k estimate 13.7 tok/s (11.0–16.4, calibrated estimate ±20%)measured (1 run, 2k context)	Q4_K_M	19.4 GB RAM	CPU only	up to 128k	CPU inference — capped at Runs slowly
Qwen3.5 35B-A3B	Runs slowly	est. 18.7 tok/s15.0–22.5calibrated estimate ±20%	Q4_K_M	22.8 GB RAM	CPU only	up to 256k	CPU inference — capped at Runs slowly
gpt-oss-20b	Runs slowly	est. 16.8 tok/s13.5–20.2calibrated estimate ±20%	MXFP4	12.3 GB RAM	CPU only	up to 128k	CPU inference — capped at Runs slowly reasoning model
Gemma 4 26B-A4B	Runs slowly	est. 13.3 tok/s10.7–16.0calibrated estimate ±20%	Q4_K_M	17.3 GB RAM	CPU only	up to 256k	CPU inference — capped at Runs slowly
Kanana 1.5 15.7B-A3B	Runs slowly	est. 12.6 tok/s10.0–15.1calibrated estimate ±20%	Q4_K_M	11.5 GB RAM	CPU only	up to 32k	CPU inference — capped at Runs slowly
gpt-oss-120b	Runs slowly	est. 12.6 tok/s10.0–15.1calibrated estimate ±20%	MXFP4	63.7 GB RAM	CPU only	up to 128k	CPU inference — capped at Runs slowly reasoning model
DeepSeek R1 Distill Llama 8B	Runs slowly	est. 7.5 tok/s6.0–9.0calibrated estimate ±20%	Q4_K_M	6.0 GB RAM	CPU only	up to 64k	reasoning model
Kanana 1.5 8B	Runs slowly	est. 7.5 tok/s6.0–9.0calibrated estimate ±20%	Q4_K_M	6.0 GB RAM	CPU only	up to 32k	—
Llama 3.1 8B	Runs slowly	est. 7.5 tok/s6.0–9.0calibrated estimate ±20%	Q4_K_M	6.0 GB RAM	CPU only	up to 128k	—
Qwen3.5 9B	Runs slowly	est. 7.3 tok/s5.8–8.8calibrated estimate ±20%	Q4_K_M	6.1 GB RAM	CPU only	up to 256k	—
Qwen3 8B	Runs slowly	est. 7.2 tok/s5.7–8.6calibrated estimate ±20%	Q4_K_M	6.2 GB RAM	CPU only	up to 32k	—
Qwen3.5 122B-A10B	Runs slowly	est. 6.0 tok/s4.8–7.2calibrated estimate ±20%	Q4_K_M	78.5 GB RAM	CPU only	up to 256k	—
Gemma 4 12B	Runs slowly	est. 5.9 tok/s4.7–7.0calibrated estimate ±20%	Q4_K_M	7.7 GB RAM	CPU only	up to 128k	—
Gemma 3 12B	Runs slowly	est. 5.7 tok/s4.6–6.9calibrated estimate ±20%	Q4_K_M	7.8 GB RAM	CPU only	up to 128k	—
HyperCLOVA X SEED Think 14B	Runs slowly	est. 4.4 tok/s3.1–5.7theoretical estimate ±30%	Q4_K_M	10.2 GB RAM	CPU only	up to 32k	reasoning model
Solar Open 100B	Runs slowly	est. 4.3 tok/s3.5–5.2calibrated estimate ±20%	Q4_K_M	63.9 GB RAM	CPU only	up to 32k	—
Qwen3 14B	Runs slowly	est. 4.3 tok/s3.5–5.2calibrated estimate ±20%	Q4_K_M	10.3 GB RAM	CPU only	up to 32k	—
Phi-4	Runs slowly	est. 4.2 tok/s3.3–5.0calibrated estimate ±20%	Q4_K_M	10.7 GB RAM	CPU only	up to 16k	reasoning model
Mistral Small 3.2 24B	Runs slowly	est. 2.9 tok/s2.3–3.4calibrated estimate ±20%	Q4_K_M	15.7 GB RAM	CPU only	up to 32k	—
Gemma 3 27B	Runs slowly	est. 2.6 tok/s2.1–3.1calibrated estimate ±20%	Q4_K_M	17.2 GB RAM	CPU only	up to 64k	—
Qwen3.5 27B	Runs slowly	est. 2.5 tok/s2.0–3.0calibrated estimate ±20%	Q4_K_M	17.6 GB RAM	CPU only	up to 64k	—
Qwen3 32B	Runs slowly	est. 2.0 tok/s1.6–2.5calibrated estimate ±20%	Q4_K_M	21.9 GB RAM	CPU only	up to 8k	—
DeepSeek R1 Distill Qwen 32B	Won't runTry Q2_K (heavy quality loss): Runs slowly	est. 3.1 tok/s2.5–3.7calibrated estimate ±20%	Q2_K	14.5 GB RAM	CPU only	up to 8k	reasoning model
EXAONE 4.0 32B	Won't run	est. 2.3 tok/s1.8–2.7calibrated estimate ±20%	Q4_K_M	19.9 GB RAM	CPU only	—	Fits, but too slow to use reasoning model
EXAONE 4.5 33B	Won't run	est. 2.2 tok/s1.7–2.6calibrated estimate ±20%	Q4_K_M	20.6 GB RAM	CPU only	—	Fits, but too slow to use reasoning model
Llama 3.3 70B	Won't run	est. 1.0 tok/s0.8–1.2calibrated estimate ±20%	Q4_K_M	45.2 GB RAM	CPU only	—	Fits, but too slow to use
Solar Open 2 250B	Won't run	—	IQ4_XS	needs 136.6 GB	—	—	Needs about 136.6 GB; 92.0 GB of RAM is free

Reasoning models spend extra tokens thinking, so their speed thresholds are 1.5× stricter (30 / 12 / 3 tok/s).

Verdict by memory size

Q4_K_M verdict for each memory configuration — open the model page for speeds
Model	16 GB	32 GB	64 GB	96 GB
EXAONE 4.0 1.2B	Runs slowly	Runs slowly	Runs slowly	Runs slowly
HyperCLOVA X SEED 1.5B	Runs slowly	Runs slowly	Runs slowly	Runs slowly
Qwen3 30B-A3B (2507)	Won't run	Runs slowly	Runs slowly	Runs slowly
Qwen3.5 35B-A3B	Won't run	Runs slowly	Runs slowly	Runs slowly
gpt-oss-20b	Won't run	Runs slowly	Runs slowly	Runs slowly
Gemma 4 26B-A4B	Won't run	Runs slowly	Runs slowly	Runs slowly
Kanana 1.5 15.7B-A3B	Runs slowly	Runs slowly	Runs slowly	Runs slowly
gpt-oss-120b	Won't run	Won't run	Won't run	Runs slowly
DeepSeek R1 Distill Llama 8B	Runs slowly	Runs slowly	Runs slowly	Runs slowly
Kanana 1.5 8B	Runs slowly	Runs slowly	Runs slowly	Runs slowly
Llama 3.1 8B	Runs slowly	Runs slowly	Runs slowly	Runs slowly
Qwen3.5 9B	Runs slowly	Runs slowly	Runs slowly	Runs slowly
Qwen3 8B	Runs slowly	Runs slowly	Runs slowly	Runs slowly
Qwen3.5 122B-A10B	Won't run	Won't run	Won't run	Runs slowly
Gemma 4 12B	Runs slowly	Runs slowly	Runs slowly	Runs slowly
Gemma 3 12B	Runs slowly	Runs slowly	Runs slowly	Runs slowly
HyperCLOVA X SEED Think 14B	Runs slowly	Runs slowly	Runs slowly	Runs slowly
Solar Open 100B	Won't run	Won't run	Won't run	Runs slowly
Qwen3 14B	Runs slowly	Runs slowly	Runs slowly	Runs slowly
Phi-4	Runs slowly	Runs slowly	Runs slowly	Runs slowly
Mistral Small 3.2 24B	Won't run	Runs slowly	Runs slowly	Runs slowly
Gemma 3 27B	Won't run	Runs slowly	Runs slowly	Runs slowly
Qwen3.5 27B	Won't run	Runs slowly	Runs slowly	Runs slowly
Qwen3 32B	Won't run	Runs slowly	Runs slowly	Runs slowly
DeepSeek R1 Distill Qwen 32B	Won't run	Won't run	Won't run	Won't run
EXAONE 4.0 32B	Won't run	Won't run	Won't run	Won't run
EXAONE 4.5 33B	Won't run	Won't run	Won't run	Won't run
Llama 3.3 70B	Won't run	Won't run	Won't run	Won't run
Solar Open 2 250B	Won't run	Won't run	Won't run	Won't run

Measured results on the DDR5-5600 dual-channel CPU

Public benchmarks we calibrate against. Their conditions (context, backend, flags) can differ from the estimates above.

Measured results on the DDR5-5600 dual-channel CPU
Model	Quant	Backend	Context	Prompt (tok/s)	Generation (tok/s)	Flags	Source	Measured
mistral-7b	Q4_K_M	llamafile	512	—	11.2	—	github.com	2024-06-15
Qwen3 30B-A3B (2507)	Q4_K_M	llama.cpp	2k	—	19.0	—	hardware-corner.net	2026-08-09

Frequently asked questions

What is the largest model this CPU setup can run?

Qwen3.5 122B-A10B at Q4_K_M (a 78.26 GB file) runs from 96 GB of RAM at est. 6.0 tok/s (4.8–7.2, calibrated estimate ±20%).

How many local LLMs run well on the DDR5-5600 dual-channel CPU?

At Q4_K_M with 8k context, with 96 GB of memory, out of 29 tracked models: 0 run great, 0 run well, 24 run slowly and 5 won't run.

Can the DDR5-5600 dual-channel CPU run a 70B model like Llama 3.3 70B?

It fits in memory but is too slow to use: est. 1.0 tok/s (0.8–1.2, calibrated estimate ±20%).

Speeds are estimates from memory bandwidth, calibrated against public benchmarks, and each one comes with an error band and a confidence label. Real results vary with drivers, backend, context length and thermals.