Will It Run AI

NVIDIA

RTX PRO 6000 Blackwell Server Edition 96GB

RTX PRO BlackwellDatacenterBlackwellPCIe 5CUDA
96GB
VRAM
1.6kGB/s
Bandwidth
120TFLOPS
FP16 Compute
4kTOPS
INT8 Inference
$9,999 MSRP
VRAM96 GBBandwidth1.6k GB/sCompute120 TFInference4k TOPSValue1.2 TF/$k
RTX PRO 6000 Blackwell Server Edition 96GBCategory AvgNVIDIA H200 141GB

Operating mode

Choose the operating mode for this hardware

Use this to bias workload recommendations toward responsiveness, background autonomy, lighter serving, or multi-GPU scale-out.

Current mode

Balanced

Balanced for general local use. Keeps the ranking neutral across personal and serving workflows.

About this GPU for AI

La RTX PRO 6000 Blackwell Server Edition es el acelerador de estación de trabajo más capaz de NVIDIA, con 96 GB de GDDR7 en arquitectura Blackwell para cargas de trabajo profesionales de IA y visualización. Sus 120 TFLOPS de cómputo FP16 y 1.597 GB/s de ancho de banda lo hacen adecuado para ejecutar modelos de 70B en Q4 con margen. Como tarjeta PCIe 5.0, cabe en plataformas de servidor estándar sin los requisitos de infraestructura de los sistemas SXM o NVLink.

Beyond LLMs

AI Capability Matrix

What AI tasks this GPU can handle — from text generation to image and video creation.

CapabilityStatusRepresentative Model
LLM Chat (7B)Runs nativelyLlama 3.1 8B Q4
LLM Coding (30B)Runs nativelyQwen 3 30B Q4
LLM Large (70B)Runs nativelyLlama 3.1 70B Q4
Image Gen (SDXL)Runs nativelySDXL 1.0 FP16
Image Gen (Flux)Runs nativelyFlux.1 Dev FP16
Image Gen (SD 3.5)Runs nativelySD 3.5 Large FP16
Video Short (25f)Runs nativelyLTX Video 2B
Video Long (100f)Tight fitWan Video 14B
massive-vramblackwell-architecturepcie-form-factorprofessional-grade

Especificaciones

Cómputo
FP16120 TFLOPS
INT84000 TOPS
ArquitecturaBlackwell
Memoria
VRAM96 GB
Ancho de banda1597 GB/s
General
FamiliaRTX PRO Blackwell
SegmentoDatacenter
InterconexiónPCIe 5
Plataforma de cómputoCUDA
MSRP$9,999

Características clave

96 GB GDDR7 VRAM — largest capacity in the RTX PRO lineupBlackwell architecture with 4th-gen Tensor Cores (FP4/FP8/FP16)1,597 GB/s memory bandwidth120 TFLOPS FP16 / 4,000 INT8 TOPSPCIe 5.0 x16 — drop-in for modern server platforms300W TDP class — single-slot power budget

Para cargas de trabajo de IA

Fortalezas
  • 96 GB VRAM fits 70B models at Q4 and 34B models at FP16 on a single card
  • Blackwell Tensor Cores with FP4 support deliver strong inference throughput per watt
  • Standard PCIe 5.0 form factor works in any modern server — no proprietary baseboard needed
  • Commercially available at a fraction of H100/A100 pricing
Consideraciones
  • GDDR7 bandwidth (1,597 GB/s) significantly below HBM-based datacenter GPUs like A100/H100
  • No NVLink support limits multi-GPU scaling to PCIe peer-to-peer speeds
  • Not suited for large-scale distributed training across GPU clusters
  • ~$10K price point still steep for individual researchers or small teams

Architecture

Blackwell

Blackwell is NVIDIA's fifth-generation RTX architecture, built on TSMC's 4NP process. It introduces 5th-generation Tensor Cores with native FP4 precision support, enabling double the inference throughput per watt compared to Ada Lovelace's FP8 operations. Key innovations include the Neural Rendering Pipeline for AI-driven shading and the debut of GDDR7 memory in consumer GPUs.

AI Relevance

FP4 Tensor Cores deliver the highest tokens-per-watt efficiency in any consumer architecture. Native FP4 quantization means models can run at lower precision with minimal quality loss, effectively doubling the effective VRAM for model weights.

Process: TSMC 4NPPlatform: CUDATensor Cores: Gen 5Precisions: FP32, FP16, BF16, FP8, FP4, INT8, INT4

Consejo de compra

¿Deberías comprar RTX PRO 6000 Blackwell Server Edition 96GB para IA local?

Excelente opción para IA local

Ejecuta 36 de 50 modelos principales bien — un todoterreno sólido para inferencia local.

96.0 GB

VRAM

$9,999

PVP

$104/GB

Coste por GB de VRAM

Mejores modelos para esta GPU

What will limit you first

Este setup está bastante equilibrado para este modelo.

No hay grandes señales de alerta

Esta recomendación tiene margen de memoria suficiente y una velocidad estimada razonable para la carga de trabajo seleccionada.

Best upgrade itinerary

Desbloquea 2 modelos adicionales que hoy no caben en tu setup.

¿Quieres más margen? NVIDIA H200 141GB (141.0 GB VRAM) es el siguiente paso.

Recommendations by Workload

Chat

S

Qwen3-Coder-Next

This model is still usable for chat, but it is not the most specialized pick. It belongs to a current frontier family for local AI. It fits natively with comfortable headroom. Known channels: huggingface, ollama, lm-studio.

Decode 90.6 tok/s · 256K ctx · llama.cppEST.
60.0 GB / 96.0 GB VRAM

Coding

S

Qwen3-Coder-Next

This model is a direct match for coding. It belongs to a current frontier family for local AI. It fits natively with comfortable headroom. Known channels: huggingface, ollama, lm-studio.

Decode 90.6 tok/s · 256K ctx · llama.cppEST.
60.8 GB / 96.0 GB VRAM

Agentic Coding

S

Qwen3-Coder-Next

This model is still usable for agentic-coding, but it is not the most specialized pick. It belongs to a current frontier family for local AI. It fits natively with comfortable headroom. Known channels: huggingface, ollama, lm-studio.

Decode 90.6 tok/s · 256K ctx · llama.cppEST.
62.2 GB / 96.0 GB VRAM

Reasoning

S

Qwen3-Coder-Next

This model is a direct match for reasoning. It belongs to a current frontier family for local AI. It fits natively with comfortable headroom. Known channels: huggingface, ollama, lm-studio.

Decode 90.6 tok/s · 256K ctx · llama.cppEST.
60.8 GB / 96.0 GB VRAM

RAG

S

Qwen 3.5 27B

This model is a direct match for rag. It belongs to a current frontier family for local AI. It fits natively with comfortable headroom. Known channels: huggingface, ollama, lm-studio.

Decode 88.0 tok/s · 131K ctx · llama.cppEST.
33.3 GB / 96.0 GB VRAM

Full Model Compatibility

AlibabaQwen3-Coder-Next
S95
80B60.8 GB91 tok/s256K ctx
moe
AlibabaQwen 3.5 122B A10B
S95
122B87.4 GB54 tok/s73K ctx
moe
MistralMistral Small 4 119B
S93
119B88.5 GB59 tok/s38K ctx
moe
AlibabaQwen 2.5 VL 72B
S92
72B59.3 GB33 tok/s33K ctx
dense
AlibabaQwen 3.6 35B A3B
S92
35B36.0 GB171 tok/s250K ctx
+1moe
MistralDevstral 2 123B Instruct
S92
123B90.9 GB19 tok/s31K ctx
dense
AlibabaQwen3-Coder 30B A3B Instruct
S92
30.5B30.6 GB203 tok/s256K ctx
moe
AlibabaQwen 3.5 27B
S91
27B30.1 GB88 tok/s131K ctx
dense
AlibabaQwen3-VL 30B A3B Instruct
S91
30B30.3 GB210 tok/s256K ctx
moe
AlibabaQwen 3.5 35B A3B
S91
35B33.3 GB185 tok/s131K ctx
moe
AlibabaQwen 3 32B
S90
32B33.9 GB75 tok/s131K ctx
dense
CohereCommand A 111B
S90
111B82.1 GB22 tok/s73K ctx
dense
MistralMagistral Small 2507
S89
24B27.6 GB99 tok/s131K ctx
dense
OpenAIGPT-OSS 120B
S89
117B86.8 GB20 tok/s46K ctx
dense
MistralDevstral Small 2 24B Instruct
S89
24B27.6 GB99 tok/s256K ctx
dense
AlibabaQwen 3 30B A3B
S89
30.5B30.6 GB203 tok/s131K ctx
moe
AlibabaQwen 3.6 27B
S89
27B27.9 GB55 tok/s262K ctx
+1dense
NVIDIANemotron 3 Nano 30B
S88
30B31.2 GB79 tok/s131K ctx
dense
Mistral AIPixtral Large 124B
S88
124B91.5 GB19 tok/s29K ctx
dense
MistralLeanstral 119B A6B
S88
119B91.9 GB54 tok/s24K ctx
moe
MistralDevstral Small 1.1
S88
24B27.6 GB99 tok/s131K ctx
dense
AlibabaQwen 3.5 9B
S88
9B18.2 GB126 tok/s131K ctx
dense
AlibabaQwen 3 14B
S87
14B21.5 GB170 tok/s131K ctx
dense
NVIDIANemotron Cascade 2 30B A3B
S87
30B31.7 GB207 tok/s262K ctx
moe
MicrosoftPhi-4-reasoning-plus 14B
S87
14.7B22.5 GB161 tok/s33K ctx
dense
GoogleGemma 4 31B
S86
30.7B43.9 GB47 tok/s73K ctx
dense
OpenAIGPT-OSS 20B
S86
21B25.8 GB258 tok/s128K ctx
moe
AlibabaQwen 3 8B
S86
8B17.6 GB112 tok/s131K ctx
dense
LG AIEXAONE 4.0 32B
A84
32B33.9 GB74 tok/s131K ctx
dense
AlibabaQwen 3.5 4B
A84
4B15.1 GB56 tok/s131K ctx
dense
GoogleGemma 4 26B A4B
A83
25.2B29.5 GB218 tok/s256K ctx
moe
MistralMinistral 3 14B
A82
14B21.5 GB169 tok/s262K ctx
multimodal
NVIDIANemotron Nano 8B
A81
8B17.3 GB112 tok/s131K ctx
dense
MicrosoftPhi-4 Mini Reasoning 4B
A80
3.8B14.3 GB53 tok/s131K ctx
dense
Jina AIJina Embeddings v3
A74
0.57B13.6 GB8 tok/s8K ctx
dense
BAAIBGE M3
A73
0.57B12.8 GB8 tok/s8K ctx
dense
AlibabaQwen 3.5 397B A17B
F0
397B255.5 GB3 tok/s4K ctx
moe
Moonshot AIKimi K2.5
F0
1000B627.9 GB2 tok/s4K ctx
moe
Moonshot AIKimi K2.6
F0
1000B627.9 GB2 tok/s4K ctx
+1moe
DeepSeekDeepSeek V4 Pro
F0
1600B874.4 GB2 tok/s4K ctx
moe
DeepSeekDeepSeek V4 Flash
F0
284B169.8 GB7 tok/s4K ctx
moe
Z.aiGLM-5.1
F0
754B489.5 GB2 tok/s4K ctx
moe
Z.aiGLM-5
F0
744B483.4 GB2 tok/s4K ctx
moe
DeepSeekDeepSeek V3.2
F0
671B420.3 GB2 tok/s4K ctx
moe
AlibabaQwen 3 235B A22B
F0
235B156.7 GB8 tok/s4K ctx
moe
AlibabaQwen3-Coder 480B A35B Instruct
F0
480B306.2 GB2 tok/s4K ctx
moe
MiniMax M2.7
F0
230B154.6 GB9 tok/s4K ctx
moe
DeepSeekDeepSeek Coder V2 236B
F0
236B213.1 GB4 tok/s4K ctx
moe
DeepSeekDeepSeek R1 671B
F0
671B479.4 GB2 tok/s4K ctx
moe
DeepSeekDeepSeek V3.1 671B
F0
671B479.4 GB2 tok/s4K ctx
moe

Casi al alcance

Modelos que podrías ejecutar con una mejora

Modelos de alta calidad que necesitan un poco más de memoria

Image & Video Generation

Diffusion Model Compatibility

51 of 52 models can generate images or video on your RTX PRO 6000 Blackwell Server Edition 96GB

ModelMax ResolutionGen TimeGrade
SD TurboImage512×512300msS
Stable Diffusion 1.5Image512×768600msS
Realistic Vision v5.1Image512×768600msS
DreamShaper 8Image512×768600msS
LCM DreamShaper v7Image512×768200msS
PixArt-SigmaImage1024×1024~2.5sS
FramePack I2VVideo1280×720~4.5s/frameS
SDXL TurboImage512×512300msS
SDXL LightningImage1024×1024900msS
Stable Diffusion XL 1.0Image1024×1024~2.5sS
Playground v2.5Image1024×1024~3.7sS
RealVisXL v5.0Image1024×1024~2.8sS
DreamShaper XLImage1024×1024~2.8sS
Juggernaut XL v9Image1024×1024~2.8sS
Animagine XL 3.1Image1024×1024~2.8sS
Pony Diffusion V6 XLImage1024×1024~2.8sS
Animagine XL 4.0Image1024×1024~2.8sS
Illustrious XLImage1024×1024~2.8sS
Wan Video 2.1 1.3BVideo480×832~1.8s/frameS
Stable Diffusion 3.5 MediumImage1024×1024~4.3sS
Flux.2 Klein 4BImage1024×1024700msS
LTX Video 2BVideo1280×720~2.1s/frameS
KolorsImage1024×1024~4.9sS
Stable CascadeImage1024×1024~6.2sS
AuraFlow v0.3Image1536×1536~11.1sS
Stable Diffusion 3.5 LargeImage1024×1024~13.6sS
Stable Diffusion 3.5 Large TurboImage1024×1024~2.5sS
CogVideoX 2BVideo720×480~2.1s/frameS
HunyuanVideoVideo720×1280~4.5s/frameS
ChromaImage1024×1024~2.5sS
Z-Image TurboImage1536×1536~2.6sS
Flux.1 DevImage1024×1024~11.1sS
Flux.1 SchnellImage1024×1024~2.2sS
LTX Video 13BVideo1280×720~4.5s/frameS
Flux.1 Kontext DevImage1024×1024~12.4sS
AnimateDiff v1.5.3Video512×768~1.1s/frameS
Cosmos Diffusion 7BVideo1024×576~3.5s/frameS
CogVideoX 5BVideo720×480~3.1s/frameS
Wan2.2 TI2V 5BVideo832×480~3.1s/frameS
Flux.2 Klein 9BImage1024×1024~1.2sS
Flux.1 Fill DevImage1024×1024~10.5sS
Mochi 1 PreviewVideo848×480~4.1s/frameS
HunyuanVideo 1.5Video720×1280~3.8s/frameS
Helios 14BVideo1280×720~4.7s/frameS
SkyReels V2 14BVideo1280×720~4.7s/frameS
Wan Video 2.1 14BVideo720×1280~4.7s/frameS
Wan Video 2.2 14BVideo720×1280~4.7s/frameS
Qwen ImageImage1024×1024~4.2sS
Qwen Image EditImage1024×1024~4.2sS
Flux.2 DevImage1024×1024~1m 57sS
MAGI-1Video1280×720~5.8s/frameS
HunyuanImage 3.0Image256×256~7.3sF

Image models estimated at 1024×1024 (28 steps, FP16). Video models estimated at 768×512 (25 frames, 30 steps, FP16). Actual performance varies with runtime and system load.

Upgrade paths

Upgrade from RTX PRO 6000 Blackwell Server Edition 96GB

See what you unlock with more powerful hardware

Opciones de mejora

Opciones de mejora

NVIDIANVIDIA H200 141GBSiguiente nivel
141 GB VRAM (+45)4800 GB/s (+3203)
B
Desbloquea 2 modelos adicionales que hoy no caben en tu setup.Desbloquea Qwen 3 235B A22B, MiniMax M2.7+55% más rápido promedio

Desbloquea 2 modelos adicionales que hoy no caben en tu setup.

Eleva la velocidad media de decodificación en torno a un 55% en los modelos que sí caben.

~$30,000 MSRP

NVIDIANVIDIA B200 180GBMejora NVIDIA
180 GB VRAM (+84)8000 GB/s (+6403)
B
Desbloquea 8 modelos adicionales que hoy no caben en tu setup.Desbloquea DeepSeek V4 Flash, Qwen 3 235B A22B, MiniMax M2.7+5 más · +89% más rápido promedio

Desbloquea 8 modelos adicionales que hoy no caben en tu setup.

Eleva la velocidad media de decodificación en torno a un 89% en los modelos que sí caben.

~$30,000 MSRP

AMD Instinct MI325X 256GBMayor salto
256 GB VRAM (+160)6000 GB/s (+4403)
B
Desbloquea 12 modelos adicionales que hoy no caben en tu setup.Desbloquea Qwen 3.5 397B A17B, DeepSeek V4 Flash, Qwen 3 235B A22B+9 más · +58% más rápido promedio

Desbloquea 12 modelos adicionales que hoy no caben en tu setup.

Eleva la velocidad media de decodificación en torno a un 58% en los modelos que sí caben.

~$20,000 MSRP

AMD Instinct MI350X 288GBMejor relación calidad-precio
288 GB VRAM (+192)8000 GB/s (+6403)
B
Desbloquea 13 modelos adicionales que hoy no caben en tu setup.Desbloquea Qwen 3.5 397B A17B, DeepSeek V4 Flash, Qwen 3 235B A22B+10 más · +77% más rápido promedio

Desbloquea 13 modelos adicionales que hoy no caben en tu setup.

Eleva la velocidad media de decodificación en torno a un 77% en los modelos que sí caben.

~$8,000 MSRP

Frequently Asked Questions

What AI models can I run on RTX PRO 6000 Blackwell Server Edition 96GB?

RTX PRO 6000 Blackwell Server Edition 96GB (96 GB VRAM) can run these top models: Qwen3-Coder-Next (score: 95/100), Qwen 3.5 122B A10B (score: 95/100), Mistral Small 4 119B (score: 93/100). See the full compatibility list above.

How much VRAM does RTX PRO 6000 Blackwell Server Edition 96GB have for AI?

RTX PRO 6000 Blackwell Server Edition 96GB has 96 GB of VRAM available for AI model inference. This determines which models and quantization levels you can run locally.

Is RTX PRO 6000 Blackwell Server Edition 96GB good for running LLMs locally?

Yes, RTX PRO 6000 Blackwell Server Edition 96GB is excellent for running LLMs locally with top compatibility scores above 80/100.

What is the best model for RTX PRO 6000 Blackwell Server Edition 96GB for coding?

For coding on RTX PRO 6000 Blackwell Server Edition 96GB, we recommend Qwen3-Coder-Next. It achieves 90.6 tokens per second with 256K context window. This model is a direct match for coding. It belongs to a current frontier family for local AI. It fits natively with comfortable headroom. Known channels: huggingface, ollama, lm-studio.

Should I upgrade from RTX PRO 6000 Blackwell Server Edition 96GB?

There are 4 upgrade path(s) from RTX PRO 6000 Blackwell Server Edition 96GB: NVIDIA H200 141GB, NVIDIA B200 180GB. Upgrading would unlock larger models and faster inference speeds.

Can RTX PRO 6000 Blackwell Server Edition 96GB run Flux for image generation?

Yes, RTX PRO 6000 Blackwell Server Edition 96GB with 96 GB of usable memory can run Flux.1 Dev at FP16 natively. Flux is a 12B parameter diffusion transformer that produces high-quality images. You can also run the Schnell variant for faster generation.

What image and video AI models can I run on RTX PRO 6000 Blackwell Server Edition 96GB?

RTX PRO 6000 Blackwell Server Edition 96GB (96 GB VRAM) can handle various AI generation tasks beyond LLMs. For image generation, SDXL and Stable Diffusion 3.5 run well. Flux.1 Dev also runs natively for state-of-the-art image quality. For video, LTX Video 2.3 can generate short clips. Check the AI Capability Matrix above for detailed compatibility.

Is RTX PRO 6000 Blackwell Server Edition 96GB good for AI image generation?

RTX PRO 6000 Blackwell Server Edition 96GB is excellent for AI image generation. With 96 GB of usable memory, it runs all major diffusion models including Flux.1, SDXL, and Stable Diffusion 3.5 at full precision. You can generate high-resolution images quickly and even handle video generation models.

Can RTX PRO 6000 Blackwell Server Edition 96GB run Qwen 3.5 27B?

Yes, RTX PRO 6000 Blackwell Server Edition 96GB with 96 GB of usable memory can run Qwen 3.5 27B at Q8 (near-lossless, ~28.9 GB) or even FP16 (~55.4 GB) depending on your context needs. This setup provides an excellent experience with this model. Use Ollama or vLLM for best results.

What is the best quantization for AI models on RTX PRO 6000 Blackwell Server Edition 96GB?

With 96 GB VRAM on RTX PRO 6000 Blackwell Server Edition 96GB, use Q8_0 for most models — it is near-lossless and you have the memory for it. For 70B+ models, Q6_K offers excellent quality. Reserve Q4_K_M for 100B+ models or when you need maximum context length.

For local LLMs on RTX PRO 6000 Blackwell Server Edition 96GB, does VRAM matter more than bandwidth?

RTX PRO 6000 Blackwell Server Edition 96GB already has strong memory bandwidth, so the next limit is often memory capacity and context headroom rather than raw decode speed. For local LLMs, fit first and bandwidth second is the right mental model.

Compare with similar