Unboxing the Performance of Ultra-High-End NVIDIA RTX 6000 Ada Professional Graphics Card

Advantech offers a diverse range of Industrial GPU Solutions, especially featuring NVIDIA's complete series of enterprise-grade professional graphics cards. As an elite partner of NVIDIA, Advantech has priority access to the latest generation of GPU cards and exclusive support from the NVIDIA technical team, providing professional consulting services to Advantech customers.

Advantech's Edge AI solutions are widely applied in various industries, accumulating rich experience in industrial and enterprise-level user collaborations. This article is a comparative evaluation of the popular professional-grade NVIDIA RTX GPU card series, helping customers assess suitable deployments for industrial IoT and various vertical applications such as manufacturing, transportation, medical, and gaming industries.

Highlights of this article:

• Comparison of specifications between NVIDIA RTX 6000 Ada and RTX A6000 
• Feature highlights of the ultra-high-end professional graphics card NVIDIA RTX 6000 Ada
• Four professional performance evaluations:

1. Professional modeling and drawing software SPECviewperf 2020 performance test.
2. AI image generation model and Stable Diffusion computational speed test.
3. Deep learning, AI model training Lambda Labs PyTorch GPU Training benchmark.
4. 3D performance 3DMark Time Spy test.

---

Comparison of NVIDIA RTX 6000 Ada and RTX A6000 specifications:

The latest NVIDIA® RTX™ 6000 GPU card uses the new Ada Lovelace architecture, providing much higher CUDA cores, third-generation RT cores, and fourth-generation Tensor cores. Compared to the previous generation RTX A6000, it has an approximately 70% increase in core numbers and performance, offering impressive FP32 single-precision floating-point computing power.   

The NVIDIA RTX 6000 Ada is equipped with 48GB of ECC video memory but has a maximum power consumption of only 300 watts, about two-thirds of the flagship gaming player GPU card. This is crucial for deployment in edge AI applications and high-end image processing, helping customers build more efficient and stable AI acceleration and graphic workstations.

Specification Comparison:

In terms of power connectors, NVIDIA RTX 6000 Ada introduces a new 12VHPWR 16-pin interface.

During the transition period, NVIDIA thoughtfully provides customers with a PCIe 8-pin x 2 to 16-pin adapter cable, allowing continued use of existing power supply.

Aesthetically, the RTX 6000 Ada maintains the same piano paint mirror design as the RTX A6000 and the "reflective" RTX GPU card is a symbol of its top-tier status.

It’s important to note that the upper side connection ports of NVIDIA RTX 6000 Ada are mainly for use with Quadro Sync multi-screen synchronous display function, no longer offering NVLink connectivity. Customers requiring NVLink functionality are advised to use previous generation NVIDIA RTX A4500 / A5000 / A5500 / A6000 GPU cards.

The latest flagship NVIDIA RTX 6000 Ada has a significant upgrade in specifications compared to the previous generation RTX A6000, with substantial increases in all aspects of theoretical peak performance. We will detail these characteristics for enterprise users.

---

Feature highlights of the ultra-high-end professional graphics card NVIDIA RTX 6000 Ada:

• Large Capacity, High-Speed 48GB GDDR6 Memory

Utilizing 48GB GDDR6 memory, RTX 6000 achieves a 25% increase in throughput for ray tracing, rendering, and AI workloads compared to the previous generation. RTX 6000 Ada offers the largest graphic memory in the industry, capable of handling the largest datasets and models for latency-sensitive professional applications. 

• Memory with Error Correcting Code (ECC)

RTX 6000 automatically detects and corrects errors, enhancing data integrity and maintaining the accuracy and stability of model computations, providing unparalleled reliability for critical tasks. This prevents data access or transmission errors from abruptly interrupting long computational processes in workstations, offering unmatched accuracy and reliability.

• NVIDIA Virtual GPU Software

NVIDIA Virtual GPU (vGPU) software allows personal workstations to be reconfigured into multiple high-performance virtual workstations, enabling remote users to share resources for driving high-end design, AI, and computational workloads.

NVIDIA® RTX™ Experience

NVIDIA RTX Experience provides a suite of productivity tools for your desktop workstation, including up to 8K desktop recording, active notifications for the latest NVIDIA RTX Enterprise driver updates, and access to gaming features.

---

Four professional performance evaluations:

Following the unboxing, we now arrive at the highlight of this article – performance evaluation. In this test, we include four NVIDIA RTX 6000 Ada, RTX A6000, RTX A4500, and RTX A4000 cards, representing professional and enterprise-level GPU products.      

The testing tools include SPECviewperf 2020 software for professional modeling and drawing, the popular AI image generation model Stable Diffusion computational speed test, Lambda Labs PyTorch GPU Training Benchmark for deep learning and AI model training, and the 3DMark Time Spy for 3D performance testing.   

Overall, NVIDIA RTX 6000 Ada shows significant performance growth in all tests compared to the previous generation RTX A6000 GPUs. This is primarily due to the new ADA Lovelace architecture, upgraded CUDA cores, third-generation RT Core, and fourth-generation Tensor Core. The core number has increased by 69% compared to the previous generation, with rendering performance tripled and AI inference performance doubled.   

This article will focus on reviewing NVIDIA RTX 6000 Ada, looking at its advantages in performance, power consumption, and system integration.

1. Professional Modeling and Drawing Software SPECviewperf 2020 Performance Test

SPECviewperf 2020 tests the drawing performance of various industry professional software, including 3ds Max, Catia, Creo, Energy, Maya, Medical, SNX, SolidWorks, and other drawing and engineering simulation software.

The test results show that RTX 6000 Ada dominates in performance across all software compared to other RTX GPU cards. Taking the commonly used modeling software in the manufacturing industry, SNX, as an example, at 1080p resolution, the performance of RTX 6000 Ada is almost double that of RTX A6000 (1080p). For medical and energy domain applications, RTX 6000 Ada provides double or more performance than the other GPU cards tested. This shows that NVIDIA RTX 6000 Ada is highly suitable for professional drawing and modeling software.

---

2. AI Image Generation Model Stable Diffusion Computational Speed Test

Generative AI is currently one of the hottest AI applications. Stable Diffusion is a deep-learning model capable of generating detailed images from text descriptions. More and more enterprise-level customers need to generate AI images in a local environment to avoid concerns about data leakage. This requires Advantech's GPU Edge Computer paired with NVIDIA GPU cards to build a private graphics processing workstation to run Stable Diffusion. 

The test results reveal that RTX 6000 Ada renders images the fastest among all tested cards. Generating 8 images at once (batch size=8), it takes less than a second to generate one image! It also maintains the same power consumption, generating more images per second compared to RTX A6000 (0.66 images→1.06 images), shortening computation time and optimizing image generation efficiency.

---

3. Deep Learning, AI Model Training Lambda Labs PyTorch GPU Training Benchmark 

PyTorch is a framework for building deep learning models, commonly used in computer vision, natural language processing, and even Tesla's AI Model Training for autonomous driving. In this test, RTX 6000 Ada still dominates all RTX GPU cards in processing various AI model training data. However, it's worth noting that highly divisible AI model training allows more GPU cards to be inserted into a system, proportionally increasing training speed, making GPU card and system compatibility very important.

The advantage of RTX 6000 Ada lies in its dual slots form factor and low power design, allowing customers to integrate more GPU cards into a server, thereby enhancing AI training performance. For example, in a 2U server SKY-6200, up to four RTX 6000 Ada cards can be inserted. However, if using GPU cards that require a 3.5 slots form factor, their different design often poses a risk of not being integrated into a 2U server.    For instance, the commonly used 4U server SKY-640 V3, if equipped with 3.5 slots GPU cards, can only integrate two GPU cards. But if equipped with RTX 6000 Ada, it can integrate four GPU cards, providing double the model training performance compared to a two-card system. Moreover, the required TGP (graphics card maximum power supply) is only 1200W, completely eliminating concerns of insufficient power supply in common 2000W servers.

Especially in terms of cooling design, RTX 6000 Ada is paired with a dual intake blower design.

Hence, by integrating multiple RTX 6000 Ada GPU cards into a system in a parallel and closely fitting configuration, an efficient intake volume can be attained, ensuring operational stability. This configuration holds the key to achieving scalability and ensuring long-term stable operation.

---

4. 3D Performance 3DMark Time Spy Test

Finally, in the 3DMark Time Spy test, the performance of RTX 6000 Ada increased by 76% compared to the previous generation A6000 (18,514 → 32,659), mainly benefiting from the upgraded core numbers.

The following figure shows that during the computation process, the TGP of RTX 6000 Ada is maintained at nearly 300W, while the GPU Clock fluctuates between 1,900MHz and 2,500MHz.

RTX 6000 Ada uses more CUDA Cores, Tensor Cores, and RT Cores. Therefore, it can maintain high efficiency with a lower and fluctuating GPU Clock, bringing the advantage of low power consumption. This is the secret to long-term stable operation of workstations.

---

Conclusion: Top-Tier GPU Paired with the World's Leading Industrial Computer, Deploying the Most Efficient and Reliable AI Workstations

The top-tier NVIDIA RTX 6000 Ada, among the four GPU cards tested, boasts the highest performance, excellent bi-directional cooling design, and low power consumption advantages, bringing optimal system integration, expandability, and software compatibility. It supports various applications to meet professional needs, providing enterprise customers with more than double the intensive computing performance while ensuring 7 day/24 hour stable operation and efficiency of workstations. 

 With the widespread application of edge computing, AI acceleration, GPU computing, and 5G, more and more AI applications need to be deployed at the edge to provide real-time AI inference computation and reduce latency. Advantech, as an NVIDIA Elite Partner, supplies the full range of NVIDIA Enterprise PCIe GPU cards, including professional RTX GPU cards, as well as GPU cards used in data centers and AI Training. It has priority access to the latest product usage rights and exclusive support from NVIDIA’s technical team in hardware, software, and driver programs.    

 More importantly, Advantech, as the world's leading industrial computer brand, has accumulated many years of experience in edge computing and AI solutions. It offers a variety of systems for customers to choose from, as well as integration services. Passing NVIDIA's strict system certification (NVIDIA-Certified and Qualified), it ensures stable system operation and software compatibility, allowing customers' various application projects to be deployed and launched in the shortest time, creating optimal value in the era of AI and accelerated computing.