Sure, but when you jump in the middle of another 2 people’s conversation to say you disagree, it’s not up to the other people to automatically know that you’re actually disagreeing with something different to what they’re discussing lol
Nvidia AI gpu’s don’t have <16GB of RAM. They have 48+. Some have over 180GB. They were talking about desktop GPUs, the RTX series. It makes no sense to congratulate Chinese GPU makers for having 16GB of RAM and being angry at Nvidia for having 188GB - not a typo, they have one hundred and eighty-eight GB of RAM - like their H100 NVL.
The <16GB complaint is a common and long standing complaint about their desktop GPUs.
and VRAM limitations almost certainly don’t apply to chinese GPUs: the process node is the only limitation in chinas manufacturing capability
i would absolutely not be surprised if they have a 256GB+ card that is equal to an 8yo nvidia/amd card at 1/5th the price and just dump more in… for the same reason that GPUs are better than CPUs for models, many GPUs is better than a single fast GPU
and that’s still only barely relevant because as i said previously, workstation GPUs etc don’t need that - there are plenty of workloads that fit the bill for a card from 8 years ago, and they never mentioned anything about large cards or ML workloads or gaming
no, they are not
They definitely are, unless you think that they’re saying that nvidias big boy AI gpus come with less than 16GB of RAM.
i’m saying that 8gb GPUs are still useful as workload accelerators in workstations etc
Cool, you’re not the original person I replied to.
right and this is a public message board in which multiple people can carry on a conversation
Sure, but when you jump in the middle of another 2 people’s conversation to say you disagree, it’s not up to the other people to automatically know that you’re actually disagreeing with something different to what they’re discussing lol
that is literally what you’re discussing… they never said anything about gaming: they just said GPUs… just admit you made an assumption mate
Nvidia AI gpu’s don’t have <16GB of RAM. They have 48+. Some have over 180GB. They were talking about desktop GPUs, the RTX series. It makes no sense to congratulate Chinese GPU makers for having 16GB of RAM and being angry at Nvidia for having 188GB - not a typo, they have one hundred and eighty-eight GB of RAM - like their H100 NVL.
The <16GB complaint is a common and long standing complaint about their desktop GPUs.
there was no mention of desktop, OR RTX
and VRAM limitations almost certainly don’t apply to chinese GPUs: the process node is the only limitation in chinas manufacturing capability
i would absolutely not be surprised if they have a 256GB+ card that is equal to an 8yo nvidia/amd card at 1/5th the price and just dump more in… for the same reason that GPUs are better than CPUs for models, many GPUs is better than a single fast GPU
and that’s still only barely relevant because as i said previously, workstation GPUs etc don’t need that - there are plenty of workloads that fit the bill for a card from 8 years ago, and they never mentioned anything about large cards or ML workloads or gaming