NVIDIA
RTX PRO 6000 Blackwell Workstation Edition 96GB
RTX PRO BlackwellWorkstationBlackwellPCIe 5CUDA
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
The RTX PRO 6000 Blackwell Workstation Edition is the most powerful single workstation GPU ever built, featuring 96 GB of ECC GDDR7 at 1,792 GB/s bandwidth with 125 TFLOPS FP16 and 4,000 INT8 TOPS. Based on the GB202 die with 24,064 CUDA cores, it can run 70B models at FP16 on a single card and fits 100B+ models at Q4 โ previously achievable only with multi-GPU data-center setups. Available from April 2025, it also supports Universal MIG for partitioning the GPU into multiple isolated inference instances.
Beyond LLMs
AI Capability Matrix
What AI tasks this GPU can handle โ from text generation to image and video creation.
| Capability | Status | Representative Model | Detail |
|---|
| LLM Chat (7B) | Runs natively | Llama 3.1 8B Q4 | โ |
| LLM Coding (30B) | Runs natively | Qwen 3 30B Q4 | โ |
| LLM Large (70B) |
workstation-gradeecc-memorylarge-vramprofessional-certifiedblackwellmig-support
Specifications
Compute
FP16125 TFLOPS
INT84000 TOPS
ArchitectureBlackwell
Memory
VRAM96 GB
Bandwidth1792 GB/s
General
FamilyRTX PRO Blackwell
SegmentWorkstation
InterconnectPCIe 5
Compute PlatformCUDA
MSRP$9,999
Key Features
96 GB ECC GDDR7 VRAM on a 512-bit busBlackwell GB202 die: 24,064 CUDA cores, 752 Tensor Cores5th-gen Tensor Cores with FP4, FP8, FP16 precision125 TFLOPS FP16 / 4,000 INT8 TOPS1,792 GB/s memory bandwidthUniversal MIG support for isolated multi-tenant inferencePCIe 5.0 x16 with up to 600W TDP
For AI Workloads
Strengths
- 96 GB ECC VRAM fits 70B models at FP16 and 100B+ models at Q4 on a single workstation card
- 1,792 GB/s bandwidth delivers fast decode throughput even for unquantized 70B inference
- Universal MIG partitioning allows multiple isolated AI workloads on one GPU โ practical for shared workstation deployments
- 4,000 INT8 TOPS and FP4 support make it one of the most capable inference platforms outside dedicated data-center hardware
Considerations
- $9,999 MSRP is approaching data-center GPU territory โ the H100 SXM becomes competitive at scale
- 600W TDP requires workstation-class chassis and power delivery
- Massive overkill for models under 30B โ most of the VRAM advantage only matters for 70B+ deployments
- Single-card throughput for 70B FP16 is still slower than H100/H200 data-center cards despite the large VRAM
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
Recommendations by Workload
Qwen3-Coder-Next is viable for Chat, but is not the most specialized choice. It is a recent-generation family, which helps on current local SOTA workloads. It fits natively with comfortable headroom. Context coverage stays within the requested workload envelope. Known distribution channels: huggingface, ollama, lm-studio.
Decode 101.7 tok/s ยท 256K ctx ยท llama.cppEST.
Qwen3-Coder-Next is a specialized fit for Coding. It is a recent-generation family, which helps on current local SOTA workloads. It fits natively with comfortable headroom. Context coverage stays within the requested workload envelope. Known distribution channels: huggingface, ollama, lm-studio.
Decode 79.4 tok/s ยท 217K ctx ยท llama.cppEST.
Just out of reach
Models you could run with an upgrade
High-quality models that need a bit more memory
397BTier 100Needs ~254.1 GB
1000BTier 100Needs ~624.2 GB
1000BTier 100Needs ~624.2 GB
1600BTier 100Needs ~873.4 GB
284BTier 98Needs ~169.2 GB
Image & Video Generation
Diffusion Model Compatibility
51 of 52 models can generate images or video on your RTX PRO 6000 Blackwell Workstation Edition 96GB
Upgrade paths
Upgrade from RTX PRO 6000 Blackwell Workstation Edition 96GB
See what you unlock with more powerful hardware
Upgrade options
Upgrade options
141 GB VRAM (+45)4800 GB/s (+3008)
BUnlocks 2 additional models that do not fit on the current setup.Unlocks Qwen 3 235B A22B, MiniMax M2.7+47% faster avg
Unlocks 2 additional models that do not fit on the current setup.
Lifts average decode speed across fitting models by about 47%.
~$30,000 MSRP
180 GB VRAM (+84)8000 GB/s (+6208)
BUnlocks 8 additional models that do not fit on the current setup.Unlocks DeepSeek V4 Flash, Qwen 3 235B A22B, MiniMax M2.7+5 more ยท +80% faster avg
Unlocks 8 additional models that do not fit on the current setup.
Lifts average decode speed across fitting models by about 80%.
~$30,000 MSRP
AMD Instinct MI325X 256GBBiggest leap
256 GB VRAM (+160)6000 GB/s (+4208)
BUnlocks 12 additional models that do not fit on the current setup.Unlocks Qwen 3.5 397B A17B, DeepSeek V4 Flash, Qwen 3 235B A22B+9 more ยท +50% faster avg
Unlocks 12 additional models that do not fit on the current setup.
Lifts average decode speed across fitting models by about 50%.
~$20,000 MSRP
AMD Instinct MI350X 288GBBest value
288 GB VRAM (+192)8000 GB/s (+6208)
BUnlocks 13 additional models that do not fit on the current setup.Unlocks Qwen 3.5 397B A17B, DeepSeek V4 Flash, Qwen 3 235B A22B+10 more ยท +69% faster avg
Unlocks 13 additional models that do not fit on the current setup.
Lifts average decode speed across fitting models by about 69%.
~$8,000 MSRP
Frequently Asked Questions
RTX PRO 6000 Blackwell Workstation Edition 96GBCategory AvgNVIDIA H200 141GB
| Image Gen (SDXL) | Runs natively | SDXL 1.0 FP16 | ~~2.4s per image |
| Image Gen (Flux) | Runs natively | Flux.1 Dev FP16 | ~~10.7s per image |
| Image Gen (SD 3.5) | Runs natively | SD 3.5 Large FP16 | ~~13.1s per image |
| Video Short (25f) | Runs natively | LTX Video 2B | ~~2.1s/frame |
| Video Long (100f) | Tight fit | Wan Video 14B | ~~6.1s/frame |
S
Qwen3-Coder-Next is a specialized fit for Agentic Coding. It is a recent-generation family, which helps on current local SOTA workloads. It fits natively with comfortable headroom. Context coverage stays within the requested workload envelope. Known distribution channels: huggingface, ollama, lm-studio.
Decode 101.7 tok/s ยท 256K ctx ยท llama.cppEST.
Qwen3-Coder-Next matches Reasoning and keeps a practical fit profile. It is a recent-generation family, which helps on current local SOTA workloads. It fits natively with comfortable headroom. Context coverage stays within the requested workload envelope. Known distribution channels: huggingface, ollama, lm-studio.
Decode 79.4 tok/s ยท 217K ctx ยท llama.cppEST.
Qwen 3.5 27B matches RAG and keeps a practical fit profile. It is a recent-generation family, which helps on current local SOTA workloads. It fits natively with comfortable headroom. Context coverage stays within the requested workload envelope. Known distribution channels: huggingface, ollama, lm-studio.
Decode 61.9 tok/s ยท 131K ctx ยท llama.cppEST.
93
119B88.5 GB66 tok/s38K ctx
Image
| MAGI-1Video | 1280ร720 | ~5.6s/frame | S |
Image models estimated at 1024ร1024 (28 steps, FP16). Video models estimated at 768ร512 (25 frames, 30 steps, FP16). Actual performance varies with runtime and system load.
Buying advice
Should you buy RTX PRO 6000 Blackwell Workstation Edition 96GB for local AI?
Excellent choice for local AI
Runs 36 of 50 top models well โ a strong all-rounder for local inference.
What will limit you first
This setup is broadly balanced for this model.
No major red flags
This recommendation has enough memory headroom and acceptable estimated speed for the selected workload.
Best upgrade itinerary
Unlocks 2 additional models that do not fit on the current setup.
Want more headroom? NVIDIA H200 141GB (141.0 GB VRAM) is the next step up.