Anonymous
8/19/2025, 8:59:17 PM
No.103667107
>>103656293
For a PC build, you have to go workstation or Mac for R1 at Q4, that is 420GB. If you run IQ2_M, that's more reasonable with 217GB needed. GLM 4.5 can be run at Q4_K_M taking around the same size at 219GB. Also, you aren't going to get more than dual channel and speeds will be abysmal, 2-3 tokens/s. Building the right workstation or buying a mac will get you quadruple that but you'll pay other costs like maintenance of the workstation and slow prefill on Mac. If you go full max out on the right motherboard to fit around 6 3090s, you can run it too at around 15 tokens speeds but the power bill...
For a PC build, you have to go workstation or Mac for R1 at Q4, that is 420GB. If you run IQ2_M, that's more reasonable with 217GB needed. GLM 4.5 can be run at Q4_K_M taking around the same size at 219GB. Also, you aren't going to get more than dual channel and speeds will be abysmal, 2-3 tokens/s. Building the right workstation or buying a mac will get you quadruple that but you'll pay other costs like maintenance of the workstation and slow prefill on Mac. If you go full max out on the right motherboard to fit around 6 3090s, you can run it too at around 15 tokens speeds but the power bill...