[FredPret]

AMD needs to launch drivers, not chips. Who’s going to develop ML models on a card that can’t connect to a popular ML model?

I look at their financial performance and it’s staggering how they’ve missed the boat - and this is during a huge boom on gaming, crypto, and AI.

Compare:

https://valustox.com/AMD

VS

https://valustox.com/NVDA

[qaq]

That is the confusing part say you need to hire 100 people at 1 mil. per year comp to get drivers to a good state. Thats 1/3 of their quarterly profit but would prob double the revenue in a few years.

[rafaelmn]

From what I've heard they aren't offering competitive salaries, so introducing people at insane comps would probably destroy existing teams(if there was no salary bumps to match)/budgets (if there was). Doubt you can do much with new hires in 1 year in such environment, by the time you start seeing results it's probably too late to capitalize on this bubble.

More likely, they wait to see how the AI HW startups shake up and then acquire the ones that have anything worth paying for.

[logicchains]

>From what I've heard they aren't offering competitive salaries

Seems like a common thing in hardware companies, they chronically underpay, which for some reason hardware/electrical engineers seem to accept, but that makes them a last-choice for competent software engineers, who have much better-paying options.

[lesuorac]

> which for some reason

Didn't you just explicitly say the reason? A SWE can go off and make Google in their Gargage; a EE can't make a fab in their garage.

[fragmede]

Even if they could make a fab, it would still be a logistical nightmare to scale from 1-1,000 users. Meanwhile, my SaaS company could have 100,000 users thanks to the cloud, and I wouldn't even have to get up from my desk.

[djmips]

AMD doesn't even own a fab.

[lesuorac]

Non-sequitor [1]. AMD not having a fab doesn't mean a EE can run a fab out of their garage.

It may be relatively easier for people to make new chips without needing AMD/Intel (see all of fang company making their own). But it's still companies with lots of money making new chips and not people in their garage.

[1]: https://en.wikipedia.org/w/index.php?title=Non_sequitur_(fal...

[p_j_w]

They own licenses to the insanely expensive software you need to build a modern IC.

[xchip]

I heard the same too (speaking for a friend)

[wongarsu]

It's not just drivers though. Nvidia has invested close to two decades into documentation, teaching materials, developer tooling and libraries for CUDA, plus all the work on gaining mindshare.

You could probably get 80% there by dedicating enough AMD developers to improving AMD support in existing AI frameworks and software, in parallel with improving drivers and whatever CUDA equivalent they are betting on right now. But it would need a massive concerted effort that few companies seem to be able to pull off (probably it's hard to align the company on the right goals)

[hardware2win]

1mil comp? Haha, what?

Salaries at semico companies are not even close to this

Also why would you even need ppl this good? People who earn 1 mil offer way, way more than just tech skills

[logicchains]

NVidia is a semico and they pay much better than AMD: https://www.indeed.com/companies/compare/Amd-vs-Nvidia-b78a5... . That's a big part of why they're so far ahead now and their drivers are much better. If AMD wants to catch up to NVidia within a reasonable timeframe, they wouldn't just need to match it, they'd need to pay way better than NVidia to attract the best people.

[djmips]

Agreed - Nvidia has made software a core pillar since the nineties.

[visarga]

That's not such a big concern, LLMs run on all things today. It won't be that hard to make them work on AMD. Before 2020 we had much more architectural diversity.

[FredPret]

The last time I tried getting Tensorflow / Pytorch to work on my good (and cheap) AMD card, I literally ended up just buying an Nvidia card.

I’m just one guy but my experience carries over to subsequent business decisions made by me, and there are many like me.

[Workaccount2]

I'll throw my hat in the ring, I didn't buy an nvidia card, I simply gave up on trying to get it working.

[FredPret]

Nvidia drivers on Linux aren’t a walk in the park either, but at least there’s a way to do it.

[dharma1]

they've had a long time to get ML working as well as with CUDA and CuDNN, mostly failed attempts so far. How close are they?

[bigbillheck]

> AMD needs to....

https://www.amd.com/en/newsroom/press-releases/2023-10-10-am...

[Workaccount2]

Whats insane is that AMD has been known for it's shit drivers for over a decade now...and nothing has happened to address this. Like surely everyone internally knows it, all the execs know it, the board knows it, investors know it...but somehow it has never been addressed.

At this point it's almost like it has to be intentional, like some perceived tradeoff ingrained in the culture that generates shit software.

[logicchains]

>At this point it's almost like it has to be intentional, like some perceived tradeoff ingrained in the culture that generates shit software.

They're underpaying their hardware engineers, and if they wanted to hire good software engineers they'd need to pay more, which would cause their hardware engineers to demand better pay too.

[FredPret]

The price of underpaying their employees like this can be seen very clearly in their revenue graph.

[machinekob]

They are not only missed AI boom, they are also seems to be overpriced by a huge margin compared to Intel or even NVDA.