What AI models can a RTX PRO 5000 Blackwell run?

2063 of 2118 indexed open-weight models fit a RTX PRO 5000 Blackwell at 65,536 context with q8_0 KV cache, the largest being command-r-35b-writer-v2 at I1-Q5_K_M. 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. 2063 of 2118 indexed models fit at 64K 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 1773vision language 186image 2audio asr 39audio tts 21video 16embedding 26

What fits at 64K context

largest quantization that fits, per model · 2063 of 2118 indexed

Model	Best quant	Params	Weights●	KV●	Total in memory◐	Headroom◐	tok/s◐
command-r-35b-writer-v2	I1-Q5_K_M	35.0B	23.29 GiB	42.50 GiB	66.90 GiB	0.06 GiB	12±22%
Mistral-Small-4-119B-2603MoE	Q4_K_S	119B	65.08 GiB	0.75 GiB	66.86 GiB	0.10 GiB	65±37%
CalmeRys-78B-Orpo-v0.1	Q5_K_M	78.0B	54.30 GiB	11.42 GiB	66.85 GiB	0.11 GiB	12±22%
calme-2.3-rys-78b	Q5_K_M	78.0B	54.30 GiB	11.42 GiB	66.85 GiB	0.11 GiB	12±22%
Gemma-4-Novelist-Eclipse-31B	BF16	32.7B	59.82 GiB	5.94 GiB	66.84 GiB	0.12 GiB	12±22%
Gemma-4-31B-StyleTune	BF16	32.7B	59.82 GiB	5.94 GiB	66.84 GiB	0.12 GiB	12±22%
Qwen3.5-REAP-212B-A17BMoE	IQ2_M	212B	64.76 GiB	1.00 GiB	66.81 GiB	0.15 GiB	60±37%
grok-2MoE	IQ1_M	270B	57.16 GiB	8.50 GiB	66.81 GiB	0.15 GiB	17±37%
Wizard-Vicuna-30B-Uncensored	I1-IQ3_M	32.5B	13.86 GiB	51.80 GiB	66.73 GiB	0.23 GiB	12±22%
archangel_sft-kto_llama30b	I1-IQ3_M	32.5B	13.86 GiB	51.80 GiB	66.73 GiB	0.23 GiB	12±22%
Skyfall-31B-v4.2	BF16	31.4B	58.41 GiB	7.17 GiB	66.70 GiB	0.26 GiB	12±22%
Qwen3.5-122B-A10B-hereticMoE	I1-Q4_K_S	123B	64.86 GiB	0.80 GiB	66.68 GiB	0.28 GiB	64±37%
c4ai-command-r-08-2024	F16	32.3B	60.17 GiB	5.31 GiB	66.59 GiB	0.37 GiB	12±22%
Laguna-S-2.1MoE	UD-Q4_K_S	118B	63.88 GiB	1.67 GiB	66.57 GiB	0.39 GiB	55±37%
Qwen3.5-REAP-262B-A17BMoE	IQ2_XXS	262B	64.50 GiB	1.00 GiB	66.55 GiB	0.41 GiB	64±37%
GLM-4.5-Air-DerestrictedMoE	Q4_0	110B	59.38 GiB	6.11 GiB	66.52 GiB	0.44 GiB	33±37%
GLM-4.5-AirMoE	Q4_0	110B	59.38 GiB	6.11 GiB	66.52 GiB	0.44 GiB	33±37%
NVIDIA-Nemotron-3-Super-120B-A12B-BF16MoE	IQ4_XS	124B	62.59 GiB	2.92 GiB	66.51 GiB	0.45 GiB	47±37%
MiMo-V2-FlashMoEKV unresolved	IQ1_M	310B	61.31 GiB	3.98 GiB	66.34 GiB	0.62 GiB	45±37%
Qwen3.6-35B-A3B-uncensored-hereticMoE	BF16	35.1B	64.61 GiB	0.66 GiB	66.28 GiB	0.68 GiB	66±37%
Darwin-35B-A3B-OpusMoE	BF16	36.0B	64.61 GiB	0.66 GiB	66.28 GiB	0.68 GiB	66±37%
Carnice-MoE-35B-A3BMoE	F16	36.0B	64.61 GiB	0.66 GiB	66.28 GiB	0.68 GiB	66±37%
Qwen3.6-35B-A3B-hereticMoE	BF16	35.1B	64.61 GiB	0.66 GiB	66.28 GiB	0.68 GiB	66±37%
Aurora-Code-1MoE	BF16	34.7B	64.61 GiB	0.66 GiB	66.28 GiB	0.68 GiB	66±37%
grug-35b-v2MoE	BF16	35.1B	64.61 GiB	0.66 GiB	66.28 GiB	0.68 GiB	66±37%
grug-35bMoE	BF16	35.1B	64.61 GiB	0.66 GiB	66.28 GiB	0.68 GiB	66±37%
WorldSim-Opus-3.6-35B-A3BMoE	BF16	35.1B	64.61 GiB	0.66 GiB	66.28 GiB	0.68 GiB	66±37%
Qwen3.6-35B-A3B-abliterated-MAXMoE	BF16	35.1B	64.61 GiB	0.66 GiB	66.28 GiB	0.68 GiB	66±37%
Huihui-Nex-N2-mini-abliteratedMoE	BF16	35.1B	64.61 GiB	0.66 GiB	66.28 GiB	0.68 GiB	66±37%
Qwen3.6-35B-A3B-AnkoMoE	BF16	35.1B	64.61 GiB	0.66 GiB	66.28 GiB	0.68 GiB	66±37%
KAT-Coder-V2.5-DevMoE	BF16	34.7B	64.61 GiB	0.66 GiB	66.28 GiB	0.68 GiB	66±37%
Ornith-1.0-35B-uncensored-hereticMoE	BF16	35.1B	64.61 GiB	0.66 GiB	66.28 GiB	0.68 GiB	66±37%
Qwen3.6-35B-A3B-abliterated-v4MoE	BF16	34.7B	64.61 GiB	0.66 GiB	66.28 GiB	0.68 GiB	66±37%
Qwen3.5-35B-A3B-ultra-uncensored-hereticMoE	BF16	35.1B	64.61 GiB	0.66 GiB	66.28 GiB	0.68 GiB	66±37%
Nex-N2-mini-ultra-uncensored-hereticMoE	BF16	35.1B	64.61 GiB	0.66 GiB	66.28 GiB	0.68 GiB	66±37%
Agents-A1MoE	F16	35.1B	64.61 GiB	0.66 GiB	66.28 GiB	0.68 GiB	66±37%
Qwen3.6-35B-A3B-java-v1MoE	BF16	34.7B	64.61 GiB	0.66 GiB	66.28 GiB	0.68 GiB	66±37%
Nex-N2-miniMoE	BF16	35.1B	64.61 GiB	0.66 GiB	66.28 GiB	0.68 GiB	66±37%
Qwen3.5-35B-A3B-BaseMoE	BF16	36.0B	64.61 GiB	0.66 GiB	66.28 GiB	0.68 GiB	66±37%
command-a-plus-05-2026-bf16MoE	IQ2_XS	219B	63.98 GiB	1.29 GiB	66.27 GiB	0.69 GiB	48±37%
Llama-3.3-70B-Instruct	Q6_K_L	70.6B	54.39 GiB	10.63 GiB	66.14 GiB	0.82 GiB	12±22%
Anubis-70B-v1.2	Q6_K_L	70.6B	54.39 GiB	10.63 GiB	66.14 GiB	0.82 GiB	12±22%
Tess-R1-Limerick-Llama-3.1-70B	Q6_K_L	70.6B	54.39 GiB	10.63 GiB	66.14 GiB	0.82 GiB	12±22%
Infinity-Instruct-7M-Gen-Llama3_1-70B	Q6_K_L	70.6B	54.39 GiB	10.63 GiB	66.14 GiB	0.82 GiB	12±22%
Athene-70B	Q6_K_L	70.6B	54.39 GiB	10.63 GiB	66.14 GiB	0.82 GiB	12±22%
Llama-4-Scout-17B-16E-InstructMoEKV unresolved	Q4_0	109B	58.72 GiB	6.38 GiB	66.13 GiB	0.83 GiB	32±37%
MiniMax-M2.1-REAP-139B-A10BMoE	I1-IQ3_M	139B	56.81 GiB	8.23 GiB	66.03 GiB	0.93 GiB	30±37%
m51Lab-MiniMax-M2.7-REAP-139B-A10BMoE	I1-IQ3_M	139B	56.81 GiB	8.23 GiB	66.03 GiB	0.93 GiB	30±37%
step-3.5-flash	IQ2_S	199B	51.70 GiB	13.30 GiB	66.03 GiB	0.93 GiB	12±22%
WizardLM-Uncensored-SuperCOT-StoryTelling-30b	Q3_K_S	32.5B	13.10 GiB	51.80 GiB	65.97 GiB	0.99 GiB	12±22%
Qwen3-VL-235B-A22B-ThinkingMoE	UD-IQ1_S	236B	58.65 GiB	6.24 GiB	65.92 GiB	1.04 GiB	33±37%
Mistral-Medium-3.5-128B	Q3_K_S	128B	53.05 GiB	11.69 GiB	65.89 GiB	1.07 GiB	12±22%
MiniMax-M2.7MoE	IQ2_XXS	229B	56.67 GiB	8.23 GiB	65.89 GiB	1.07 GiB	33±37%
Mixtral-8x22B-Instruct-v0.1MoE	Q3_K_S	141B	57.28 GiB	7.44 GiB	65.78 GiB	1.18 GiB	18±37%
Mixtral-8x22B-v0.1MoE	Q3_K_S	141B	57.28 GiB	7.44 GiB	65.78 GiB	1.18 GiB	18±37%
Mixtral-8x22B-v0.1MoE	Q3_K_S	141B	57.27 GiB	7.44 GiB	65.77 GiB	1.19 GiB	18±37%
Qwen3-VL-235B-A22B-InstructMoE	UD-IQ1_S	236B	58.49 GiB	6.24 GiB	65.77 GiB	1.19 GiB	33±37%
Phi-3-mini-4k-instructKV unresolved	F32	3.8B	52.01 GiB	12.75 GiB	65.76 GiB	1.20 GiB	12±22%
Apertus-70B-Instruct-2509	Q6_K	70.6B	53.95 GiB	10.63 GiB	65.75 GiB	1.21 GiB	12±22%
Devstral-2-123B-Instruct-2512	IQ3_M	125B	52.89 GiB	11.69 GiB	65.74 GiB	1.22 GiB	12±22%

●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?: 2063 of 2118 indexed open-weight models fit a RTX PRO 5000 Blackwell at 65,536 context with q8_0 KV cache, the largest being command-r-35b-writer-v2 at I1-Q5_K_M. 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.