What are the full specifications of the RTX PRO 5000 Blackwell 72GB?

NVIDIA Blackwell architecture, 14,080 CUDA cores, fifth-gen Tensor Cores, fourth-gen RT Cores — up to 2,064 AI TOPS, 65 TFLOPS single-precision, 196 TFLOPS RT Core. 72GB GDDR7 with ECC, 384-bit interface, 1.3 TB/s bandwidth. PCIe 5.0 ×16; 4× DisplayPort 2.1b; 3× NVENC (9th gen) + 3× NVDEC (6th gen) with 4:2:2. 300W total board power, active dual-slot 4.4″ × 10.5″ full height. Multi-Instance GPU up to 2× 36GB or 1× 72GB; RTX PRO Sync compatible.

Why 72GB — what can it run that smaller professional cards can't?

The 72GB is the reason to choose this card — 50% more than the 48GB tier. NVIDIA positions it for desktop agentic AI: training, fine-tuning and prototyping larger models locally, and serving them from a workstation instead of the data centre for data privacy and low latency. In practice, 72GB holds models in the 30–70B class with headroom for context and concurrency that 24–48GB cards run out of, and it can split into two isolated 36GB instances for multi-user or multi-workload teams. Real capacity depends on model and quantization.

RTX PRO 5000 vs RTX PRO 6000 — which should I choose?

Both are Blackwell. The RTX PRO 6000 is the flagship: 24,064 CUDA cores, 96GB, up to 4,000 AI TOPS, four-way MIG and 600W. The 5000 gives you 72GB and 14,080 cores at half the power (300W) and a lower price, in a dual-slot card. Choose the 6000 when you need maximum single-card throughput, the largest models or four-way partitioning; choose the 5000 when 72GB covers your models and power, space or budget matter.

How does it compare to the previous-generation RTX 5000 Ada?

A substantial generational step: 72GB GDDR7 with ECC versus the RTX 5000 Ada's 32GB GDDR6 — more than double the memory — on the newer Blackwell architecture with fifth-gen Tensor and fourth-gen RT cores and PCIe 5.0. NVIDIA states up to 2.1× LLM inference, 2× text generation and 3.5× image generation over the previous generation. If your work is memory-bound, the doubled-plus VRAM alone is the headline.

What warranty and support come with this card from EMARQUE?

The card carries a 3-year manufacturer warranty. EMARQUE handles local RMA routing in Malaysia — if a hardware fault appears, the warranty claim is routed on your behalf so you deal with one local team rather than the manufacturer's process. Pre-sale and post-sale questions go through the same team: WhatsApp 012 627 2280 or business@emarque.co.

Can EMARQUE build a workstation or server around this GPU?

Yes. EMARQUE supplies NVIDIA RTX PRO Blackwell cards standalone, or integrated and validated inside an EMARQUE workstation or AI PRO multi-GPU server — with thermals, power and BIOS configured for the card count, built in Malaysia. For a scoped configuration: WhatsApp 012 627 2280 or business@emarque.co.

1 2

NVIDIA RTX PRO 5000 Blackwell 72GB GDDR7 ECC Professional Graphics Card

Name: NVIDIA RTX PRO 5000 Blackwell 72GB GDDR7 ECC Professional Graphics Card
Brand: NVIDIA
SKU: PRO5000-72GB
Price: 47469.00 MYR
Availability: InStock

RM47,469.00

Sale Sold out

Shipping calculated at checkout.

NVIDIA RTX PRO 5000 Blackwell — 72GB GDDR7, up to 2,064 AI TOPS

The NVIDIA RTX PRO 5000 Blackwell is a professional GPU built for desktop agentic AI and demanding creative workflows — 72GB of GDDR7 memory with ECC and 14,080 CUDA cores on the Blackwell architecture, engineered to keep larger models resident locally for training, fine-tuning and inference, alongside high-fidelity rendering and simulation. It is NVIDIA's newest memory tier in the line: 50% more VRAM than the 48GB card, in a 300W dual-slot design.

NVIDIA Blackwell architecture — 14,080 CUDA cores, fifth-gen Tensor Cores, fourth-gen RT Cores
72GB GDDR7 with ECC — 384-bit interface, 1.3 TB/s bandwidth — keep large models and datasets resident on-card
Up to 2,064 AI TOPS — 65 TFLOPS single-precision, 196 TFLOPS RT Core performance
Multi-Instance GPU — split into up to 2× 36GB isolated instances for multi-user or multi-workload teams
PCIe 5.0 ×16 · 4× DisplayPort 2.1b · RTX PRO Sync compatible
300W dual-slot — NVIDIA-stated up to 2.1× LLM inference and 3.5× image generation versus the previous generation

Specifications

GPU architecture: NVIDIA Blackwell · CUDA cores: 14,080 · Tensor Cores: fifth generation · RT Cores: fourth generation
AI performance: up to 2,064 AI TOPS · Single-precision: 65 TFLOPS · RT Core performance: 196 TFLOPS
GPU memory: 72GB GDDR7 with ECC · Memory interface: 384-bit · Memory bandwidth: 1.3 TB/s (1,344 GB/s)
System interface: PCIe 5.0 ×16 · Display connectors: 4× DisplayPort 2.1b
Video engines: 3× NVENC (ninth gen), 3× NVDEC (sixth gen), 4:2:2 support
Total board power: 300W · Thermal: active · Form factor: 4.4″ × 10.5″, dual slot, full height
Multi-Instance GPU: up to 2× 36GB / 1× 72GB · RTX PRO Sync compatible

Part of the NVIDIA RTX PRO Blackwell family: the RTX PRO 4500 (32GB) and the flagship RTX PRO 6000 (96GB).

Manufacturer warranty. Supplied in Malaysia by EMARQUE.

WhatsApp Us: 012 627 2280

Built by Enthusiasts

Over 10 Years of Experience

Verified Components

57-Point QA/QC + Stress Test

Instalments Available

Credit Card / BNPL

Secured Shipping

Throughout Malaysia

Manufacturer Warranty

Per Product Specifications

7 Days 1-1 Exchange

On Product Defect

View full details

FAQ

Frequently asked questions

Everything you need to know about this NVIDIA RTX PRO Blackwell GPU — specifications, support, and Malaysian fulfilment.

What are the full specifications of the RTX PRO 5000 Blackwell 72GB?

NVIDIA Blackwell architecture, 14,080 CUDA cores, fifth-gen Tensor Cores, fourth-gen RT Cores — up to 2,064 AI TOPS, 65 TFLOPS single-precision, 196 TFLOPS RT Core. 72GB GDDR7 with ECC, 384-bit interface, 1.3 TB/s bandwidth. PCIe 5.0 ×16; 4× DisplayPort 2.1b; 3× NVENC (9th gen) + 3× NVDEC (6th gen) with 4:2:2. 300W total board power, active dual-slot 4.4″ × 10.5″ full height. Multi-Instance GPU up to 2× 36GB or 1× 72GB; RTX PRO Sync compatible.
Why 72GB — what can it run that smaller professional cards can't?

The 72GB is the reason to choose this card — 50% more than the 48GB tier. NVIDIA positions it for desktop agentic AI: training, fine-tuning and prototyping larger models locally, and serving them from a workstation instead of the data centre for data privacy and low latency. In practice, 72GB holds models in the 30–70B class with headroom for context and concurrency that 24–48GB cards run out of, and it can split into two isolated 36GB instances for multi-user or multi-workload teams. Real capacity depends on model and quantization.
RTX PRO 5000 vs RTX PRO 6000 — which should I choose?

Both are Blackwell. The RTX PRO 6000 is the flagship: 24,064 CUDA cores, 96GB, up to 4,000 AI TOPS, four-way MIG and 600W. The 5000 gives you 72GB and 14,080 cores at half the power (300W) and a lower price, in a dual-slot card. Choose the 6000 when you need maximum single-card throughput, the largest models or four-way partitioning; choose the 5000 when 72GB covers your models and power, space or budget matter.
How does it compare to the previous-generation RTX 5000 Ada?

A substantial generational step: 72GB GDDR7 with ECC versus the RTX 5000 Ada's 32GB GDDR6 — more than double the memory — on the newer Blackwell architecture with fifth-gen Tensor and fourth-gen RT cores and PCIe 5.0. NVIDIA states up to 2.1× LLM inference, 2× text generation and 3.5× image generation over the previous generation. If your work is memory-bound, the doubled-plus VRAM alone is the headline.
What warranty and support come with this card from EMARQUE?

The card carries a 3-year manufacturer warranty. EMARQUE handles local RMA routing in Malaysia — if a hardware fault appears, the warranty claim is routed on your behalf so you deal with one local team rather than the manufacturer's process. Pre-sale and post-sale questions go through the same team: WhatsApp 012 627 2280 or business@emarque.co.
Can EMARQUE build a workstation or server around this GPU?

Yes. EMARQUE supplies NVIDIA RTX PRO Blackwell cards standalone, or integrated and validated inside an EMARQUE workstation or AI PRO multi-GPU server — with thermals, power and BIOS configured for the card count, built in Malaysia. For a scoped configuration: WhatsApp 012 627 2280 or business@emarque.co.