NVIDIA RTX PRO 6000 Blackwell 96GB GDDR7 workstation
With 24,064 CUDA cores and a massive 96GB of ECC GDDR7 memory on a 512-bit bus, this Blackwell architecture card delivers 126 FP32 TFLOPS for data-intensive single-GPU workflows. Its passive, double-flow-through cooling fits a 2-slot, 305mm form factor, enabling dense multi-GPU server deployments without throttling. This card is best for AI developers training large language models and engineers running complex real-time simulations that require error-corrected memory integrity.
현재 판매처 최신 가격 없음
요약
The 30-Second Version
The RTX PRO 6000 Blackwell is the fastest workstation GPU we've ever tracked, with a PassMark G3D score of 39,287 and 96GB of ECC GDDR7. It's the 100th percentile for AI, ray tracing, and VRAM. But it costs $11,829 to $20,000 and draws 600W, making it a terrible value for anyone who doesn't absolutely need that much memory.
Pros & Cons
장점
- 96GB ECC GDDR7 is the largest VRAM pool we've ever tracked, 100th percentile 100th
- PassMark G3D score of 39,287 tops our entire database 100th
- 126 FP32 TFLOPS makes it a monster for AI and compute workloads 100th
- 1790 GB/s memory bandwidth keeps massive models running smoothly 100th
- ECC memory means data integrity for professional and scientific work
단점
- 600W TDP and passive cooling land it in the 15th percentile for thermals
- Price ranges from $11,829 to $20,000, the worst affordability score we've seen
- Only slight compute gains over the RTX 5090 according to one owner
- No power adapter included in OEM packaging
- Efficiency sits at the 25th percentile, so expect a real power bill
사용자 의견
The Word on the Street
시간에 따라 사용자 평판이 어떻게 변했는가
독점고객이 실제로 리뷰를 작성한 시점을 기준으로 합니다. 초기의 호평이 유지되었는지 확인할 수 있습니다.
날짜가 있는 고객 리뷰 9건을 기준으로 달력 분기별로 묶었습니다. 기간별 분석은 영어로 제공됩니다.
근거 자료
Performance
Let's talk numbers. The RTX PRO 6000 Blackwell sits at the 100th percentile for ray tracing, rasterization, AI, and VRAM in our database. That's not a typo. It's literally the best we've tested across every one of those categories. The 24,064 CUDA cores and 188 RT cores chew through workloads that would make a 5090 sweat, and the 126 FP16 TFLOPS means AI inference is absurdly fast. The 96GB of ECC GDDR7 on a 512-bit interface delivers 1790 GB/s of bandwidth, which is why local LLM users are reporting such smooth experiences with large models.
But the compute score tells a more nuanced story. At the 97th percentile, it's still elite, but one user noted only slight improvements over an RTX 5090 in certain compute tasks. That's not a knock on this card, it's a reality check on how good the 5090 already is. Where the PRO 6000 separates itself is memory capacity and ECC support. If your workload needs more than 32GB of VRAM, there's no comparison. The 2617MHz boost clock is solid, and PCIe 5.0 x16 ensures the card isn't bottlenecked by the interface. Just remember: 600W TDP means you need a 1000W PSU and a 16-pin connector that can actually deliver that power.
Specifications
Full Specifications
Chipset
| GPU Chipset | NVIDIA RTX PRO 6000 Blackwell |
| Brand | NVIDIA |
| Architecture | Blackwell |
| Class | workstation |
| Process | 5 |
Memory
| VRAM | 96 GB |
| Memory Type | GDDR7 |
| Memory Bus | 512 |
| Bandwidth | 1790 |
| ECC | Yes |
Compute
| CUDA Cores | 24064 |
| Tensor Cores | 752 |
| RT Cores | 188 |
| Base Clock | 1590 |
| Boost Clock | 2617 |
| FP16 TFLOPS | 126 |
| FP32 TFLOPS | 126 |
Performance
| PassMark G3D | 39287 |
Power & Cooling
| TDP | 600 |
| Recommended PSU | 1000 |
| Power Connectors | 1x 16-Pin PCI Connector |
| Cooler | Passive |
| Fans | 0 |
Connectivity
| Interface | PCI Express 5.0 x16 |
| DisplayPort | 4 |
| HDMI | 0 |
| Max Displays | 4 |
| Max Resolution | 7680 x 4320 |
Physical
| Length | 305 |
| Slot Width | 2 |
| Form Factor | Full Height |
| Weight | 2.8 kg / 6.1 lbs |
Features
| Ray Tracing | Yes |
| Upscaling | DLSS 4 |
| VR Ready | Yes |
| DirectX | 12 |
| RGB | No |
vs Competition
Against the ASUS TUF RTX 5090, the PRO 6000 wins on VRAM (96GB vs 32GB) and ECC support, but the 5090 is dramatically cheaper and nearly as fast in many compute tasks. The MSI Ventus RTX 5070 Ti and Gigabyte AERO RTX 5080 aren't even in the same league for memory capacity, but they're also a fraction of the price. The Sapphire NITRO+ and ZOTAC SOLID OC are gaming-focused cards that can't touch the PRO 6000's AI or workstation scores, but they'll run games just as well for a tenth of the cost. If you need 96GB of VRAM, there's no competition. If you don't, there's no reason to buy this.
| Spec | NVIDIA RTX PRO 6000 Blackwell | ASUS TUF Gaming TUF-RTX5090-32G-GAMING | MSI Ventus RTX 5070 Ti 16G VENTUS 3X PZ OC | Gigabyte AORUS GV-N5080AORUS M-16GD | ZOTAC APOCALYPSE ZT-B50800C-10P | Sapphire NITRO+ 11322-01-40G |
|---|---|---|---|---|---|---|
| Chipset | NVIDIA RTX PRO 6000 Blackwell | GeForce RTX 5090 | GeForce RTX 5070 Ti | GeForce RTX 5080 | GeForce RTX 5080 | Radeon RX 7900 XTX |
| Vram | 96GB GDDR7 | 32GB GDDR7 | 16GB GDDR7 | 16GB GDDR7 | 16GB GDDR7 | 24GB GDDR6 |
| Boost Clock Mhz | 2617 | 2407 | 2482 | 2805 | 2730 | 2680 |
| Tdp W | 600 | 575 | 300 | 360 | 360 | 339 |
| Memory Bus Bits | 512 | 512 | 256 | 256 | 256 | 384 |
| Ray Tracing | true | true | true | true | true | true |
| GPU Class | workstation | consumer | consumer | consumer | consumer | consumer |
| Compare | Compare | Compare | Compare | Compare |
| 제품 | AI | Rt | Vram | Raster | Compute | Feature | Thermal | User Sentiment | Efficiency | Connectivity | 사용자 평판 | Affordability |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| NVIDIA RTX PRO 6000 Blackwell | 99.8 | 99.9 | 99.8 | 99.9 | 97.3 | 86.7 | 14.5 | 45.5 | 24.5 | 75 | 65.6 | 0.2 |
| ASUS TUF Gaming TUF-RTX5090-32G-GAMING Compare | 96.7 | 90 | 97.4 | 98.6 | 97.4 | 96.4 | 72.8 | 86.5 | 26.8 | 99.2 | 99.1 | 3.4 |
| MSI Ventus RTX 5070 Ti 16G VENTUS 3X PZ OC Compare | 79 | 92.1 | 85.3 | 88.4 | 87.8 | 96.4 | 91.5 | 81 | 73.4 | 94.1 | 88.9 | 20.7 |
| Gigabyte AORUS GV-N5080AORUS M-16GD Compare | 85 | 95.2 | 88.5 | 94.1 | 93.1 | 96.4 | 91.5 | 72.3 | 58.9 | 94.1 | 88.9 | 15.5 |
| ZOTAC APOCALYPSE ZT-B50800C-10P Compare | 85 | 97.5 | 88.5 | 94.1 | 93.1 | 96.4 | 91.5 | 99.5 | 58.9 | 94.1 | 99.1 | 19.7 |
| Sapphire NITRO+ 11322-01-40G Compare | 94.7 | 96.2 | 91.5 | 83.6 | 95.4 | 84.4 | 91.5 | 86.5 | 54.5 | 98.7 | 78.4 | 15.2 |
가격
Value & Pricing
Here's where things get painful. The affordability score is 46.2/100, which is the weakest area of this card by a mile. In fact, it's the 0th percentile for affordability in our database. That means it's literally the worst value proposition we've ever tracked. The price spread is wild: $11,829 on the low end and $20,000 on the high end across four US listings. That's an $8,171 difference for what appears to be the same card. If you're buying, shop around aggressively. The $11,829 listing is the one to target. At $20,000, you're paying a massive premium for what might just be OEM packaging differences. For context, you could buy multiple RTX 5090s for that price, though you'd still be capped at 32GB of VRAM per card.
더 보기
Overview
The NVIDIA RTX PRO 6000 Blackwell is, on paper, the most powerful workstation GPU we've ever tracked. A PassMark G3D score of 39,287 puts it at the absolute top of our charts for raster, ray tracing, and AI workloads. With 24,064 CUDA cores, 96GB of ECC GDDR7 on a 512-bit bus, and 126 FP32 TFLOPS, this thing isn't just fast, it's in a different conversation entirely. The 93.3/100 AI score and 88.6/100 workstation score back that up. But here's the catch: it costs somewhere between $11,829 and $20,000 depending on where you look, and the efficiency and thermal numbers tell a very different story.
This is a card built for one thing: stuffing massive models into VRAM and letting them run. 96GB of memory means you can load LLMs that would choke a 32GB RTX 5090, and the 1790 GB/s bandwidth keeps tokens flowing. But the 600W TDP and passive cooling design mean you'll need serious airflow, and the 25th percentile efficiency ranking is a reminder that raw power has a real power bill attached. If you're rendering 8K video or training models locally, this is the card. If you're gaming, well, keep reading.
Common Questions
Q: Is the RTX PRO 6000 Blackwell worth it over an RTX 5090?
Only if you need more than 32GB of VRAM. The PRO 6000 has 96GB of ECC GDDR7, which is triple the 5090's capacity. But one owner reported only slight compute improvements over the 5090, and the PRO 6000 costs $11,829 to $20,000. For most users, the 5090 is the smarter buy.
Q: What power supply do I need for this card?
NVIDIA recommends a 1000W PSU. The card has a 600W TDP and uses a single 16-pin PCIe connector. It's passively cooled, so you'll also need strong case airflow to keep temperatures in check.
Q: Can I use this card for gaming?
You can, but it's a waste of money. The PRO 6000 is built for AI, rendering, and simulation workloads. Its gaming performance is excellent, but cards like the RTX 5090 deliver similar frame rates for a fraction of the price. Buy this for work, not play.
Who Should Skip This
Gamers should absolutely skip this. The RTX PRO 6000 Blackwell is not cost-effective for gaming, and the 46.2/100 budget score is the worst we've ever recorded. If you're running an RTX 5090 already, the compute gains here are marginal according to at least one owner. And if you don't have a 1000W PSU with strong case cooling, the 600W TDP and passive design will be a headache. This card is for AI researchers, 3D artists, and simulation engineers who need 96GB of ECC VRAM. Everyone else should buy something cheaper and put the savings toward a better monitor or a vacation.
Verdict
The RTX PRO 6000 Blackwell is the best workstation GPU we've ever tested, and the numbers prove it. 100th percentile across ray tracing, raster, AI, and VRAM. A PassMark score that tops our entire database. But the 600W power draw, passive cooling, and $11,829+ price tag mean this is a tool for professionals who can write it off as a business expense. The 45th percentile user sentiment score suggests even some buyers are lukewarm, mainly due to the cost and the fact that the RTX 5090 gets you most of the way there for a lot less money. If your workload demands 96GB of ECC VRAM, buy it. If not, save your money.