←back to thread

Ask HN: Why hasn’t AMD made a viable CUDA alternative?

193 points spacebanana7 | 7 comments | 4/1/2025, 2:37:00 PM | HN request time: 1.76s | source | bottom

I appreciate developing ROCm into something competitive with CUDA would require a lot of work, both internally within AMD and with external contributions to the relevant open source libraries.

However the amount of resources at stake is incredible. The delta between NVIDIA's value and AMD's is bigger than the annual GDP of Spain. Even if they needed to hire a few thousand engineers at a few million in comp each, it'd still be a good investment.

Show context
johnnyjeans ◴[4/1/2025, 2:53:00 PM] No.43547510[source]▶

> The delta between NVIDIA's value and AMD's is bigger than the annual GDP of Spain.

Nvidia is massively overvalued right now. AI has rocketed them into absolute absurdity, and it's not sustainable. Put aside the actual technology for a second and realize that public image of AI is at rock bottom. Every single time a company puts out AI-generated materials, they receive immense public backlash. That's not going away any time soon and it's only likely to get worse.

Speaking as someone that's not even remotely anti-AI, I wouldn't touch the shit with a 10 foot pole because of how bad the public image is. The moment that capital realizes this, that bubble is going to pop and it's going to pop hard.

replies(5): >>43547559 #>>43547590 #>>43547630 #>>43547711 #>>43548927 #
1. spacebanana7 ◴[4/1/2025, 3:08:00 PM] No.43547711[source]▶

> AI has rocketed them into absolute absurdity, and it's not sustainable

Why isn't it sustainable? Their biggest customers all have strong finances and legitimate demand. Google and Facebook would happily run every piece of user generated content through an LLM if they had enough GPUs. Same with Microsoft and every enterprise document.

The VC backed companies and Open AI are more fragile, but they're comparatively small customers.

replies(3): >>43547801 #>>43548050 #>>43548097 #
2. pjc50 ◴[4/1/2025, 3:15:00 PM] No.43547801[source]▶

> Google and Facebook would happily run every piece of user generated content through an LLM if they had enough GPUs. Same with Microsoft and every enterprise document.

.. But how much actual value derives from this?

replies(1): >>43547917 #
3. spacebanana7 ◴[4/1/2025, 3:24:00 PM] No.43547917[source]▶

Youtube could conceivably put multi language subtitles on every video. Potentially even dub them.

But the "real value" would come from making adverts better targeted and more interactive. It's hard to quantity as a person outside of the companies, but the intuition for a positive value is pretty strong.

replies(1): >>43548093 #
4. ndiddy ◴[4/1/2025, 3:35:00 PM] No.43548050[source]▶

IMO the closest analogue for Nvidia now is Cisco during the dot-com boom. Cisco sold the physical infrastructure required for Internet companies to operate. Investors all bought in because they figured it was a safe bet. Individual companies may come and go, but if the Internet keeps growing, companies will always need to buy networking equipment. Despite the Internet being way bigger than it was in 2000, and Cisco being highly profitable, Cisco's share price has never exceeded the peak it was at during the dot-com boom.

5. johnnyjeans ◴[4/1/2025, 3:38:00 PM] No.43548093{3}[source]▶

> Youtube could conceivably put multi language subtitles on every video.

They already do this, it's opt-in.

> But the "real value" would come from making adverts better targeted and more interactive.

Is there any evidence to suggest that a transformer would be better at collaborative filtering than the current deep learning system that was custom engineered and built for this?

6. fancyfredbot ◴[4/1/2025, 3:38:00 PM] No.43548097[source]▶

Google may well want to run more of their content through an LLM, but they will not be using Nvidia hardware to do it, they'll be using their TPUs.

Amazon are on their third generation of in-house AI chips and Anthropic will be using those chips to train the next generation of Claude.

In other words, their biggest customers are looking for cheaper alternatives and are already succeeding in finding them.

replies(1): >>43549546 #
7. a5ehren ◴[4/1/2025, 5:41:00 PM] No.43549546[source]▶

Google and Amazon still have to buy tons of Nvidia HW to provide in their clouds. No one writes to their custom chips besides internal teams because the software stack doesn't exist.