AMD Strix Halo “FP11” APU Reference Platform Spotted With Massive 128 GB Memory Config

AMD's Strix Halo APUs might feature some massive memory pools based on the latest shipping logs which point out up to 128 GB capacities.

AMD's Strix Halo To Be An Enthusiast-Tier APU Platform: Featuring Powerful CPU, GPU Cores With Lots of Memory & Untapped Power

We have covered AMD's Strix Halo APUs in the past with the most recent post about them talking about the reference "Maple" Reference Evaluation Boards. These REVB products were discovered within the shipping logs from nbd.ltd. The last entries featured 32 GB and 64 GB configurations but the latest one doubles the memory pool which could be our hint at what the most premium configs would be getting.

Related Story AMD Rejects Tiny Corp’s Request of Benchmarking Instinct MI300X AI Accelerator On MLPerf, Raising New Questions

So starting with the details, the AMD Strix Halo APUs are said to be compatible with the FP11 (BGA) platform and will come in TDPs ranging from 55W and up to 130W. AMD is likely using the full untapped TDPs of these chips to evaluate their maximum potential but it would be awesome if we get chips that can support such high TDPs to push them to their limits since Strix Halo is designed for enthusiast platforms.

In terms of memory, the highest capacity that has been listed so far is up to 128 GB and that's a huge pool. Even most high-end desktop PCs these days don't feature more than 64 GB capacities and this is workstation territory so if AMD is going for larger pools of memory, that would be nice but it would also certainly mean that the cost of these high-end devices based on Strix Halo will be very high too. We also previously mentioned that the 128 GB memory is definitely not on-board or on-package like Intel's Lunar Lake but we are likely looking at soldered designs (at least for these REVB platforms).

It will be interesting if Strix Halo comes with a SO-DIMM DDR5 option but it will most likely be going the LPDDR5x route considering that 40 compute units iGPU which it's expected to