Rtx5000 computing power
recently, I saw a test that the rtx3070 notebook only earns 1.2 yuan an hour.
1050ti computing power is less than 1 / 4 of 3070, which is about 24%
power consumption is 35% of 3070.
energy consumption is not cost-effective. One hour's income is definitely less than 1 yuan, except for the electricity charge. Once the loss of notebook is eliminated, it is definitely a loss
can surpass, but it is not recommended to play games
RTX 8000 is a high-end professional GPU based on NVIDIA Turing architecture and NVIDIA RTX platform. It can provide hardware accelerated ray tracing, deep learning and advanced coloring functions for professionals. It is built for professionals. It is not recommended to play with professional cards
Quadro RTX 8000 is the most advanced GPU of NVIDIA Turing Architecture series, with ultra fast on-board video memory technology and optimized software driver for professional applications
RTX 8000 includes 72 RT cores, which can accelerate realistic ray tracing rendering. Meanwhile, 576 tensor cores are added, which can provide deep learning performance up to 130.5 tflops
using NVIDIA nvlink & 8482; The technology can also connect two RTX 8000 to share GPU and video memory. The performance can be expanded by up to 100 GB / s connection performance and up to 96 GB gddr6 combined video memory, so as to provide processing support for super large scene files, rendering and AI
Moreover, Quadro RTX 8000 supports AI optimization software, such as deep learning framework caffe2, mxnet, cntk, tensorflow, etc., which can greatly speed up training time and improve multi node training efficiency. GPU acceleration function libraries such as cudnn, cublas, tensorflow, etc{rrrrrrr}
extended data:
problems in manufacturing process:
integrated circuits develop to nano scale process, constantly approaching the physical limit, resulting in the so-called red wall problem:
first, line delay is more and more important than gate delay. Long line not only has the problem of transmission delay, but also has the problem of energy consumption
Secondly, the feature size is so small that defects in chip manufacturing are inevitable. We should study fault tolerance and error avoidance technology from three aspects: defect tolerance, fault tolerance and error tolerance Thirdly, leakage current and power consumption become very important, and self-management technology of power consumption should be adopted. Modern graphics processor chips have made remarkable progress in overcoming the red wall problem in several aspects: using a large number of regular SIMD array structure; Its distributed memory is close to the operation unit, which reces the long-term effect; Its hardware multithreading conceals part of the impact of storage latencyNVIDIA® Quadro RTX™ 4000 adopts NVIDIA Turing & 8482; Architecture and NVIDIA RTX platform can provide excellent performance and function in single slot PCI-E configuration
it is based on tu106 core, the overall specification is similar to RTX 2070 graphics card, 2304 CUDA cores, 36 RT cores, and 288 tensor cores, floating-point performance is 7.1tflops, video memory capacity is 8GB gddr6, but the frequency is reced to 6.5gbps
Quadro RTX 4000 has been specially designed, constructed and tested to adapt to the demanding professional visual computing workflow. Quadro has been verified in OEM workstation and has been recognized by professional software application. It can provide professionals with the required performance, stability and reliability. As shown in the following figure:
extended data:
graphics card type: workstation graphics card; Video memory type: gddr6; Graphics card type: professional
graphics chip: Quadro RTX 4000; Video memory capacity: 8GB
bit width: 256bit; Power interface: 8pin
other in addition to the motherboard, memory, solid-state hard disk, chassis power supply, etc., all use big brands, the total price should be more than 5000, if you choose a niche brand, control below 5000 is also OK
Rtx2070s
use PS, AI, C4d, AE, PR, occasionally use some engineering 3D software, you can choose rtx2070s, Adobe family generally as long as the display memory is large, CUDA (a card has no CUDA, only OpenCL). Unless the project with 10bit quality or 16 bit color depth or above (i.e. reaching broadcast or movie level), medium and high level game graphics card is OK
The same is true for C4d, which can't meet the requirement of double precision (after 6 decimal places), so it doesn't need Quadro grade professional card. CAD generally takes the value of 6 digits after the decimal point. Adobe series software can't use professional graphics card. Take PR as an example, professional graphics card optimizes the performance of OpenGL, and the acceleration of PR graphics card comes from CUDA
extended data:
C4d, whether using MAC or PC, can run Adobe Photoshop computer. It's just the speed of rendering time, and rendering is mainly divided into CPU render and GPU render
The difference is that rtx8000 is 48gb video memory, rtx6000 is 24GB video memory, rtx5000 is tu104 full version core, with 3072 CUDA cores