How many GPUs do you have on your local system/server/AI PC?
Posted by panchovix@reddit | LocalLLaMA | View on Reddit | 14 comments
Posted by panchovix@reddit | LocalLLaMA | View on Reddit | 14 comments
mukz_mckz@reddit
1x5090 1x5060ti 16gb
Legal-Ad-3901@reddit
16 mi50s, 2 strix halo, 5060ti, rtx 6000 pro Blackwell. They need AA for llm
seamonn@reddit
What's llama.cpp support for these, these days? Can you run latest models?
FullstackSensei@reddit
Been great since last September or so. I run "only" six in one machine and have no complaints besides PP speed on 200B+ models.
sophlogimo@reddit
Single GPU... but technically an APU, because Strix Halo.
Kahvana@reddit
2x ASUS PRIME 5060 Ti 16GB's on Asus ProArt X870E Creator Wifi with AMD Ryzen 5 9600X.
Decent performance and very energy efficient for it's price, parallel tensor has been a real big boon from llama.cpp.
I wonder how it compares to 2x RX 9060 XT 16GB for energy and performance.
Plane-Marionberry380@reddit
total VRAM plus watts at the wall is probably the real goblin metric. GPU count makes a 4x 3060 box look comparable to a single 6000 Pro until the power bill walks in holding a knife.
Bulky-Priority6824@reddit
2 5060ti 16gb running infernece 3070 on frigate 5070ti game but I never game so it just basically idles 2 3060 12gb in a shoebox that the 5060ti replaced
koushd@reddit
total vram probably a better question.
panchovix@reddit (OP)
For sure! Will do in some days.
rpkarma@reddit
A single GB10 based system with 128GB of accessible shared RAM. Though I’m likely to buy a second lol
DeProgrammer99@reddit
Technically three, but who counts the Ryzen 7600's graphics processor?
general_sirhc@reddit
Now do a poll of dedicated model memory.
E.g. VRAM for GPU users or Unified memory
panchovix@reddit (OP)
Me myself, 8 GPUs (4x5090, 2x4090 and A40/A6000). Thinking of selling 3x5090 for a single RTX 6000 PRO, but they go for 14000USD here on Chile (and got each 5090 at 2500USD time ago).