Then their position doesn't make much sense to me! They used assertive wordings like the leader in perf/watt, the best in class in perf/watt, and the most advanced GPU, despite not having the performance or the power advantage, it's the first time I have seen them do this, it's like their marketing was dissociated from reality!
Have to be careful to ensure that "CPU limited" situations aren't in the data being presented.
No issues at all. Typical heavy load clocks are between 2.6 and 2.8 on a 4090 running stock clocks, mostly landing in 2.7-2.75 range.
I think nvidia has spies inside amd.. jensen hinted to that when he trashed vega
They don't need spies. They work in the same industry which means that they use the same tools and people are changing places from one IHV to another all the time. It is fairly easy to predict / model what the competition will be able to achieve if you know the base from which they are working and more or less the target which they are aiming at.
TlDw: The power bug is so bad the card starts massively underclocking the memory to maintain die power.
how tf did they ship this? Well I guess all those attempts at trying to parse out performance per silicon have to take this into account as well.Even at 450 watts it looses to a 4090 and in efficiency.. amd can't talk about efficiency anymore nvidia will be more efficient on the whole product stack
We know the efficiency of Navi 32/33 already?
Probably the GPU gets so unstable that memory downclocks due to stalling pipelines / uncorrectable errors etc. Considering it's Jay, he probably did a slapdash attempt at OC'ing and it eventually got so unstable this weird behaviour appeared. We have to wait until the people at igorslab or OCN get how to circumvent the new layer of (idiotic) blocks against power fiddling (why on Earth did AMD try to do this now when they are on a backstep I can't fathom)
