So It seems AMD will sell gpus with flashed LLM as ROM.https://www.servethehome.com/amd-to-acquire-taalas-for-model-specific-ai-inference-chips/
>>109638536Use case for hardware that gets obsolete after 2 months?
>>109638598It will cost 100 usd to replace the chip, rather than 5k per nvidicuck shit.
>>109638610It will likely cost more than you make in an entire decade. They're going to be low volume runs, and likely each system will have to be rack-scale to support models big enough to be useful.
>>109638690>Making a LLM into a physical chip requires more tech than building a 5090 RTXok
>>109638610The best this shit can run right now is llama 3.1 8B. If this ever goes somewhere and you seriously believe AMD won't milk it to kingdom come you need to be put down and your surviving family members forcibly sterilized to prevent any further mishaps.
>>109638737>It can't improve chudIt only need to load DeepSeek 4 and I'll buy it.
>>109638750Nigger if this improves there are ZERO reasons not to price it 0.01$ beneath whatever HXXX generation Nvidia offers by that point assuming either company will still be relevant by that point.
