CUDA Cores vs Stream Processors Explained

If you have ever looked at the specifications of an Nvidia and AMD Graphics cards then you must have seen the terms CUDA Cores and Stream Processors listed there. Both these terms are same in some respect but they are also quite different if you go into the core technical aspect.

CUDA Cores and Stream Processors are one of the most important parts of the GPU and they decide how much power your GPU has. Here in this post I am going to explain CUDA Cores and Stream Processors in very simple words and also list down that various graphics cards that support them.

cuda-cores-vs-stream-processors

What are CUDA Cores & Stream Processors?

CUDA Cores & Stream Processors are also called processor cores or pixel pipelines of a GPU. You have already heard about multi-core processors such as dual core processors, quad-core processors, hexa core processors and octa core processors in computers and mobile phones. These multi-core processors consist of very few cores that performs various individual tasks and increases the parallel processing and multi-tasking. This ultimately leads to better performance and efficiency.

cuda-cores-stream-processors

Similarly a GPU or Graphics Processing Unit is made up of hundreds and thousands of cores that perform various complex operations and computations related to pixels (pixel processing). These Cores are known as CUDA Cores or Stream Processors. NVIDIA calls them CUDA Cores and in AMD they are known as Stream Processors. Generally these Pixel Pipelines or Pixel processors denotes the GPU power. More is the number of these cores the powerful will be the card is, given that both the cards have same GPU Architecture. Both CUDA Cores & Stream Processors have almost same use in graphics cards from Nvidia and AMD but technically they are different. This is because of the difference in the GPU Architecture of both Nvidia and AMD graphics cards.

gpu-multi-cores

CUDA (Compute Unified Device Architecture) is mainly a parallel computing platform and application programming interface (API) model by Nvidia. It accesses the GPU hardware instruction set and other parallel computing elements. The physical individual cores inside the GPU that execute CUDA API are known as CUDA Cores. Graphics cards having Tesla, Fermi, Keper, Maxwell, Pascal and Volta GPU architectures supports CUDA.

Must Read: Important Technical Terms related to Graphics Cards

What are Tensor Cores?

Volta GPU architecture based graphics card Nvidia Titan V comes with Tensor Cores (640 tensor cores) in addition to CUDA Cores. These Tensor Cores help in AI, Deep Learning / Machine Learning and can deliver around 110 TeraFLOPS (TFLOPS) of performance. This is over a 5X performance increase over the older Pascal GPU architecture from Nvidia.  Tensor Cores are responsible for evaluating large complex mathematical problems and computations. Nvidia Titan V graphics cards also comes with 12GB HBM2 memory which is currently one of the fastest memory around.

tensor-cores

Note: Pascal GPU architecture does not support Tensor Cores .

CUDA Cores vs. Stream Processors

It must be noted that CUDA Cores is not equivalent to Stream Processors in terms of power and number. That means two graphics cards having same number of CUDA Cores, Stream Processors, Memory, Clock Frequencies cannot have same performance. This is again due to difference in the working of GPU architecture of the graphics cards. If we have two 10 series Graphics cards G1 and G2 with same memory and clock from Nvidia having 380 and 480 CUDA Cores then G2 will be more powerful because of the extra 100 CUDA Cores.

So,

G2 > G1*

*in terms of performance given that the both cards are in same family or have same GPU architecture.

So, if CUDA Cores of one graphics card is more than the other graphics card given that both graphics cards are in the same family then the graphics card with higher number of CUDA Cores will be more powerful.

How many Stream Processors equals to CUDA  Cores?

If we have a Nvidia Graphics card with 500 CUDA Cores and an AMD Graphics Card with 600 Stream Processors then we cannot conclude that AMD Graphics Card is more powerful. Here to find out which graphics card is more powerful we have put them to real word gaming benchmarks. The card that scores more frame rates or fps in games will be the winner.

So,

X no. of CUDA Cores ≠ X no. of Stream Processors

 

Some Important Points to Remember

  • CUDA Cores are Nvidia’s GPU Muti-core units
  • Stream Processors are AMD’s GPU Muti-core units
  • We cannot equate CUDA Cores and Stream Processors
  • GPU Architecture matters a lot along with number of GPU Cores

Final Words

At last I would like to say that I have tried my best to clearly explain about the similarities and differences between CUDA Cores and Stream Processors. Both CUDA Cores and Stream processors are nothing but processor cores in a GPU. Their working, functions, shape and size may differ according to the GPU architecture. If you have any queries regarding them then you can ask me by leaving a comment below.

17 Comments

  1. dunlie April 29, 2017
    • Akshat Verma April 29, 2017
  2. dunlie April 29, 2017
    • Akshat Verma April 29, 2017
      • dunlie May 1, 2017
      • Akshat Verma May 1, 2017
  3. dunlie May 1, 2017
    • Akshat Verma May 1, 2017
      • dunlie May 1, 2017
      • Akshat Verma May 1, 2017
  4. dunlie May 1, 2017
  5. Brijesh May 4, 2017
  6. Saiham Chwodhury July 5, 2017
    • Akshat Verma July 5, 2017
      • Saiham Chowdhury July 5, 2017
  7. jimmy May 6, 2018
    • Akshat Verma May 6, 2018

Leave a Reply