Making AMD GPUs competitive for LLM inference

Re:Making AMD GPUs competitive for LLM inference by greenknight on Wednesday August 09, 2023 @07:25PM (#37069767)

> AMD RX 580 was released in April 2018

It was actually Apr 18, 2017 -- https://en.wikipedia.org/wiki/Radeon_500_series

[ Parent | on HN ]

Re:Making AMD GPUs competitive for LLM inference by elabajaba on Thursday August 10, 2023 @12:43AM (#37072054)

And the 580 was just a rebranded 480, which was released in June 2016.

[ Parent | on HN ]

Re:Making AMD GPUs competitive for LLM inference by ktta on Thursday August 10, 2023 @03:20AM (#37073029)

This just makes the case for supporting it even further, because then they'd be supporting multiple years worth of hardware with just the effort for one uarch

[ Parent | on HN ]

Re:Making AMD GPUs competitive for LLM inference by imtringued on Thursday August 10, 2023 @08:20AM (#37075265)

Nothing like having a lottery of 2-6 years of support for your hardware to make your customers confident they are getting value out of the products they are buying.

The manufacturer can smugly proclaim they offered six years of support for a product that was on the shelf four years into the driver's lifecycle.

[ Parent | on HN ]

Re:Making AMD GPUs competitive for LLM inference by latchkey on Friday August 11, 2023 @01:42PM (#37091530)

Speaking from having run tens of thousands of both 580 and 480, they weren't just rebranded. Maybe on paper they seemed similar, but they didn't run similar.

[ Parent | on HN ]

Re:Making AMD GPUs competitive for LLM inference by bibanez on Saturday August 12, 2023 @10:18AM (#37100449)

I’m curious, where would you run so many gpus? Quality control?

[ Parent | on HN ]

Re:Making AMD GPUs competitive for LLM inference by latchkey on Saturday August 12, 2023 @11:30AM (#37101173)

7 datacenters in the US

[ Parent | on HN ]