AMD Vega Memory Architecture Q&A With Jeffrey Cheng

At the AMD Computex 2017 Press Conference, AMD President & CEO Dr. Lisa Su announced that AMD will launch the Radeon Vega Frontier Edition on 27 June 2017, and the Radeon RX Vega graphics cards at the end of July 2017. We figured this is a great time to revisit the new AMD Vega memory architecture.

Now, who better to tell us all about it than AMD Senior Fellow Jeffrey Cheng, who built the AMD Vega memory architecture? Check out this exclusive Q&A session from the AMD Tech Summit in Sonoma!

Updated @ 2017-06-11 : We clarified the difference between the AMD Vega’s 64-bit flat address space, and the 512 TB addressable memory. We also added new key points, and time stamps for the key points.

Originally posted @ 2017-02-04

Don’t forget to also check out the following AMD Vega-related articles :


The AMD Vega Memory Architecture

Jeffrey Cheng is an AMD Senior Fellow in the area of memory architecture. The AMD Vega memory architecture refers to how the AMD Vega GPU manages memory utilisation and handles large datasets. It does not deal with the AMD Vega memory hardware design, which includes the High Bandwidth Cache and HBM2 technology.


AMD Vega Memory Architecture Q&A Summary

Here are the key takeaway points from the Q&A session with Jeffrey Cheng :

  • Large amounts of DRAM can be used to handle big datasets, but this is not the best solution because DRAM is costly and consumes lots of power (see 2:54).
  • AMD chose to design a heterogenous memory architecture to support various memory technologies like HBM2 and even non-volatile memory (e.g. Radeon Solid State Graphics) (see 4:40 and 8:13).[adrotate group=”2″]
  • At any given moment, the amount of data processed by the GPU is limited, so it doesn’t make sense to store a large dataset in DRAM. It would be better to cache the data required by the GPU on very fast memory (e.g. HBM2), and intelligently move them according to the GPU’s requirements (see 5:40).
  • The AMD Vega’s heterogenous memory architecture allows for easy integration of future memory technologies like storage-class memory (flash memory that can be accessed in bytes, instead of blocks) (see 8:13).
  • The AMD Vega has a 64-bit flat address space for its shaders (see 12:0812:36 and 18:21), but like NVIDIA, AMD is (very likely) limiting the addressable memory to 49-bits, giving it 512 TB of addressable memory.
  • AMD Vega has full access to the CPU’s 48-bit address space, with additional bits beyond that used to handle its own internal memory, storage and registers (see 12:16). This ties back to the High Bandwidth Cache Controller and heterogenous memory architecture, which allows the use of different memory and storage types.

  • Game developers currently try to manage data and memory usage, often extremely conservatively to support graphics cards with limited amounts of graphics memory (see 16:29).
  • With the introduction of AMD Vega, AMD wants game developers to leave data and memory management to the GPU. Its High Bandwidth Cache Controller and heterogenous memory system will automatically handle it for them (see 17:19).
  • The memory architectural advantages of AMD Vega will initially have little impact on gaming performance (due to the current conservative approach of game developers). This will change when developers hand over data and memory management to the GPU. (see 24:42).[adrotate group=”2″]
  • The improved memory architecture in AMD Vega will mainly benefit AI applications (e.g. deep machine learning) with their large datasets (see 24:52).

Don’t forget to also check out the following AMD Vega-related articles :

Go Back To > Computer Hardware + Systems | Home


Support Tech ARP!

If you like our work, you can help support our work by visiting our sponsors, participating in the Tech ARP Forums, or even donating to our fund. Any help you can render is greatly appreciated!

5 thoughts on “AMD Vega Memory Architecture Q&A With Jeffrey Cheng

  1. serpico

    “The AMD Vega has a 64-bit flat address space, giving it potentially up to 512 TB of addressable memory.” 64-bit byte addressable addresses give a approx 18*10^18 Bytes (thats about 10,000 times more than 512 TB). Did you mean 49-bit addressing? Or maybe there is a split like with current x86-64 processors where the processors are 64-bit but only 48bits of memory address lines are provided?

    1. Dr. Adrian Wong Post author

      It has a 64-bit address space for its shaders. But it will have access to just 512 TB of addressable memory. That means it’s restricted to 49-bit addressing, as you pointed out.

      We just updated the article to reflect that, as well as added new details. Thanks for the heads-up! 😀

  2. Pingback: Daily Roundup: 2017-02-03 -

  3. Pingback: The AMD Radeon RX 580 Performance Comparison - Tech ARP

  4. Pingback: Daily Roundup: 2017-06-12 -

Leave a ReplyCancel reply