EVGA

AMD boost Radeon's ML performance by up to 4.4x with TensorFlow-DirectML's production

Author
Post
Essentials Only Full Version
rjohnson11
EVGA Forum Moderator
  • Total Posts : 102262
  • Reward points : 0
  • Joined: 2004/10/05 12:44:35
  • Location: Netherlands
  • Status: online
  • Ribbons : 84
2021/09/16 00:14:39 (permalink)
AMD boost Radeon's ML performance by up to 4.4x with TensorFlow-DirectML's production release | OC3D News (overclock3d.net)
 
Earlier this month, Microsoft released TensorFlow-DirectML version 1.15.5, the production release of DirectML, which alongside AMD's latest Radeon Software drivers, delivered the company some major performance boosts on their latest GPU hardware. TensorFlow-DirectML was created by Microsoft and co-engineered with the help of AMD, Intel and Nvidia. It is designed to enable hardware-accelerated training across DirectX 12 graphics cards, creating a general-purpose acceleration platform for developers to use across a wide range of systems. 

AMD has been working with Microsoft to optimise TensorFlow-DirectML with its latest release and optimise its Radeon Software stack (GPU driver) to deliver the strongest possible performance in AI workloads. When comparing their new version of TensorFlow-DirectML (version 1.15.5) and their AMD Radeon Software 21.8.1 drivers with TensorFlow-DirectML's previous version (version 1.15.4) with AMD's older Radeon Software 21.5.2 drivers, AMD has seen staggering performance gains as large as 4.4x with their RX 6900 XT graphics card. 
 
Across various TensorFlow-DirectML workloads, AMD has seen greater than 3x performance gains across a variety of workloads on their RX 6000 series graphics cards. 

Today, AMD's RX 6600 XT delivers more performance than AMD's RX 6900 XT did before the release of TensorFlow DirectML's latest version and AMD's latest Radeon Software drivers. That's a big deal for AMD and all future users of TensorFlow-DirectML. Microsoft is working with developers and hardware vendors to continue optimising TensorFlow-DirectML, and Microsoft plans to deliver the same benefits to the TensorFlow 2 codebase. Microsoft has discussed their plans to TensorFlow-DirectML in more detail here. 
 
When looking at combined TensorFlow-DirectML performance in a combined ML training and inferencing workload, AMD's latest Radeon RX 6000 series graphics cards have again delivered greater than 3x performance benefits. These changes have come thanks to AMD's work on their Radeon Software drivers and Microsoft's efforts to optimise TensorFlow-DirectML since its initial preview release. 

TensorFlow-DirectML is available to deliver AI hardware acceleration in Windows and in the Windows Subsystem for Linux (WSL). More information about this project is available here. 
 
The increased gains will definitely benefit developers. 
 


AMD Ryzen 9 7950X,  Corsair Mp700 Pro M.2, 64GB Corsair Dominator Titanium DDR5  X670E Steel Legend, MSI RTX 4090 Associate Code: H5U80QBH6BH0AXF. I am NOT an employee of EVGA

#1

1 Reply Related Threads
    Cool GTX
    EVGA Forum Moderator
    • Total Posts : 30983
    • Reward points : 0
    • Joined: 2010/12/12 14:22:25
    • Location: Folding for the Greater Good
    • Status: online
    • Ribbons : 122
    Re: AMD boost Radeon's ML performance by up to 4.4x with TensorFlow-DirectML's production 2021/09/16 14:50:46 (permalink)
    good to see the AMD software mature, nice jump

    Learn your way around the EVGA Forums, Rules & limits on new accounts Ultimate Self-Starter Thread For New Members

    I am a Volunteer Moderator - not an EVGA employee

     https://foldingathome.org -->become a citizen scientist and contribute your compute power to help fight global health threats

    RTX Project EVGA X99 FTWK Nibbler EVGA X99 Classified EVGA 3080Ti FTW3 Ultra


    #2
    Jump to:
  • Back to Mobile