Elastic Desktop Service (EDS) Enterprise Edition offers cloud-based graphics workstation instances powered by the NVIDIA RTX 5880 GPU, available in two configurations: Graphics Workstation Ultimate and Workstation Pro. Built on the NVIDIA Ada Lovelace architecture, the RTX 5880 combines third-generation RT Cores, fourth-generation Tensor Cores, next-generation CUDA Cores, and 48 GB of GDDR6 GPU memory to deliver high-throughput rendering, graphics, and compute performance — without the upfront cost of physical hardware.
Key advantages
| Advantage | Details |
|---|---|
| Flexible specifications | GPU memory from 2 GB to 192 GB; vCPU and system memory scale with workload — from lightweight content moderation to large-scale industrial simulation and AI model training |
| Powerful performance | RTX 5880 outperforms comparable GPU models by up to 10% in FP32 compute; broad professional software compatibility |
| Low-latency remote experience | 11 global regions; proprietary ASP protocol with adaptive encoding delivers distortion-resistant remote desktop access |
| Data security | All data stays in the cloud — nothing is written to local devices; anti-screen capture, screen recording audits, and domain and application whitelists protect against data leakage across terminal, transmission, and network layers |
| High availability | SLA of ≥99.975% service availability per cloud computer; 99.99% connection success rate; minute-level migration and recovery |
| Flexible billing | Four billing options: subscription, 360-hour plan, 250-hour plan, and pay-as-you-go — match payment to actual usage patterns |
GPU performance
The RTX 5880 leads the L20 and A10 across all key GPU specifications, making it suited for graphics rendering, industrial software, architectural visualization, and AI inference.
| GPU | Architecture | Primary use case | GPU memory | Memory bandwidth | CUDA cores | FP32 | TF32 | FP16 | FP8 |
|---|---|---|---|---|---|---|---|---|---|
| RTX 5880 | Ada Lovelace | Professional graphics, design, and AI | 48 GB GDDR6 | 960 GB/s | 14,080 | 69.3 | 138.5 | 277 | 554 |
| L20 | Ada Lovelace | AI compute and data center acceleration | 48 GB GDDR6 | 864 GB/s | 11,776 | 59.5 | 59.5 | 119 | 239 |
| A10 | NVIDIA Ampere | Interactive rendering and virtual workstations | 24 GB GDDR6 | 600 GB/s | 9,216 | 31.2 | 62.5 | 125 | — |
Compute values are in TFLOPS.
FP32 is the standard metric for graphics rendering quality, particularly for scenarios requiring high-fidelity visuals such as AAA games, film-grade CG, and deep learning.
The RTX Ada 5880 delivers performance comparable to the RTX Ada 6000. Its third-generation RT Cores reduce Omniverse real-time rendering latency by up to 50% compared to other GPUs in its class, making it suited for embodied intelligence workloads.
Choose a configuration
Both Graphics Workstation Ultimate and Workstation Pro use the AMD Genoa 9T24 CPU (up to 3.7 GHz) and the NVIDIA RTX 5880 GPU. Use the table below to find the configuration that matches your role and workload.
| Workload intensity | Typical roles | Typical tasks | Recommended configuration |
|---|---|---|---|
| Light | Content reviewers, lightweight 3D tasks | Content moderation, 4K display, GPU-required apps with low GPU memory | Graphics Workstation Ultimate – 4–8 vCPUs, 8–16 GB memory, 2 GB GPU memory |
| Moderate | Designers, annotation engineers, live streamers | Simple parts design, autonomous driving data annotation, digital human livestreaming | Graphics Workstation Ultimate – 8 vCPUs, 16 GB memory, 4 GB GPU memory |
| Heavy | CAD/CAE engineers, game developers, simulation engineers | Complex parts design, autonomous driving system development, light simulation, game development | Graphics Workstation Ultimate – 16–32 vCPUs, 32–64 GB memory, 8–16 GB GPU memory |
| Intensive | Industrial designers, robotics engineers, simulation teams | Large-scale industrial design, heavy simulation, robotics development | Graphics Workstation Ultimate – 32–64 vCPUs, 64–128 GB memory, 24 GB GPU memory |
| Professional / AI | AI researchers, embodied intelligence teams, scientific computing | Massive-scale rendering, AI inference, large model simulation | Workstation Pro – 32–96 vCPUs, 64–384 GB memory, 48–192 GB GPU memory |
Pricing
Hibernation is not currently supported. Cloud computers purchased under a monthly usage plan default to auto-shutdown on disconnect. Cancel or modify this policy in the console after purchase. For details, see Manage policies and Create a cloud computer.
Pricing covers compute resources only and excludes storage. The final price is shown on the purchase page.
Workstation Pro - Network-enhanced is in invite-only preview and supports UI customization for hardware terminals only. To request access or inquire about hardware terminal customization, submit a ticket.
| Specification | Monthly (USD/month) | 360-hour plan (USD/month) | 250-hour plan (USD/month) | Pay-as-you-go (USD/hour) |
|---|---|---|---|---|
| Graphics Cloud Computer – 8 vCPUs, 16 GB memory, 4 GB GPU memory | 156 | 105 | 86 | 0.390489 |
| Graphics Workstation Ultimate – 16 vCPUs, 32 GB memory, 8 GB GPU memory | 293 | 195 | 162 | 0.732594 |
| Graphics Workstation Ultimate – 16 vCPUs, 32 GB memory, 12 GB GPU memory | 371 | 375 | 250 | 0.92808 |
| Graphics Workstation Ultimate – 32 vCPUs, 64 GB memory, 12 GB GPU memory | 469 | 473 | 315 | 1.17244 |
| Graphics Workstation Ultimate – 32 vCPUs, 64 GB memory, 16 GB GPU memory | 547 | 352 | 291 | 1.367932 |
| Graphics Workstation Ultimate – 32 vCPUs, 64 GB memory, 24 GB GPU memory | 782 | 528 | 430 | 1.954398 |
| Graphics Workstation Ultimate – 64 vCPUs, 128 GB memory, 24 GB GPU memory | 977 | 645 | 537 | 2.443120 |
| Workstation Pro – 32 vCPUs, 64 GB memory, 48 GB GPU memory | 1,173 | 782 | 645 | 2.931842 |
| Workstation Pro – 64 vCPUs, 128 GB memory, 48 GB GPU memory | 1,564 | 1,055 | 860 | 3.909286 |
| Workstation Pro – 64 vCPUs, 256 GB memory, 48 GB GPU memory | 1,955 | 1,329 | 1,075 | 4.886729 |
| Workstation Pro – 64 vCPUs, 256 GB memory, 96 GB GPU memory | 2,932 | 1,955 | 1,613 | 7.330338 |
| Workstation Pro – 96 vCPUs, 384 GB memory, 192 GB GPU memory | 5,669 | 3,714 | 3,118 | 14.172444 |
| Workstation Pro - Network-enhanced – 32 vCPUs, 64 GB memory, 48 GB GPU memory | 1,290 | 860 | 709 | 3.225075 |
| Workstation Pro - Network-enhanced – 64 vCPUs, 256 GB memory, 96 GB GPU memory | 3,225 | 2,150 | 1,773 | 8.063421 |
| Workstation Pro - Network-enhanced – 96 vCPUs, 384 GB memory, 192 GB GPU memory | 6,236 | 4,086 | 3,429 | 15.589737 |
Use cases
EDS graphics workstations support professional GPU workloads across six domains.
Creative design and media
Supports Adobe Photoshop, Illustrator, and 3ds Max, covering the full workflow from conceptual design to final output. Suited for AAA game asset production, film-grade CG rendering, and high-fidelity visual creation.
CAD and engineering
Compatible with AutoCAD, SolidWorks, and CATIA for engineering modeling and complex drafting. Supports end-to-end collaboration from design through implementation.
Advanced manufacturing and automotive
Supports Siemens NX and PTC Creo for complex component design. GPU-accelerated parametric modeling and reverse engineering meet the precision requirements of mechanical design and virtual testing for autonomous driving systems.
Industrial simulation (CAE)
Compatible with Altair HyperWorks and ANSYS for multi-body dynamics analysis and topology optimization. Handles structural design of large equipment and heavy simulation validation under extreme conditions.
AI development and inference
Supports ONNX Runtime, PyTorch, and TensorFlow. Works with NVIDIA cuDNN, TensorRT, and cuBLAS operator libraries, and LLM inference frameworks including SGLang and vLLM. Low-latency compute enables real-time AI inference at the edge.
Embodied intelligence and digital twins
Supports NVIDIA Omniverse, Isaac Sim, and Isaac Lab for robotics development, digital twin simulation, and virtual testing for autonomous driving. The RTX 5880's third-generation RT Cores reduce Omniverse rendering latency by up to 50% compared to other GPUs in its class, supporting the full workflow from algorithm validation to deployment.
Supported regions
Graphics Workstation Ultimate and Workstation Pro are available in the following regions:
China (Beijing)
China (Shanghai)
China (Hangzhou)
China (Shenzhen)
China (Chengdu)
China (Guangzhou)
China (Qingdao)
China (Ulanqab)
China (Hong Kong)
Singapore
Japan (Tokyo)
FAQ
Which GPU is suitable for embodied intelligence projects?
The RTX Ada 5880 is built for this workload. Its third-generation RT Cores cut real-time rendering latency in Omniverse by up to 50% compared to other GPUs in its class. The graphical Ubuntu OS interface fits standard robotics development environments. Choose Workstation Pro configurations with 48–192 GB GPU memory to handle the complex algorithms and real-time data processing that embodied intelligence requires.
Can EDS run simulation software?
Yes. EDS supports Altair HyperWorks and ANSYS for fluid dynamics, structural analysis, and electromagnetic simulation. Workstation Pro configurations provide the CPU and GPU memory headroom for heavy simulation workloads.
I use AutoCAD and SolidWorks. Which configuration should I choose?
For 2D drafting and light 3D work in AutoCAD, Graphics Workstation Ultimate with 4–8 GB GPU memory is sufficient. For complex assemblies in SolidWorks, use a 16 or 24 GB GPU memory configuration to handle large model sets without performance degradation.
Our team develops autonomous driving technology. How much GPU memory do we need?
For data annotation, 4 GB GPU memory is typically enough. For system development — sensor fusion, path planning, and model training — choose 16–24 GB GPU memory configurations. Large-scale simulation and multi-model inference benefit from Workstation Pro with 48 GB or more.
We run digital human livestreams. What configuration should we use?
Graphics Workstation Ultimate with 4–8 GB GPU memory handles real-time avatar rendering and video encoding for most livestreaming setups. If you run multiple concurrent accounts or high-definition avatars simultaneously, move to 8–12 GB GPU memory configurations.
RTX 4090 vs. EDS RTX Ada 5880 — what is the difference?
The RTX 4090 is optimized for gaming and light professional graphics. The RTX Ada 5880 is designed for professional CAD, scientific simulation, and AI inference, with 48 GB of VRAM or more to handle large datasets and complex models that the 4090 cannot fit in memory.
What are the cost advantages over physical GPU servers?
Pay-as-you-go billing eliminates upfront hardware investment. You avoid hardware maintenance, firmware updates, and physical capacity planning. Monthly usage plans (250-hour and 360-hour) reduce the per-hour cost for predictable workloads, while pay-as-you-go works best for burst or experimental usage.
Can a single account run multiple instances at the same time?
Yes. A single user can run multiple cloud computer instances simultaneously, which is useful for automated livestreaming studios, parallel annotation pipelines, and multi-environment development setups.