![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
Lisa Su Saved AMD. Now She Wants Nvidia's AI Crown
- Thread starter erek
- Start date
NVidia is selling everything they make and there is still excess market demand. AMD will get market share just because their stuff will be available to buy - it would be hard for them to screw it up and NOT gain share.
Maybe a bit like AMD seem to always make really nice gain during crypto craze, if the demand stay crazy high anyone making anything that can ship and work as an opportunity, like previous time it can be short lived.
But that assuming that the current situation will not be short lived and be done by 2026 at best (with peak demand possibly calming down, used product a la A100 staying relevant for a very long time and creating a large in the field base, lot of top-near top fab projects around the world, what used to compete for the latest and (game console, video gaming card) greatest node stopping to do so.
Airbrushkid
2[H]4U
- Joined
- Aug 27, 2007
- Messages
- 2,430
Nvidia is doing fine on their own. They don't need her.
Nvidia is no longer able to sell to China. That is like 25% capacity that will go into other markets.
If anything is left front company that feed a China black market will buy them we can imagine, it is not like an front or will really want to use it AI start-up would be easy to distinguish.
Russia has continued to buy for 3 years banned stuff way easier to control that GPU in a market in which company that did not exist yesterday has suddenly a $800million GPU budget are perfectly legit and normal, so is them shutting down suddendly and selling all the GPUs, the VC money having been possible only because it will be easy to sell them at a really good price in case of failure.
China has been stockpiling Nvidia datacenter chips, reported to be 2 years of demand. Also, China is huge compared to Russia. You are making it sound way too easy to circumvent sanctions at such a scale.
yes it would be a fraction, just not all the china sales would be lost, i doubt they can just fill current demand for som of them, let alone future.
Nah she doesn't need a stupid icon like he does. These CEOs are not visionaries. They go to board meetings, and 20 and 30 somethings give them the hot great ideas and the CEOs undersigned them and get all the wealth and credit.
If I owned a tech giant I'd give incentive such as, your idea flourishes and makes us billions, you get 1% of the total profit as a bonus in the 3rd year of profit for your super profitable idea. You could be the frigging janitor for all that matters.
1% of a 2 billion profit is a suer damn nice bonus.
OutOfPhase
Supreme [H]ardness
- Joined
- May 11, 2005
- Messages
- 6,581
Of course they are generally not visionaries. But they need to make the hard calls on a million "great ideas" and figure out which will work out.
I think she's doing a pretty dang good job at that.
I was contracted out to some oil companies for a short while in the early 2000s and they had a program where if you noticed something they were actively doing that was costing them money that didn't need to be wasted if you could prove it and give them a plan they would pay you the difference it saved for a year, or at least a portion of it.
Xar
Limp Gawd
- Joined
- Dec 15, 2022
- Messages
- 227
With mutual respect towards one another.
Not sure, if I should post it here or under AI software forum:
A statement by Lamini dev:
If you are an AI startup blocked on GPUs, send me a note. At Lamini, we figured out how to use AMD GPUs, which gives us a relatively large supply compared to the rest of the market.
IMO, ROCm is 90% of the way there, but the market treats it like it is 0% of the way there. We filled in the 90%->100% gaps for LLM finetuning and inference with a team of a few good hackers.
It was essential to use MI class servers, e.g. MI210/MI250/MI300. I read that ROCm on Radeon/RDNA has gotten better recently, but last year it was not pretty and we have moved on completely to CNDA by now.
https://x.com/GregoryDiamos/status/1736663783635124343?s=20
A statement by Lamini dev:
If you are an AI startup blocked on GPUs, send me a note. At Lamini, we figured out how to use AMD GPUs, which gives us a relatively large supply compared to the rest of the market.
IMO, ROCm is 90% of the way there, but the market treats it like it is 0% of the way there. We filled in the 90%->100% gaps for LLM finetuning and inference with a team of a few good hackers.
It was essential to use MI class servers, e.g. MI210/MI250/MI300. I read that ROCm on Radeon/RDNA has gotten better recently, but last year it was not pretty and we have moved on completely to CNDA by now.
https://x.com/GregoryDiamos/status/1736663783635124343?s=20