Publish Date : 2019/04/22

Quadro RTX 5000 Hands-on Review


The application of GPU in the field of traditional graphic visualization has been continuously expanding. Whether it is 3D design, high-resolution multi-screen splicing display, or special effects rendering, more and more people are turning to more stable professional graphics cards. As a cutting-edge industry, deep learning and big data also use a large number of professional GPUs to speed up the development of training. NVIDIA's latest Turing architecture products have been available since last year. So what are the changes to the new Quadro RTX 5000 professional graphics card, and what have been improved? Let's find out in the unboxing and review.

The box

RTX 5000 box appearance

  • The style of the box has been changed to a green-white-gray color scheme; the gray grid with a gradient color on the front is very modern and stereoscopic. The front and top of the package are marked with the model of the graphics card.
  • The RTX word has a metallic reflective effect that highlights one of the most important features of this generation of graphics cards – Real-Time Ray Tracing.
  • The box is heavy, and the material of the package is very sturdy.

Unboxing

Packaging and accessories

  • The inner packaging is also different from before. The outer casing and the shock-proof sponge are integrated, and the graphics card is placed in it, giving the impression that it is firm and solid.
  • There is also a transparent anti-static bag outside the graphics card. There is a user manual, a support manual, an 8pin to double 6pin power cable, a DP to DVI adapter and a DP to HDMI adapter in the accessory box. This is a blessing for many users with DVI monitors or HDMI monitors; there is no need to purchase separate adapters.

Card front

  • The appearance of the graphics card is mainly in green, silver and black, which is consistent with the packaging style. The card is also heavy, giving people a very good feel. The model of the card can be seen on the front and back of the card.
  • The Quadro RTX 5000 is still a dual slot graphics card with a turbo fan. The card is powered by 8pin + 6pin, which is sufficient for the RTX 5000 with a maximum power consumption of 265W.

Connectors at the top of the card

  • There are 3 connectors at the top of the card: NVLink, SYNC, and Stereo.
  • NVLink replaces SLI and can be used with Quadro NVLink 2-Slot or 3-Slot. NVLink not only has the function of multi-card synchronous output, but also enables memory pooling when supported by the application, so that the memory on two cards are combined into one larger memory.
  • The SYNC connector is used with the Quadro SYNC II card to achieve multi-card display signal synchronization output. For users of large screen splicing, they can continue using their previous setup.
  • The Stereo connector should be used with a 3D stereo bracket to output 3D stereo signals.
  • The bus interface is PCIE 3.0 and is backward compatible with PCIE 2.0. Most of the motherboards on the market today use this interface.

Display connectors

  • There is a new change in the display connectors. The 4DP+ DVI configuration has been changed to the 4DP + VirtualLink configuration.
  • The DP connector supports the DP 1.4 protocol with a maximum resolution of 7680 x 4320. The VirtualLink connector is the standard interface used by next-generation VR devices. A single connector can perform multiple functions including power supply, display transmission, and signal control.

Graphics card PCB

Performance Test

Component Model
Motherboard Gigabyte Z390 AORUS Master
CPU I9 9900K
DISK NVME SSD 512GB
MEM 64GB DDR4
Graphic RTX 5000
Power ATX 1000W
System Windows 10 64 1809/Ubuntu 16.4
Driver version 419.71

Test software

SPECviewPerf 13
Superposition Benchmark
Vray Benchmark
CUDA-Z 0.10.251
3Dmark Port Royal
OctaneBench 2019 Preview
NVIDIA Tensorflow example

Specifications

  RTX 5000 P5000
CUDA Cores 3072 2560
Tensor Core 384 N/A
RT Core 48 N/A
GPU Memory 16 GB GDDR6 16 GB GDDR5X
Graphics Bus PCI Express 3.0 x 16 PCI Express 3.0 x 16
Bridge Mode NVLink SLI
Display connectors DP1.4(4), VirtualLink (1) DP1.4(4) DVI-D (1)
VRReady YES YES
SYNC Quadro SYNC II Quadro SYNC II
Power consumption Total board power:265W Total board power:180W

a. SPECviewperf 13

SPECviewperf 13 is a benchmark software that is widely used to measure graphics performance based on professional applications. The program tests the performance of professional graphics software based on OpenGL and DirectX, and SPECviewperf 13 brings 9 new professional graphics test scenes.

The SPECviewperf 13 test is closer to real-world applications. Some of the test scenarios even contain more than 60 million fixed-point data, which fully reflects the professional graphics performance of the card. The default configuration is used in the test.

From the test results, the RTX 5000 consistently outperformed the Quadro P5000 in all benchmarks, and the performance of the snx even increased by more than 40%. It seems that the Turing architecture adds more than just the efficiency of ray tracing and deep learning. There is also tremendous increase in the performance of professional applications.

b. SuperPosition Benchmark

This program is more like a complex game environment, where the graphics card's DX and OpenGL rendering performance and stability with different lighting effects are put to the test.

In terms of DirectX performance, RTX 5000 is nearly 45% faster than P5000. In terms of OpenGL, RTX 5000 has improved by about 50% over P5000. The performance of the two major graphics APIs has increased by a wide margin, and it is believed that RTX 5000 has higher usability in the professional 3D visualization field.

c. V-Ray Benchmark

Chaos Group's V-Ray has long been recognized by the majority of users in the rendering field. Due to the improved GPU rendering performance, Chaos Group introduced the V-Ray GPU NEXT version on V-Ray Next, which supports calling NVIDIA CUDA core for rendering. With the constantly evolving technology, the quality of GPU rendering is almost the same as that of CPU. GPUs are powerful, rendering time and cost is lower, and multi-card rendering is supported, so many renderers are adding their own GPU rendering capabilities. This benchmark only tests the rendering performance of a single card. The shorter the rendering time means the better performance.

Test screenshot

Test result

From the test results, using the latest RTX 5000 saves 35% of time compared with the P5000. The RTX 5000 is more efficient in V-Ray rendering.

d. CUDA-Z

Just like the CPU-Z and GPU-Z we are familiar with, CUDA-Z is a collection of basic information about NVIDIA GPU. It can be used on GeForce, Quadro and Tesla cards.

Test screenshot

CUDA performance

In the CUDA-Z test, the capability that is used the most is single-precision floating-point operations. If you are using double-precision scientific calculations, it is recommended to use GV 100 or GP 100 GPUs that have high double-precision performance. The RTX 5000's single-precision performance reaches 11.7T, a 36% increase over the previous P5000. Such a powerful single-precision floating-point performance is unmatched by the CPU, which is why more and more applications have shifted computing from the CPU to the GPU.

e. 3DMark Port Royal

Test scene

Test result

Currently NVIDIA's RTX rendering feature can be benchmarked in 3DMark's Port Royal. The Quadro P5000 cannot run this benchmark because it does not have an RT Core.

The test scene has a large amount of metal material, and the reflection effect is amazing. The RTX 5000 renders at around 28 FPS, which is fairly smooth. At present, the game industry already has BF5 using this technology, and it is believed that it will be widely used in the industrial manufacturing field and the later stage of film and television production in the future.

f. OctaneBench 2019 Preview

The OC renderer is a GPU-accelerated renderer for 3D design and animation. It can be used with 3D modeling and special effects software such as 3ds Max, CINEMA 4D, NUKE, and MODO. The OC renderer is a rendering software that supports Out of Core. The latest OctaneBench 2019 Preview is a software that supports RT Core to accelerate ray-traced rendering. We can look at the difference in rendering speed between RTX on and off.

Test screenshot

The software renders the same scene with RTX on and off respectively. It can be seen in the test results that the rendering speed with RTX on is nearly three times faster than that with RTX off. It is obvious that the rendering speed has been significantly improved with the help of RT Core.

g. NVIDIA Tensorflow example

We chose an example of NVIDIA Tensorflow to test the performance of the card. With the same parameter settings, the more pictures the graphics card trains in one second, the better the performance in terms of instance deep learning.

As you can see in the screenshot above, the RTX 5000 can process up to 441 images per second at full load.

In the screenshot above, the P5000 processes up to 194 images per second.

From the performance of data processed per second, the RTX 5000 is 2.2 times the speed of the P5000. It's a lot faster because of the call to Tensor Core for calculations. It can be seen that Tensor Core still plays a big role in the acceleration of deep learning. All High-end Quadro RTX cards have Tensor Core, so they can be used for the applications with graphics plus AI. For example, AI denoise, AI image recognition, or AI inference can be implemented while rendering.

Conclusion

The major features of the RTX 5000 graphics card are:

  • RT Core supports RTX real-time ray tracing rendering, which can be used to render a picture with Ambilight realism. The rendering speed is very fast, and it will be a good driving force in many visualized industrial design processes.
  • The Tensor Core can accelerate deep learning and AI inference. Today with the rapid development of artificial intelligence, many applications will become more powerful with the help of Tensor Core.
  • Traditional graphics application performance optimization and stability are retained and performance is vastly improved. Users that are still using the Fermi, Kepler, and Maxwell architectures can consider upgrading to the RTX series of graphics cards.