Today's server GPUs are typically connected by the PCI Express (PCIe) bus, which provides a communication bandwidth of 12 GB per second. However, NVIDIA's new NVLink ultrafast interconnect technology is set to replace PCIe, delivering 5 to 12 times more bandwidth. NVLink allows GPUs to communicate directly with each other, significantly accelerating large-scale simulations. Additionally, NVLink enables high-speed communication between GPUs and CPUs, allowing both processor types to access system memory at the same speed. This results in further benefits for accelerated computing, enhancing overall system performance.


each node A100, 8 A100s are connected using NVlink and 2nd set of 8 A100s are connected through NVswitch

NVIDIA technology in our supercomputer: NVLink and its associated technology, NVSwitch.

  • NVLink: Interconnects up to eight GPUs in our DGX systems, providing 10 times the speed of traditional PCIe connections.
  • NVSwitch: Extends NVLink's capabilities, allowing interconnection of up to 16 A100 GPUs, achieving up to 10 petaflops of compute power.

Key Benefits:

  • Enhanced Performance: NVLink offers significantly faster interconnectivity, boosting processing capabilities.
  • Scalability: Enables scaling workloads across multiple GPUs efficiently.
  • High Memory Bandwidth: Each GPU provides 2 terabytes of memory bandwidth, capable of processing 82 4K videos per second.

Together, NVLink and NVSwitch provide world-class performance and scalability for parallel processing

To understand how the claim of processing 2800 4K videos per second translates into a bandwidth of 2 terabytes per second (TB/s), let's break down the calculations:

How many videos can this process?

Specifications of a 4K Video

  1. Resolution: 4K resolution typically means 3840 x 2160 pixels.

Standard 4K UHD (Ultra High Definition): The most common 4K resolution is 3840 x 2160 pixels. This is exactly four times the pixel count of 1080p (1920 x 1080 pixels), providing a much sharper image.

  1. Color Depth: Assume a color depth of (8bits x3 channels) or 24 bits per pixel (8 bits per channel for RGB).
  2. Frame Rate: A common frame rate for 4K video is 30 frames per second (fps).

Data Rate Calculation for a Single 4K Video

To calculate the data rate required for a single 4K video:

  1. Pixels per Frame:
  2. 3840×2160=8,294,400 pixels
  3. Bits per Frame:
  4. 8,294,400 pixels×24 bits per pixel=199,065,600 bits
  5. Bytes per Frame (since 1 byte = 8 bits):
  6. 199,065,600 bits/8=24,883,200 bytes
  7. Bytes per Second:
  8. 24,883,200 bytes×30 frames per second=746,496,000 bytes per second
  9. Convert to megabytes per second (MB/s) (since 1 MB = 1,048,576 bytes):
  10. 746,496,000/1,048,576≈712.5 MB/s

Total Bandwidth for 82 4K Videos

To find the total bandwidth required for 2800 4K videos:

712.5 MB/s per video×2800=1995000MB/s

Convert to gigabytes per second (GB/s) (since 1 GB = 1,024 MB):

1995000 MB/s /1,024≈1948 GB/s

Comparison to 2 TB/s Bandwidth

2 terabytes per second is equivalent to 2,048 gigabytes per second (since 1 TB = 1,024 GB). The key specification here is the memory bandwidth, which is the rate at which data can be read from or written to the A100 GPU's memory.


approx 2800 4k video can be processed per sec.

This calculation indicates that 2 TB/s can theoretically handle approximately 2800 4K videos per second.



Older Post
Newer Post
--> -->