AMD's Next-Gen Van Gogh APU Shows '256-Bit DDR5' Memory Support

AMD
(Image credit: AMD)

AMD's next-generation client accelerated processing unit (APU) supports a 256-bit DDR5 memory interface, according to a Linux Kernel boot log published by AMD itself. If the information is correct, then AMD's upcoming APU could offer a massive boost in memory-intensive workloads, such as graphics. 

AMD's Van Gogh is a rather enigmatic APU that is believed to integrate general-purpose cores based on the Zen 2 microarchitecture as well as a graphics engine based on the RDNA2 architecture, which makes it similar to the custom system-on-chips (SoCs) that AMD developed for Microsoft's Xbox Series X as well as Sony's PlayStation 5. The processor is expected to be made using TSMC's N7 process technology, though this information does not come from AMD. To add some more mystery, AMD's Van Gogh was rumored to feature a DDR5 memory subssystem and then strangely lost this capability, but gained LPDDR5 support.

This month, one of AMD's software engineers posted a Linux Kernel boot log for AMD's quad-core Van Gogh APU (marked as the AMD Eng Sample: 100-000000405-03_35/24_N, family: 0x17, model: 0x90, stepping: 0x1) which says that the processor supports a 256-bit DDR5 memory interface. 

A 256-bit memory interface is nothing new for high-end desktop CPUs as well as game consoles, but for typical processors for client PCs such a wide DRAM interface is considered an overkill. Meanwhile, if AMD wants to offer a breakthrough graphics performance with an APU, a 256-bit DDR5 interface might be be a logical choice. What is somewhat illogical is to use four Zen 2 general-purpose cores with a high-end APU.

From an operating system point of view the type of memory used by the platform is not very important, which is why there are modern CPUs that can work equally well with DDR4 as well as LPDDR4X. DDR5 has numerous similarities with LPDDR4 but with a different addressing range than existing types of memory, and LPDDR5 and DDR5 also share similarities. It's possible that a development system could incorrectly detect LPDDR5 as DDR5 (especially with a pre-production BIOS). That could also result in the board detecting multiple 16-bit LPDDR5 or 32-bit DDR5 memory channels as several 64-bit DDR4 channels, showing a '256-bit DDR5 interface.'

So we're back to square one, and Van Gogh's memory interface remains yet another mystery about this APU. Given the 4-core CPU, we suspect this isn't actually a 256-bit DDR5 interface. Most likely it's a 128-bit interface, possibly even a mobile optimized 64-bit interface with LPDDR5. Or, it could actually be an APU with more memory bandwidth and a larger GPU — anything is possible at this point.

The development motherboard used to run the Van Gogh APU is codenamed Chachani-VN, but not a lot is known about its specifications. Meanwhile, there are some other Van Gogh platform codenames revealed before, including Celadon (an FP6 mobile platform), Mytle (a desktop AM4 platform), and Artistic (an unknown desktop platform). Only the Celadon is mentioned in the boot log, so we are most likely dealing with a mobile platform.

Keeping in mind that AMD's Van Gogh APU has not been formally announced and AMD does not comment on unreleased products, so it is impossible for us to get any official details about the processor.

Anton Shilov
Contributing Writer

Anton Shilov is a contributing writer at Tom’s Hardware. Over the past couple of decades, he has covered everything from CPUs and GPUs to supercomputers and from modern process technologies and latest fab tools to high-tech industry trends.

  • TheJoker2020
    What utter gibberish :tearsofjoy::tearsofjoy::tearsofjoy:
    Reply
  • InvalidError
    256bits LPDDR5 would enable it to approach 1650S/RX5500 level performance and would be fine for a Ryzen 3300X-class CPU. It would make sense for an embedded platform SoC with sufficiently good entry-level graphics to skip tossing in an entry-level dGPU.
    Reply
  • TheJoker2020
    InvalidError said:
    256bits LPDDR5 would enable it to approach 1650S/RX5500 level performance and would be fine for a Ryzen 3300X-class CPU. It would make sense for an embedded platform SoC with sufficiently good entry-level graphics to skip tossing in an entry-level dGPU.
    This is a big IF, and as the rambling above notes, this could be entirely false, indeed it could even have a 64-bit interface.!
    Reply
  • zodiacfml
    plausible. gaming SoC for portable gaming
    Reply
  • InvalidError
    zodiacfml said:
    plausible. gaming SoC for portable gaming
    Or a NUC, an HTPC, an STB, a smart-TV, an AiO PC, notebook, laptop, etc. Anything that could benefit from having decent gaming/GFX chops.

    An embedded platform with memory soldered on the motherboard would also skip many of the additional costs of DDR5 DIMMs: don't need data buffer chips when the bus is single-point-to-single-point with no memory slot or CPU socket to pass through, may also be able to skip the command-address buffer if all the DRAM chips for one channel are in one DRAM stack.
    Reply