What AI models can a RTX PRO 5000 Blackwell run?

2070 of 2118 indexed open-weight models fit a RTX PRO 5000 Blackwell at 32,768 context with q8_0 KV cache, the largest being GLM-Z1-Rumination-32B-0414 at BF16. That covers text, vision-language, image, video and speech models.

How much usable memory does a RTX PRO 5000 Blackwell actually have?

Its nameplate is 72 GB, but about 66.96 GiB is available to a model once driver and compositor overhead is accounted for.

Is a RTX PRO 5000 Blackwell fast for local AI?

Its memory bandwidth is 1344 GB/s, and that figure — not teraflops — is what governs token generation speed. Capacity decides what you can run; bandwidth decides how fast it runs.

NVIDIA · workstation

RTX PRO 5000 Blackwell

RTX PRO 5000 Blackwell has 72 GB of VRAM at 1344 GB/s — about 66.96 GiB usable after driver and compositor overhead. 2070 of 2118 indexed models fit at 32K context with q8_0 KV.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed

Memory

72 GB

GDDR7

Bandwidth

1344 GB/s

384-bit bus

Tensor FP16

295 TF

dense

TDP

300 W

$4569 MSRP

Context4K 8K 16K 32K 64K 128K

KV cachef16 q8_0 q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context

text 1780vision language 186image 2audio asr 39audio tts 21video 16embedding 26

What fits at 32K context

largest quantization that fits, per model · 2070 of 2118 indexed

Model	Best quant	Params	Weights●	KV●	Total in memory◐	Headroom◐	tok/s◐
GLM-Z1-Rumination-32B-0414	BF16	33.1B	61.74 GiB	4.05 GiB	66.88 GiB	0.08 GiB	12±22%
command-a-plus-05-2026-bf16MoE	IQ2_S	219B	65.10 GiB	0.76 GiB	66.86 GiB	0.10 GiB	51±37%
NVIDIA-Nemotron-3-Super-120B-A12B-BF16MoE	IQ4_NL	124B	64.39 GiB	1.46 GiB	66.85 GiB	0.11 GiB	54±37%
GLM-4.6VMoE	Q4_1	108B	62.66 GiB	3.05 GiB	66.74 GiB	0.22 GiB	41±37%
Ornith-Agents-A1-3.7-35B-A3B-dare_ties_v4MoE	F16	34.7B	65.34 GiB	0.33 GiB	66.68 GiB	0.28 GiB	69±37%
Ornith-Agents-A1-3.6-35B-A3B-dare_tiesMoE	F16	34.7B	65.34 GiB	0.33 GiB	66.68 GiB	0.28 GiB	69±37%
GLM-4.7-REAP-218B-A32BMoE	IQ2_S	218B	59.49 GiB	6.11 GiB	66.64 GiB	0.32 GiB	30±37%
Mistral-Small-4-119B-2603MoE	Q4_K_S	119B	65.08 GiB	0.37 GiB	66.49 GiB	0.47 GiB	69±37%
GLM-4.5VMoE	I1-Q4_1	108B	62.40 GiB	3.05 GiB	66.48 GiB	0.48 GiB	41±37%
Step-3.5-Flash-REAP-121B-A11B	I1-Q3_K_L	121B	58.48 GiB	6.92 GiB	66.43 GiB	0.53 GiB	12±22%
Caller	BF16	32.8B	61.04 GiB	4.25 GiB	66.38 GiB	0.58 GiB	12±22%
Dumpling-Qwen2.5-32B	BF16	32.8B	61.04 GiB	4.25 GiB	66.38 GiB	0.58 GiB	12±22%
OpenThinker-32B	F16	32.8B	61.04 GiB	4.25 GiB	66.38 GiB	0.58 GiB	12±22%
INTELLECT-2	BF16	32.8B	61.04 GiB	4.25 GiB	66.38 GiB	0.58 GiB	12±22%
openhands-lm-32b-v0.1	BF16	32.8B	61.04 GiB	4.25 GiB	66.38 GiB	0.58 GiB	12±22%
LongWriter-Zero-32B	BF16	32.8B	61.04 GiB	4.25 GiB	66.38 GiB	0.58 GiB	12±22%
OpenCodeReasoning-Nemotron-32B	BF16	32.8B	61.04 GiB	4.25 GiB	66.38 GiB	0.58 GiB	12±22%
OpenCodeReasoning-Nemotron-32B-IOI	BF16	32.8B	61.04 GiB	4.25 GiB	66.38 GiB	0.58 GiB	12±22%
Qwen2.5-Coder-32B-Instruct-abliterated	F16	32.8B	61.04 GiB	4.25 GiB	66.38 GiB	0.58 GiB	12±22%
QwQ-32B-ArliAI-RpR-v4	BF16	32.8B	61.04 GiB	4.25 GiB	66.38 GiB	0.58 GiB	12±22%
OpenThinker2-32B	BF16	32.8B	61.04 GiB	4.25 GiB	66.38 GiB	0.58 GiB	12±22%
Qwen2.5-Coder-32B	F16	32.8B	61.04 GiB	4.25 GiB	66.38 GiB	0.58 GiB	12±22%
Qwen2.5-32B-Instruct	F16	32.8B	61.04 GiB	4.25 GiB	66.38 GiB	0.58 GiB	12±22%
QwQ-32B-Preview	BF16	32.8B	61.04 GiB	4.25 GiB	66.38 GiB	0.58 GiB	12±22%
Qwen2.5-32b-RP-Ink	F16	32.8B	61.04 GiB	4.25 GiB	66.38 GiB	0.58 GiB	12±22%
deepseek-r1-qwen-2.5-32B-ablated	BF16	32.8B	61.04 GiB	4.25 GiB	66.38 GiB	0.58 GiB	12±22%
Rombos-LLM-V2.5-Qwen-32b	F16	32.8B	61.04 GiB	4.25 GiB	66.38 GiB	0.58 GiB	12±22%
DeepSeek-R1-Distill-Qwen-32B-abliterated	BF16	32.8B	61.04 GiB	4.25 GiB	66.38 GiB	0.58 GiB	12±22%
Qwen2.5-32B-ArliAI-RPMax-v1.3	F16	32.8B	61.04 GiB	4.25 GiB	66.38 GiB	0.58 GiB	12±22%
DeepSeek-R1-Distill-Qwen-32B	F16	32.8B	61.04 GiB	4.25 GiB	66.38 GiB	0.58 GiB	12±22%
Qwen2.5-VL-32B-Instruct	BF16	33.5B	61.04 GiB	4.25 GiB	66.38 GiB	0.58 GiB	12±22%
EVA-Qwen2.5-32B-v0.2	F16	32.8B	61.04 GiB	4.25 GiB	66.38 GiB	0.58 GiB	12±22%
EVA-Qwen2.5-32B-v0.1	F16	32.8B	61.04 GiB	4.25 GiB	66.38 GiB	0.58 GiB	12±22%
cogito-v1-preview-qwen-32B	BF16	32.8B	61.03 GiB	4.25 GiB	66.38 GiB	0.58 GiB	12±22%
QwQ-32B-Snowdrop-v0	BF16	32.8B	61.03 GiB	4.25 GiB	66.38 GiB	0.58 GiB	12±22%
OpenBuddy-R1-0528-Distill-Qwen3-32B-Preview0-QAT	BF16	32.8B	61.03 GiB	4.25 GiB	66.37 GiB	0.59 GiB	12±22%
KAT-Dev	BF16	32.8B	61.03 GiB	4.25 GiB	66.37 GiB	0.59 GiB	12±22%
Qwen3-VL-32B-Instruct	BF16	33.4B	61.03 GiB	4.25 GiB	66.37 GiB	0.59 GiB	12±22%
Qwen3-VL-32B-Thinking	BF16	33.4B	61.03 GiB	4.25 GiB	66.37 GiB	0.59 GiB	12±22%
Qwen3-VL-32B-Instruct-ultra-uncensored-heretic	BF16	33.4B	61.03 GiB	4.25 GiB	66.37 GiB	0.59 GiB	12±22%
Qwen3-32B	BF16	32.8B	61.03 GiB	4.25 GiB	66.37 GiB	0.59 GiB	12±22%
DeepSWE-Preview	BF16	32.8B	61.03 GiB	4.25 GiB	66.37 GiB	0.59 GiB	12±22%
HuatuoGPT-o1-72B	Q6_K	72.7B	59.93 GiB	5.31 GiB	66.37 GiB	0.59 GiB	12±22%
Rombo-LLM-V3.0-Qwen-72b	Q6_K	72.7B	59.93 GiB	5.31 GiB	66.37 GiB	0.59 GiB	12±22%
Qwen2.5-72B-Instruct-abliterated	Q6_K	72.7B	59.93 GiB	5.31 GiB	66.37 GiB	0.59 GiB	12±22%
EVA-Qwen2.5-72B-v0.2	Q6_K	72.7B	59.93 GiB	5.31 GiB	66.37 GiB	0.59 GiB	12±22%
MiroThinker-v1.0-72B	Q6_K	72.7B	59.93 GiB	5.31 GiB	66.37 GiB	0.59 GiB	12±22%
Qwen2.5-Math-72B-Instruct	Q6_K	72.7B	59.93 GiB	5.31 GiB	66.37 GiB	0.59 GiB	12±22%
Qwen2.5-72B-Instruct	Q6_K	72.7B	59.93 GiB	5.31 GiB	66.37 GiB	0.59 GiB	12±22%
Qwen2.5-72B	Q6_K	72.7B	59.93 GiB	5.31 GiB	66.37 GiB	0.59 GiB	12±22%
Kimi-Dev-72B	Q6_K	72.7B	59.93 GiB	5.31 GiB	66.37 GiB	0.59 GiB	12±22%
magnum-v4-72b	Q6_K	72.7B	59.93 GiB	5.31 GiB	66.37 GiB	0.59 GiB	12±22%
KAT-Dev-72B-Exp	Q6_K	72.7B	59.93 GiB	5.31 GiB	66.37 GiB	0.59 GiB	12±22%
Chuluun-Qwen2.5-72B-v0.01	Q6_K	72.7B	59.93 GiB	5.31 GiB	66.37 GiB	0.59 GiB	12±22%
Homer-v1.0-Qwen2.5-72B	Q6_K	72.7B	59.93 GiB	5.31 GiB	66.37 GiB	0.59 GiB	12±22%
Qwen2.5-VL-72B-Instruct	Q6_K	73.4B	59.93 GiB	5.31 GiB	66.37 GiB	0.59 GiB	12±22%
Tower-Plus-72B-ultra-uncensored-heretic	I1-Q6_K	72.7B	59.93 GiB	5.31 GiB	66.37 GiB	0.59 GiB	12±22%
Chronos-Platinum-72B	Q6_K	72.7B	59.93 GiB	5.31 GiB	66.37 GiB	0.59 GiB	12±22%
UI-TARS-72B-DPO	Q6_K	73.4B	59.93 GiB	5.31 GiB	66.37 GiB	0.59 GiB	12±22%
Qwen3.5-REAP-212B-A17BMoE	IQ2_M	212B	64.76 GiB	0.50 GiB	66.31 GiB	0.65 GiB	64±37%

●From the file◐Predictedwhat these mean

Speed is modeled, not measured: decode is memory-bandwidth bound, so tokens per second is bytes read per token against achievable bandwidth. Mixture-of-experts models carry a wider band because only the routed experts are read each step, and few have been measured publicly.

Questions people ask

What AI models can a RTX PRO 5000 Blackwell run?: 2070 of 2118 indexed open-weight models fit a RTX PRO 5000 Blackwell at 32,768 context with q8_0 KV cache, the largest being GLM-Z1-Rumination-32B-0414 at BF16. That covers text, vision-language, image, video and speech models.
How much usable memory does a RTX PRO 5000 Blackwell actually have?: Its nameplate is 72 GB, but about 66.96 GiB is available to a model once driver and compositor overhead is accounted for.
Is a RTX PRO 5000 Blackwell fast for local AI?: Its memory bandwidth is 1344 GB/s, and that figure — not teraflops — is what governs token generation speed. Capacity decides what you can run; bandwidth decides how fast it runs.