NVIDIA RTX PRO 6000 Blackwell Workstation Edition, AI & Machine Learning Benchmarks & Specs

96GB · AI Score 53.1/100 · first-party measured on 12 AI workloads

53.1 AI Score ✓ Measured

Every number on this page is first-party: NVIDIA RTX PRO 6000 Blackwell Workstation Edition was run on our pinned 12-workload AI suite on 2026-07-11, with under 0.5% run-to-run variance. On Llama 3.1 8B (Q4_K_M) NVIDIA RTX PRO 6000 Blackwell Workstation Edition delivers about 258.37 tokens/sec. Stepping up to Qwen3 32B it holds roughly 70.13 tok/s. The full Llama 3.3 70B still runs, at about 34.87 tok/s. For image generation, SDXL runs at 13.97 it/s, and FLUX.1-dev at 3.23 it/s. All 12 workloads fit in 96GB. There is no model in our suite this card has to turn down. NVIDIA RTX PRO 6000 Blackwell Workstation Edition isn't a retail purchase for most people. It's rented by the hour. You can run this exact card on RunPod.

AI & Machine Learning benchmark results

Text Generation tok/s 123

SmolLM2-135M1252.77

Qwen1.5-0.5B1025.58

LFM2.5-1.2B989.91

gemma-3-270m965

Qwen2.5-Coder-0.5B910.13

Qwen2.5-0.5B905.27

Llama 3.2 1B892.74

Qwen3-0.6B806.77

Qwen3 0.6B788.97

Qwen2.5-Coder-1.5B650.4

Qwen2-1.5B648.49

Qwen2.5-1.5B636.39

Workload	Result	Telemetry	Data
SmolLM2-135M	1252.77 tok/s	93 W46°CQ4_K_M	✓ Measured
Qwen1.5-0.5B	1025.58 tok/s	103 W46°CQ4_K_M	✓ Measured
LFM2.5-1.2B	989.91 tok/s	125 W53°CQ4_K_M	✓ Measured
gemma-3-270m	965 tok/s	97 W43°CQ4_K_M	✓ Measured
Qwen2.5-Coder-0.5B	910.13 tok/s	104 W45°CQ4_K_M	✓ Measured
Qwen2.5-0.5B	905.27 tok/s	112 W43°CQ4_K_M	✓ Measured
Llama 3.2 1B	892.74 tok/s	140 W38°CQ4_K_M	✓ Measured
Qwen3-0.6B	806.77 tok/s	100 W44°CQ4_K_M	✓ Measured
Qwen3 0.6B	788.97 tok/s	62 W36°CQ4_K_M	✓ Measured
Qwen2.5-Coder-1.5B	650.4 tok/s	113 W54°CQ4_K_M	✓ Measured
Qwen2-1.5B	648.49 tok/s	117 W48°CQ4_K_M	✓ Measured
Qwen2.5-1.5B	636.39 tok/s	115 W44°CQ4_K_M	✓ Measured
DeepSeek-R1 Distill 1.5B	635.1 tok/s	169 W45°CQ4_K_M	✓ Measured
LFM2.5-8B-A1B	617.09 tok/s	129 W48°CQ4_K_M	✓ Measured
Qwen3-1.7B	593.5 tok/s	143 W52°CQ4_K_M	✓ Measured
Qwen3 1.7B	575.77 tok/s	105 W45°CQ4_K_M	✓ Measured
gemma-3-1b	567.1 tok/s	128 W44°CQ4_K_M	✓ Measured
Llama-3.2-3B-Instruct-uncensored	440.15 tok/s	145 W54°CQ4_K_M	✓ Measured
Hermes-3-Llama-3.2-3B	439.65 tok/s	165 W49°CQ4_K_M	✓ Measured
Llama 3.2 3B	435.63 tok/s	167 W47°CQ4_K_M	✓ Measured
gemma-2-2b	417.07 tok/s	138 W47°CQ4_K_M	✓ Measured
gemma-2-2b-it-abliterated	416.68 tok/s	143 W52°CQ4_K_M	✓ Measured
phi-2	410.12 tok/s	164 W49°CQ4_K_M	✓ Measured
SmolLM3-3B	409.73 tok/s	140 W52°CQ4_K_M	✓ Measured
SmolLM3 3B	406.45 tok/s	176 W46°CQ4_K_M	✓ Measured
Qwen2.5-Coder-3B	401.24 tok/s	155 W55°CQ4_K_M	✓ Measured
AI21-Jamba-Reasoning-3B	400.02 tok/s	161 W48°CQ4_K_M	✓ Measured
Qwen2.5-3B	398.72 tok/s	153 W45°CQ4_K_M	✓ Measured
Phi-4-mini	393.81 tok/s	194 W56°CQ4_K_M	✓ Measured
Phi-4 Mini 3.8B	391.04 tok/s	172 W48°CQ4_K_M	✓ Measured
gpt-oss-20b	375.09 tok/s	153 W44°CQ4_K_M	✓ Measured
Qwen3 4B	362.81 tok/s	3.3 GB peak93 W39°C3.9 tok/WQ4_K_M	✓ Measured
Phi-3.5-mini	362.79 tok/s	180 W46°CQ4_K_M	✓ Measured
DeepSeek-Coder-V2-Lite	352.52 tok/s	155 W52°CQ4_K_M	✓ Measured
Qwen3-4B-Instruct-2507	334.71 tok/s	179 W51°CQ4_K_M	✓ Measured
Qwen3-4B-Thinking-2507	334.65 tok/s	192 W49°CQ4_K_M	✓ Measured
Qwen3-4B-Instruct-2507	333.54 tok/s	193 W52°CQ4_K_M	✓ Measured
Nemotron-3-Nano-30B-A3B	325.43 tok/s	138 W50°CQ4_K_M	✓ Measured
Qwen3-Coder 30B A3B	317.68 tok/s	111 W45°CQ4_K_M	✓ Measured
Qwen3-30B-A3B	311.76 tok/s	145 W51°CQ4_K_M	✓ Measured
Qwen3 30B A3B (Q3_K_M)	309.61 tok/s	145 W52°CQ3_K_M	✓ Measured
Qwen3 30B A3B	308.74 tok/s	132 W47°CQ4_K_M	✓ Measured
Gemma 3 4B	297.73 tok/s	169 W46°CQ4_K_M	✓ Measured
Dolphin X1 Trinity Nano 6B	282.02 tok/s	124 W39°CQ4_K_M	✓ Measured
Llama-2-7B	263.84 tok/s	241 W56°CQ4_K_M	✓ Measured
Llama 3.1 8B	258.37 tok/s	5.2 GB peak241 W43°C1.07 tok/WQ4_K_M	✓ Measured
Qwen2.5-Coder-7B-Instruct-abliterated	257.07 tok/s	204 W50°CQ4_K_M	✓ Measured
Qwen2.5-7B	256.63 tok/s	228 W47°CQ4_K_M	✓ Measured
Qwen2.5-Coder 7B	256.11 tok/s	195 W49°CQ4_K_M	✓ Measured
DeepSeek-R1 Distill 7B	256.04 tok/s	169 W50°CQ4_K_M	✓ Measured
KAT-Coder-V2.5-Dev	253.35 tok/s	152 W45°CQ4_K_M	✓ Measured
Mistral-7B-Instruct-v0.2	253.34 tok/s	244 W56°CQ4_K_M	✓ Measured
Mistral-7B-Instruct-v0.3	253.2 tok/s	222 W53°CQ4_K_M	✓ Measured
Mistral-7B-Instruct-v0.1	252.55 tok/s	238 W55°CQ4_K_M	✓ Measured
Mistral 7B v0.3	247.33 tok/s	175 W43°CQ4_K_M	✓ Measured
L3-8B-Stheno-v3.2	239.84 tok/s	224 W52°CQ4_K_M	✓ Measured
DarkIdol-Llama-3.1-8B-Instruct-1.2-Uncensored	239.45 tok/s	227 W55°CQ4_K_M	✓ Measured
Dolphin X1 8B	239.4 tok/s	202 W51°CQ4_K_M	✓ Measured
dolphin-2.9-llama3-8b	239.29 tok/s	217 W53°CQ4_K_M	✓ Measured
Dolphin 3.0 Llama 3.1 8B	239.28 tok/s	179 W50°CQ4_K_M	✓ Measured
Ornith-1.0-35B	237.7 tok/s	155 W45°CQ4_K_M	✓ Measured
Meta-Llama-3.1-8B	237.69 tok/s	213 W46°CQ4_K_M	✓ Measured
DeepSeek-R1 Distill Llama 8B	235.07 tok/s	154 W44°CQ4_K_M	✓ Measured
Qwen-AgentWorld-35B-A3B	234.82 tok/s	162 W45°CQ4_K_M	✓ Measured
Josiefied-Qwen3-8B-abliterated-v1	227.15 tok/s	213 W51°CQ4_K_M	✓ Measured
Qwen3-8B	226.88 tok/s	223 W57°CQ4_K_M	✓ Measured
DeepSeek-R1-0528-Qwen3-8B	226.85 tok/s	228 W48°CQ4_K_M	✓ Measured
Qwen3 8B	225.95 tok/s	198 W49°CQ4_K_M	✓ Measured
GLM-4.7-Flash	215.61 tok/s	163 W46°CQ4_K_M	✓ Measured
Qwen3-Next-80B-A3B-Thinking	214.04 tok/s	135 W49°CQ4_K_M	✓ Measured
Qwen3-Coder-Next	211.5 tok/s	147 W46°CQ4_K_M	✓ Measured
Qwen3-Coder-Next-abliterated	210.47 tok/s	142 W47°CQ4_K_M	✓ Measured
Qwen3-Next-80B-A3B	210.17 tok/s	136 W48°CQ4_K_M	✓ Measured
Qwen3-Coder-Next	205.69 tok/s	145 W49°CQ4_K_M	✓ Measured
Qwen3-Next-80B-A3B-Thinking	203.29 tok/s	144 W47°CQ4_K_M	✓ Measured
Ornith-1.0-9B	201.69 tok/s	232 W49°CQ4_K_M	✓ Measured
GLM-4.7-Flash-REAP-23B-A3B	199.14 tok/s	156 W50°CQ4_K_M	✓ Measured
Phi-4 14B (Q3_K_M)	165.43 tok/s	286 W55°CQ3_K_M	✓ Measured
Mistral-Nemo-Instruct-2407	164.28 tok/s	262 W52°CQ4_K_M	✓ Measured
NemoMix-Unleashed-12B	164.21 tok/s	252 W51°CQ4_K_M	✓ Measured
gemma-2-9b	156.29 tok/s	240 W55°CQ4_K_M	✓ Measured
Gemma 3 12B (Q3_K_M)	153.93 tok/s	257 W53°CQ3_K_M	✓ Measured
DeepSeek-R1 Distill 14B (Q3_K_M)	149.81 tok/s	272 W54°CQ3_K_M	✓ Measured
Qwen2.5-Coder 14B	147.74 tok/s	8.7 GB peak239 W47°C0.62 tok/WQ4_K_M	✓ Measured
Phi-4 14B	142.68 tok/s	220 W49°CQ4_K_M	✓ Measured
Hermes-4-14B	139.56 tok/s	277 W53°CQ4_K_M	✓ Measured
Qwen3-14B	139.45 tok/s	263 W52°CQ4_K_M	✓ Measured
Qwen3 14B	139.01 tok/s	175 W49°CQ4_K_M	✓ Measured
Gemma 3 12B	138.09 tok/s	189 W45°CQ4_K_M	✓ Measured
Gemma 4 12B	137.73 tok/s	201 W46°CQ4_K_M	✓ Measured
Qwen2.5-Coder-14B-Instruct-abliterated	135.88 tok/s	281 W52°CQ4_K_M	✓ Measured
EVA-Qwen2.5-14B-v0.2	135.83 tok/s	271 W55°CQ4_K_M	✓ Measured
Uncensored	135.68 tok/s	273 W53°CQ4_K_M	✓ Measured
Qwen2.5-14B	135.66 tok/s	275 W49°CQ4_K_M	✓ Measured
DeepSeek-R1 Distill 14B	135.41 tok/s	212 W49°CQ4_K_M	✓ Measured
StarCoder2 15B	122.09 tok/s	211 W52°CQ4_K_M	✓ Measured
Mistral Small 24B (Q3_K_M)	108.88 tok/s	334 W57°CQ3_K_M	✓ Measured
Codestral 22B (Q3_K_M)	107.67 tok/s	338 W58°CQ3_K_M	✓ Measured
Cydonia-24B-v4.3	93.76 tok/s	301 W54°CQ4_K_M	✓ Measured
Dolphin-Mistral-24B-Venice-Edition	93.71 tok/s	295 W53°CQ4_K_M	✓ Measured
Dolphin Mistral 24B Venice	93.67 tok/s	157 W53°CQ4_K_M	✓ Measured
Dolphin 3.0 R1 Mistral 24B	93.62 tok/s	199 W53°CQ4_K_M	✓ Measured
Mistral Small 24B	93.55 tok/s	230 W50°CQ4_K_M	✓ Measured
Devstral Small 24B	93.52 tok/s	136 W48°CQ4_K_M	✓ Measured
Codestral 22B	92.7 tok/s	150 W51°CQ4_K_M	✓ Measured
Qwen2.5-Coder 32B (Q3_K_M)	75.78 tok/s	351 W58°CQ3_K_M	✓ Measured
Gemma 3 27B	72.1 tok/s	226 W60°CQ4_K_M	✓ Measured
Qwen3 32B	70.13 tok/s	19.1 GB peak200 W51°C0.35 tok/WQ4_K_M	✓ Measured
Qwen2.5-Coder 32B	66.07 tok/s	239 W57°CQ4_K_M	✓ Measured
QwQ 32B	66.07 tok/s	218 W59°CQ4_K_M	✓ Measured
DeepSeek-R1 Distill 32B	66.06 tok/s	260 W61°CQ4_K_M	✓ Measured
DeepSeek-R1-Distill-Qwen-32B-abliterated	66.03 tok/s	320 W54°CQ4_K_M	✓ Measured
Qwen2.5-32B	66 tok/s	321 W53°CQ4_K_M	✓ Measured
Olmo-3.1-32B-Think	65.71 tok/s	321 W58°CQ4_K_M	✓ Measured
Dolphin 2.9.1 Yi 1.5 34B	63.74 tok/s	186 W55°CQ4_K_M	✓ Measured
Llama 3.3 70B	34.87 tok/s	40.1 GB peak193 W56°C0.18 tok/WQ4_K_M	✓ Measured
Llama-3.3-70B-Instruct-abliterated	32.53 tok/s	363 W59°CQ4_K_M	✓ Measured
DeepSeek-R1-Distill-Llama-70B	32.52 tok/s	358 W59°CQ4_K_M	✓ Measured
Meta-Llama-3.1-70B	32.52 tok/s	352 W58°CQ4_K_M	✓ Measured
Hermes-4-70B	32.51 tok/s	353 W57°CQ4_K_M	✓ Measured
Qwen2.5-72B	29.65 tok/s	356 W57°CQ4_K_M	✓ Measured
Laguna-XS-2.1	✕ Won't fit needs ~26 GB	VRAM-gated at this precision	✓ Measured
Nanbeige4.2-3B	✕ Won't fit needs ~4 GB	VRAM-gated at this precision	✓ Measured

Image Generation images/min 6

FLUX.1 Schnell42.41

Z-Image Turbo (1024px)31.43

Stable Diffusion XL27.94

Z-Image Turbo15.6

FLUX.1 dev6.921

Krea 2 Turbo4.71

Workload	Result	Telemetry	Data
FLUX.1 Schnell	42.41 images/min	503 W53°C	✓ Measured
Z-Image Turbo (1024px)	31.43 images/min	541 W62°C	✓ Measured
Stable Diffusion XL	27.94 images/min	14.8 GB peak575 W55°C2.2 s/img	✓ Measured
Z-Image Turbo	15.6 images/min	25.9 GB peak584 W57°C3.8 s/img	✓ Measured
FLUX.1 dev	6.92 images/min	36.8 GB peak597 W66°C8.7 s/img	✓ Measured
Krea 2 Turbo	4.71 images/min	491 W66°C	✓ Measured

Fine-Tuning train tok/s 4

TinyLlama 1.1B LoRA18556.4

SmolLM2 1.7B LoRA16255.4

Qwen2.5 1.5B LoRA15753.9

Qwen2.5 7B LoRA6041.8

Workload	Result	Telemetry	Data
TinyLlama 1.1B LoRA	18556.4 train tok/s	271 W44°C	✓ Measured
SmolLM2 1.7B LoRA	16255.4 train tok/s	330 W46°C	✓ Measured
Qwen2.5 1.5B LoRA	15753.9 train tok/s	332 W44°C	✓ Measured
Qwen2.5 7B LoRA	6041.8 train tok/s	513 W52°C	✓ Measured

LLM Serving serve tok/s 4

TinyLlama 1.1B served7678.7

Qwen2.5 1.5B served5786.7

SmolLM2 1.7B served5642

Qwen2.5 7B served2282.8

Workload	Result	Telemetry	Data
TinyLlama 1.1B served	7678.7 serve tok/s	175 W42°C	✓ Measured
Qwen2.5 1.5B served	5786.7 serve tok/s	233 W43°C	✓ Measured
SmolLM2 1.7B served	5642 serve tok/s	236 W42°C	✓ Measured
Qwen2.5 7B served	2282.8 serve tok/s	362 W46°C	✓ Measured

Image Editing images/min 2

Workload	Result	Telemetry	Data
FLUX.1 Kontext dev	3.09 images/min	37.8 GB peak598 W76°C19.5 s/img	✓ Measured
Qwen-Image-Edit	2.64 images/min	60.3 GB peak598 W74°C22.7 s/img	✓ Measured

Depth Estimation images/min 2

Workload	Result	Telemetry	Data
Depth Anything V2 Small	1374.58 images/min	88 W34°C	✓ Measured
Depth Anything V2 Large	1328.11 images/min	92 W35°C	✓ Measured

Segmentation images/min 2

Workload	Result	Telemetry	Data
SAM ViT-Base	1079.89 images/min	104 W38°C	✓ Measured
SAM ViT-Huge	193.47 images/min	298 W47°C	✓ Measured

Video Generation frames/s 1

Workload	Result	Telemetry	Data
LTX-Video (distilled)	16.36 frames/s	60.4 GB peak542 W63°C5.9 s/clip	✓ Measured

Background Removal images/min 1

Workload	Result	Telemetry	Data
BiRefNet	1445.92 images/min	100 W33°C	✓ Measured

Upscaling images/min 1

Workload	Result	Telemetry	Data
Swin2SR 4x Upscaler	52.83 images/min	370 W44°C	✓ Measured

How we measured this. Every result comes from our own pinned, reproducible AI suite, 12 workloads: the Qwen3-4B to Llama-70B LLM ladder (llama.cpp, Q4_K_M), SDXL / Z-Image / FLUX-dev generation, FLUX-Kontext / Qwen-Edit editing, and LTX / Wan video, run first-party on rented hardware with under 0.5% run-to-run variance. Peak VRAM, power draw, temperature and tokens-per-watt are captured per workload. “Won’t fit” rows are real data: where a model exceeds the card’s VRAM at the tested precision we record a hard gate rather than silently dropping to a smaller quant. Measured 2026-07-11 · harness 2.0.0-standalone.

NVIDIA RTX PRO 6000 Blackwell Workstation Edition specifications

Architecture	Blackwell
CUDA cores	24,064
VRAM	96GB GDDR7 ECC
Memory bus	512-bit
Memory bandwidth	1792 GB/s
Boost clock	2,617 MHz
TDP	600 W
Process	4nm (TSMC 4N)
Interface	PCIe 5.0 x16
Release date	2025-03-18
Launch MSRP	$8,565

Verdict, NVIDIA RTX PRO 6000 Blackwell Workstation Edition on real AI workloads

NVIDIA RTX PRO 6000 Blackwell Workstation Edition scores 53.1/100, #9 of 102. It ran all 12 workloads. Every figure here is our own measurement.

Relative performance: where the NVIDIA RTX PRO 6000 Blackwell Workstation Edition lands

100% = this card, AI & Machine Learning headline metric (AI Score). #1 of 20 desktop cards in this vertical.

GPU	%	AI Score
NVIDIA RTX PRO 6000 Blackwell Workstation Edition	100%	53.1
NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition	85%	45.3
NVIDIA RTX PRO 5000 Blackwell	60%	31.6
NVIDIA RTX A6000	40%	21
NVIDIA RTX PRO 4500 Blackwell	24%	12.9

One tier down: NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition (-15%)

← All AI & Machine Learning GPU rankings

What this card can build

Whole-job timings, composed from our measured per-model results on this card.

Workflow	Time	Energy	Basis
50-image depth pass	4 s	0.06 Wh	measured
500-image masking run	2.6 min	12.84 Wh	measured
24-frame storyboard	3.6 min	16.08 Wh	all 2 stages measured
200-product catalogue cutout	4 min	23.56 Wh	all 2 stages measured
60-second AI short film	4.5 min	19.26 Wh	all 3 stages measured
6-panel comic page	5.1 min	28.55 Wh	all 3 stages measured
Character sheet, 12 poses	6.1 min	40.17 Wh	all 2 stages measured
Full codebase review	6.8 min	27 Wh	measured
20 long-form articles	13.4 min	42.96 Wh	measured
40-product photo shoot	16.7 min	142.82 Wh	all 2 stages measured
40-product shoot, start to finish	17.6 min	147.54 Wh	all 4 stages measured
100-photo restoration batch	33.4 min	322.75 Wh	measured
100-photo restore and enlarge	35.3 min	334.41 Wh	all 2 stages measured