@I_loves_deep_nn Probably Global Foundries? Samsung is probably too big, I think

@9550pro Is this LP cores "compact" core Zen6c or like intel-style LPE cores?

@okyfirmansyah Seems like their engineer are compiler-brained talking in echo chamber then see every problem is compiler problem, just like oop engineer think everything in oop lol

@okyfirmansyah @VideoCardz I guess it is too much for client desktop (and likely laptop since there is LP core) to have numa arrangements, for laptop it make sense for quad channel memory controller since amd strix halo laptop has quad channel LPDDR5, for desktop maybe higher clocked memory support but idk

@x86deadandback Still dual channel LPDDR5X?

Yeah probably amd is already losing on NPU software to Intel. OpenVINO solution is maybe the easiest way to gain adoption for NPU. Honestly amd engineers and executives need to try what does it feel to develop an application

@AnhPhuH Yes my mistake I didn't notice that, only see G1a and thought it is strix halo model

@AnhPhuH wait it is HX 375, not 395 😅, but isn't hp G1a uses strix halo chipset?

@IanCutress Wow didn't know that qcom actually has more employees than nvidia

RT @chris_j_paxton: Deepseek has been incredibly transparent. $5.5m was the cost of the final training run, its in the paper, who is still…

@highyieldYT @Apple @techinsightsinc Ah yes correct of course they can do that, I didn't think of that

@hkproj @euaccofficial ah i see, probably i didn't understand that correctly