Spec Card SPEC CARD
NVIDIA Quadro RTX 8000
48GB former flagship rendering card (full-clip Turing TU102, first-generation ray tracing GPU): one of the most widely circulated 48GB-class cards on the used rendering farm / virtual workstation (vGPU/vWS) market; large volumes hit the secondhand market after the Ampere RTX A6000 launched in late 2020, making it the direct predecessor of the A6000
Overview
The NVIDIA Quadro RTX 8000 is a professional workstation graphics card built on the Turing-architecture TU102 core in its full-clip design, with 4,608 CUDA cores, 576 second-generation Tensor Cores and 72 first-generation RT Cores — among the first GPUs in the industry to offer hardware ray tracing acceleration. It carries 48GB of ECC GDDR6 memory on a 384-bit bus, delivers 16.3 TFLOPS of single-precision compute and 130.5 TFLOPS of Tensor performance. Two cards can be linked via NVLink to form a 96GB pooled memory pool. Its large memory capacity and professional driver positioning make it a common choice for rendering farms and virtual workstations (vGPU/vWS); after the RTX A6000 launched, large volumes moved to the secondhand market, making it one of the most widely circulated 48GB-class cards.
Core Specs
| Architecture | Turing (TU102, TSMC 12nm FFN, 18.6B transistors) |
| Memory | 48GB GDDR6 with ECC |
| Memory Bandwidth | 672GB/s (GDDR6 14Gbps) |
| Memory Bus Width | 384bit |
| Compute | FP32 16.3 TFLOPS (Boost) / Tensor 130.5 TFLOPS / FP16 32.6 TFLOPS / INT8 206.1 TOPS / RTX-OPS 84T / FP64 = 1/32 of FP32 ≈ 509 GFLOPS |
| CUDA cores | 4608 (72 SM) / Tensor Cores 576 (Turing Gen 2) / RT Cores 72 (Gen 1) |
| Power | Total board power 295W / graphics power 260W |
| Interconnect | NVLink 100GB/s bidirectional (TU102 dual x8 links, pair two cards for 96GB, NVLink bridge sold separately) / PCIe 3.0 x16 |
| Launch Year | 2018 (announced 2018-08-13 at SIGGRAPH Vancouver [media often log 08-14], Q4 2018 launch, expected market price $10,000) |
| Form Factor | Dual-slot full-height, 4.4"H x 10.5"L, active cooling, 1x 8-pin external power connector |
| Display output | 4x DisplayPort 1.4 + VirtualLink (USB Type-C) |
| Encode/decode | 1 Encode / 1 Decode |
| VR | Yes (VR Ready, first VirtualLink port in the industry) |
| API | DirectX 12.0 (FL 12_1) / OpenGL 4.6 / Vulkan 1.1 / CUDA / OpenCL / DirectCompute |
Specs verified against official documentation (
official benchmarks , updated 2026-10-10) · actual price subject to the seller's quote
Use Cases
- Offline rendering and render farm nodes: 48GB ECC memory suits large scenes, textures and geometry data, while ECC reduces the risk of data errors during long batch renders
- Virtual workstation (vGPU / vWS) deployments: as a large-memory professional card, it can allocate GPU resources to multiple remote users in virtualized environments for remote design and visualization workstations
- Content creation and visual effects (VFX/DCC): supports workflows in modeling, texturing and lighting previews that rely on memory capacity and ray tracing acceleration
- Deep learning inference and small-scale training: Tensor Cores provide INT8/FP16 acceleration, and 48GB of memory accommodates fairly large models; dual-card NVLink can further expand the memory pool
Same-Family Models
| Model | Memory | Power |
|---|
| NVIDIA T4 Compare | 16GB GDDR6 with ECC (enabled by default) | 70W (total board power, default) |
| NVIDIA A100 Compare | Dual versions: 40GB HBM2 / 80GB HBM2e (80GB added at SC20 on 2020-11-16), with ECC | SXM4 400W (same for 40GB/80GB); PCIe version 250W for 40GB / 300W for 80GB |
| NVIDIA H20 Compare | 96GB HBM3 (consistent across Flopper, VipraTech, and Tencent News sources) | 400W TDP (three sources: Flopper, VipraTech spec sheet, and Tencent News' "400W thermal design power"; VipraTech's body text also cites 350W, a conflicting figure that has been discarded) |
| NVIDIA GB300 Compare | 288GB HBM3e (per GPU package; 576GB for the superchip; HGX board-level 2.1TB / rack-level 20TB coexist) | TGP up to 1400W |
| NVIDIA A30 Compare | 24 GB HBM2 | 165 W |
| NVIDIA GeForce RTX 4090 Compare | 24GB GDDR6X | 450W TGP (315W average gaming, 850W recommended system power) |
Video Overview VIDEO
Procurement & Inspection Notes
- Rack space and power: this is a dual-slot full-height card measuring roughly 4.4 inches tall and 10.5 inches long — verify chassis/server shroud and slot spacing before purchase; it uses a single 8-pin power connector, so confirm the PSU provides one, or use the bundled 8-pin-to-dual-6-pin adapter if necessary
- Interconnect and cooling: if planning an NVLink dual-card setup (100 GB/s bidirectional, 96GB combined memory), note the NVLink bridge is sold separately and the application itself must support cross-card memory access; the dual-slot active cooling design demands good chassis airflow — evaluate inlet/outlet temperatures when densely deploying render nodes
- Used-unit inspection checklist: verify it is a genuine Quadro RTX 8000 professional card, not a re-purposed card with the same core; check ESD/fan condition and memory ECC error logs; confirm ports are 4x DP 1.4 plus VirtualLink (USB Type-C — check compatibility notes); the interface is PCIe 3.0 x16, usable directly on older platforms and backward compatible on newer ones
FAQ
Is a used Quadro RTX 8000 still worth buying? How does it compare with the RTX A6000?
The RTX 8000 is the direct predecessor of the RTX A6000, with the same 48GB memory positioning. The A6000 is on a newer architecture with more CUDA cores and improved memory specs, but costs more. The RTX 8000 circulates in volume on the used market and still offers 48GB per card, so it remains practical for rendering and virtual workstation workloads where memory capacity is the main requirement. If your workflow depends on newer-architecture features or higher compute, go for a later generation.
What are the limitations of pairing two Quadro RTX 8000 cards with NVLink for 96GB of memory?
NVIDIA officially supports connecting two RTX 8000 cards via NVLink, aggregating 100 GB/s bidirectional bandwidth and pooling both cards' memory into 96GB for applications — provided the application itself supports cross-GPU pooled memory access. Also note the NVLink bridge is a separately sold accessory you need to source at purchase, and the physical spacing between the two cards must match the chosen bridge.
Can the Quadro RTX 8000 fit in a regular gaming motherboard and case?
It uses a standard PCIe 3.0 x16 interface, compatible with mainstream platforms, but it is a dual-slot full-height professional card about 10.5 inches long — check case length clearance and whether it blocks adjacent slots before installing. Power needs only a single 8-pin connector; if the PSU lacks one, use the bundled 8-pin-to-dual-6-pin adapter. Its active cooling has certain airflow requirements, so watch heat exhaust in smaller cases.