There is always financing, there are always people willing to go to the competitor at some wage, there is always a way if the leadership wants to.
If it was just a straight up fab bottleneck? Yeah maybe you buy that for a year or two.
“During Q1, Nvidia reported $5.6 billion in cost of goods sold (COGS). This resulted in a gross profit of $20.4 billion, or a margin profile of 78.4%.”
They literally bought Xilinx for their software engineering team. That's at least a thousand firmware engineers and software engineers focused on software stack improvements. That was two years ago. And on top of Xilinx they've been hiring staff like crazy for years now.
The issue was that they basically let everyone go who wasn't building hardware for their essential product lines (CPU & GPU) other than a skeleton crew to keep the software at least mostly functioning. And as much as this seems like it was a bad decision, AMD was probably weeks from bankruptcy by the time they got Zen out the door even despite doing this. Had they not done so, they'd almost certainly closed up entirely.
So for the last ~5 years minimum now they've been building back their software teams and trying to recuperate what they lost in institutional knowledge. That all takes time to do even if you hire back twice as many engineers as you lost.
And so now we are here. Things are clearly improving but nowhere near acceptable yet. But there's a trend of improvement.
> How long am I supposed to wait, as my still-modern AMD GPU sits still-unsupported?
Which GPU do you have? At least according to these docs, on linux the upper chunk of RDNA3 is supported officially but from experience, basically all 6xxx or 7xxxx cards are unofficially supported if you build it for your target arch. 5xxx cards get the short end of the stick and got skipped (they were a rough launch) but Radeon VII cards should also still be officially supported (with support shifting to unofficial status in the next release).
And given that ROCm is pretty core to AMD's support for the windows AI stack (via ONNX), you can assume any new GPUs released from here on out will be supported.
It's 5xxx. And "rough launch" is not an excuse. They've had plenty of time. Is it that different from the other RDNA cards?
The unofficial support for so many cards is not a good situation either.
Edit: Actually, no, I know it's not that different, because some versions of ROCm largely work on RDNA1 if you trick them. They are just refusing to do the extra bit of work to patch over the differences.
I mean it apparently works on RDNA1 now after some effort but they never really attempted to support it because they initially only supported workstation RDNA cards but they didn't have a workstation RDNA1 release.
I wish they had comprehensive support for basically all recent GPU releases but tbh I'd rather they focus on perfecting support for the current and upcoming generations than spread their efforts too thin.
And ideally with time backports to the older cards will come with time but it's really not a priority over issues on the current generation because those RDNA1 cards were never actually supported in the first place.
Financing is not the bottleneck. Organizational capacity might well be, though. As an organization, AMDs survival depended not on competing with nVidia but on competing with Intel. Now they are established, in what must be one of the greatest come from behind successes in tech history. 8 years ago, Intel was worth 80 times as much as AMD, today AMD has surpassed them:
> With margins that high?
There is always financing, there are always people willing to go to the competitor at some wage, there is always a way if the leadership wants to.
People love to pop-off on stuff they really know anything about. Let me ask you: what financing do you imagine is available? Like literally what financing do you propose for a publically traded company? Like do you realize they can't actually issue new shares without putting it to a shareholder vote? Should they issue bonds? No I know they should run an ICO!!!
And then what margins exactly? Do you know what the margin is on MI300? No. Do you know whether they're currently selling at a loss to win marketshare? No.
I would the happiest boy if hn, in addition to policing jokes and memes, could police arrogance.
They didn't imply it was notably easy. Your reply there only makes sense if you were trying to say it's nearly impossible. If you're just saying it's kinda hard then your post is weirdly hostile for no reason, reading theirs in an extreme way just so you can correct it harder.
There is always financing, there are always people willing to go to the competitor at some wage, there is always a way if the leadership wants to.
If it was just a straight up fab bottleneck? Yeah maybe you buy that for a year or two.
“During Q1, Nvidia reported $5.6 billion in cost of goods sold (COGS). This resulted in a gross profit of $20.4 billion, or a margin profile of 78.4%.”
That’s called an “induced market failure”.