Thanks Chester, another great dive and (for me) down memory lane with Trinity! One aspect of modern APUs I am curious about is if and how iGPU-assisted compute is actually implemented nowadays. Judging by the now-ubiquitous NPU blocks in designs from both AMD and Intel, the idea of GPU-assisted compute seems to have been abandoned (outside AI), but is it ? Would love to hear your and anyone else's take!
I haven't seen that much GPU compute offload even today, except for machine learning stuff and hardware accelerated video encode/decode. Obviously still used for a lot of scientific applications (plenty of supercomputer clusters come with GPUs), but that's not on client.
Fun fact, Xbox Series APUs still has Garlic & Onion. Another one in a string of dumb Microsoft hardware decisions. So you get the double whammy of having to manually segment memory on Series S and it only has 7GBs of RAM.
Loved my Llano A6-3400m. Gamed on it.
Thanks Chester, another great dive and (for me) down memory lane with Trinity! One aspect of modern APUs I am curious about is if and how iGPU-assisted compute is actually implemented nowadays. Judging by the now-ubiquitous NPU blocks in designs from both AMD and Intel, the idea of GPU-assisted compute seems to have been abandoned (outside AI), but is it ? Would love to hear your and anyone else's take!
I haven't seen that much GPU compute offload even today, except for machine learning stuff and hardware accelerated video encode/decode. Obviously still used for a lot of scientific applications (plenty of supercomputer clusters come with GPUs), but that's not on client.
Thanks Chester! Yes, that's my impression as well.
Fun fact, Xbox Series APUs still has Garlic & Onion. Another one in a string of dumb Microsoft hardware decisions. So you get the double whammy of having to manually segment memory on Series S and it only has 7GBs of RAM.