NVIDIA RTX PRO™ 5000 Blackwell Generation
PRODUCT DESCRIPTION
NVIDIA RTX PRO™ 5000 Blackwell Generation 48 GB GDDR7 with ECC (900-5G153-2550-000) - 3 Years Local Warranty
RTX PRO 5000 Blackwell 48 GB — 48 GB GDDR7 with ECC for Professional AI and Visualisation
The NVIDIA RTX PRO™ 5000 Blackwell pairs 48 GB GDDR7 with ECC with 1,344 GB/s of memory bandwidth across a 384-bit interface and 14,080 CUDA cores. It is the balance point of the RTX PRO Blackwell desktop range: enough VRAM for serious model work and large scenes, in a standard 300 W card that fits a normal workstation.
2,064 AI TOPS from Fifth-Generation Tensor Cores
Fifth-generation Tensor Cores deliver 2,064 AI TOPS (effective FP4 TOPS with sparsity, peak rates based on GPU Boost Clock), with 65 TFLOPS of FP32 compute and 196 TFLOPS of fourth-generation RT Core performance for ray tracing and neural rendering. 3x NVENC (ninth generation) and 3x NVDEC (sixth generation) handle encode and decode for video and streaming pipelines.
Multi-Instance GPU for Shared Workstations
MIG support lets this card be partitioned — up to 2x 24 GB or 1x 48 GB (maximum 2 instances) — so two users or two containerised workloads can share one GPU with isolated memory and compute rather than contending for it.
Standard Workstation Fit, Backed Locally
4.4in (H) x 10.5in (L), dual slot, full height with active cooling on a PCIe 5.0 x16 interface, powered through a 1x PCIe CEM5 16-pin connector. Supplied by SourceIT Pte Ltd in Singapore with a GST invoice and 3 years local warranty — supported here rather than through an overseas RMA queue.
NVIDIA RTX PRO™ 5000 Blackwell Datasheet — Technical Specifications
General
| Part Number | 900-5G153-2550-000 |
| Model | NVIDIA RTX PRO™ 5000 Blackwell |
| GPU Architecture | NVIDIA Blackwell |
| Product Family | NVIDIA RTX PRO Blackwell Workstation / Professional Desktop GPUs |
| Warranty | 3 Years Local Warranty (Singapore) |
Memory
| GPU Memory | 48 GB GDDR7 with ECC |
| Memory Interface | 384-bit |
| Memory Bandwidth | 1,344 GB/s |
| ECC | Yes — error-correcting code memory |
Compute & Performance
| CUDA Cores | 14,080 |
| Tensor Cores | Fifth-generation Tensor Cores |
| RT Cores | Fourth-generation RT Cores |
| Single-Precision (FP32) Performance | 65 TFLOPS |
| RT Core Performance | 196 TFLOPS |
| AI Performance | 2,064 AI TOPS (effective FP4 TOPS with sparsity, peak rates based on GPU Boost Clock) |
| Multi-Instance GPU (MIG) | Yes — up to 2x 24 GB or 1x 48 GB (maximum 2 instances) |
Media Engines
| Encoders (NVENC) | 3x NVENC (ninth generation) |
| Decoders (NVDEC) | 3x NVDEC (sixth generation) |
Display
| Display Connectors | 4x DisplayPort 2.1b |
| Maximum Simultaneous Displays | 4x 4096 x 2160 @ 120 Hz; 4x 5120 x 2880 @ 60 Hz; 2x 7680 x 4320 @ 60 Hz |
Power, Form Factor & Interface
| Maximum Power Consumption | 300 W total board power |
| Graphics Bus | PCIe 5.0 x16 |
| Form Factor | 4.4in (H) x 10.5in (L), dual slot, full height |
| Thermal Solution | Active |
| Power Connector | 1x PCIe CEM5 16-pin |
Specifications are taken from NVIDIA's published datasheet for this model. Fields NVIDIA does not publish for a given card are shown as such rather than estimated.
Compatibility & Software Support
| Graphics APIs | DirectX 12 (Shader Model 6.6), OpenGL 4.6, Vulkan 1.4 |
| Compute APIs | CUDA 12.8, OpenCL 3.0, DirectCompute |
| Operating Systems | Windows 10/11 64-bit; Linux — RHEL 8.10, SUSE Linux Enterprise Desktop 15.6, OpenSUSE 15.6, Fedora 41, Ubuntu 22.04; FreeBSD 14.2; Solaris 11 U4 (per NVIDIA RTX PRO Blackwell Quick Start Guide) |
| Drivers | NVIDIA RTX Enterprise Driver — Production Branch and New Feature Branch |
| Professional Applications | Certified across leading CAD, DCC, BIM, medical imaging and simulation ISV applications |
| AI Frameworks | CUDA, cuDNN, TensorRT, PyTorch, TensorFlow, NVIDIA NIM and NGC containers |
| Omniverse / Digital Twin | Supported via NVIDIA Omniverse and OpenUSD workflows |
| Chassis Check | 4.4in (H) x 10.5in (L), dual slot, full height — confirm slot clearance, power headroom and airflow before ordering |
SourceIT can verify fit against your specific workstation or server model before you order. For ISV certification status on a named application version, check the NVIDIA professional product literature library.
Package Contents & NVIDIA RTX PRO Blackwell Ordering Information
What is in the box: NVIDIA's RTX PRO Blackwell Quick Start Guide lists a Quick Start Card and an auxiliary power cable (2x PCIe 8-pin PSU to 1x PCIe Gen5 16-pin GPU adapter). NVIDIA does not publish a per-SKU box-contents list on the datasheet.
NVIDIA RTX PRO Blackwell range available from SourceIT:
| Part Number | Model & Memory |
| 900-5G144-2500-000 | RTX PRO 6000 Blackwell Workstation Edition — 96 GB GDDR7 with ECC |
| 900-5G153-2500-000 | RTX PRO 6000 Blackwell Max-Q Workstation Edition — 96 GB GDDR7 with ECC |
| 900-2G153-0000-000 | RTX PRO 6000 Blackwell Server Edition — 96 GB GDDR7 with ECC |
| 900-5G153-2570-000 | RTX PRO 5000 Blackwell 72 GB — 72 GB GDDR7 with ECC |
| 900-5G153-2550-000 | RTX PRO 5000 Blackwell 48 GB — 48 GB GDDR7 with ECC (this item) |
| 900-5G147-2550-000 | RTX PRO 4500 Blackwell — 32 GB GDDR7 with ECC |
| 900-5G147-2570-000 | RTX PRO 4000 Blackwell — 24 GB GDDR7 with ECC |
| 900-5G195-2501-000 | RTX PRO 4000 Blackwell SFF Edition — 24 GB GDDR7 with ECC |
| 900-5G195-2551-000 | RTX PRO 2000 Blackwell — 16 GB GDDR7 with ECC |
Choosing between them: memory capacity is usually the deciding factor for AI work and scene size, while form factor and power decide what will physically fit. The Max-Q variant exists to give full 96 GB capacity at 300 W for chassis that cannot take a 600 W card or that need more than one GPU. The Server Edition is passively cooled and belongs in a server, not a workstation. The SFF Edition trades bandwidth and PCIe lanes for a 70 W low-profile board.
Datasheet & Downloads
Frequently Asked Questions — RTX PRO 5000 Blackwell 48 GB
How much GPU memory does it have, and why does that matter?
48 GB GDDR7 with ECC. For AI work, VRAM capacity decides which models fit on the card at all — a model that does not fit runs an order of magnitude slower or not at all. For visualisation it decides scene and texture budget. ECC means single-bit memory errors are corrected rather than silently corrupting a long job.
How does it compare to the RTX PRO 4000 Blackwell?
The RTX PRO 4000 steps up to 24 GB of GDDR7, 8,960 CUDA cores, 672 GB/s and 1,178 AI TOPS at 145 W. The RTX PRO 2000 delivers 16 GB, 4,352 cores, 288 GB/s and 545 AI TOPS at 70 W. If your models and scenes fit in 16 GB, the 2000 is the efficient choice; if they do not, no amount of compute compensates.
What power supply and chassis clearance do I need?
This card is 300 w total board power in a 4.4in (H) x 10.5in (L), dual slot, full height form factor with active cooling, powered via 1x PCIe CEM5 16-pin. Check both physical clearance and PSU headroom against your workstation before ordering — send us your machine model and we will confirm fit.
Does it support Multi-Instance GPU?
Yes — up to 2x 24 GB or 1x 48 GB (maximum 2 instances). Each instance gets isolated memory and compute, so several users, containers or services can share one card with hard boundaries instead of contending for it.
Is this a professional card or a gaming card?
Professional. RTX PRO Blackwell cards carry ECC memory, NVIDIA RTX Enterprise drivers with ISV application certification, longer driver support branches and a professional warranty. Consumer GeForce cards have none of these, which is why they are not accepted in most enterprise and regulated deployments regardless of raw frame rate.
Does it come with local warranty in Singapore?
Yes. Every RTX PRO 5000 Blackwell 48 GB sold by SourceIT comes with 3 years local Singapore warranty and a GST invoice.
