Hacker News new | ask | show | jobs
by jauntywundrkind 746 days ago
Such a weird & cruel modernity, where these releases are purely in the abstract. No, you still won't be able to buy a MI300X in Q4 2024. The enhanced edition will absolutely not be available.

(I miss the old PC era where the world at large was benefiting in tandem from new things happening (or falling behind from not adapting)).

5 comments

The focus is on hyperscalers and cloud service providers now. Even Groq stopped selling to retail.
Welcome (back) to the age of the mainframe.

Except we call it "cloudframe" now.

I think that's where short-sighted financial gain leads AMD to. Where's the money? -- datacenter. So let's focus the good stuff on datacenter exclusivelly. What about "the rest" (gamers, hobbist, students)? There's no money there, let's give theme crap RDNA that we make sure can't be used for any real work; just pretent we're catering for their needs.

I think their "consumer GPU" did so bad recently that AMD could just as well, you know, simply liquidate the "consumer GPU" division and stop pretending.

I'm in the "consumer GPU" market myself; what AMD GPU do I buy today? -- Radeon Pro VII, launched in 2020 and the best AMD consumer GPU I can find today.

It's such a divide. I could optimize my software for such powerful GPUs as the Mi300 line.. but why do that, given that probably I won't even see one such GPU in my lifetime.

The RX 7900s are pretty good. You get 24GB of RAM in a consumer GPU. If you're interested in GenAI that's a good offering for your "gamers, students, hobbyists" category.

And they announced a workstation version with 48GB: https://www.phoronix.com/news/AMD-Radeon-PRO-W7900-Dual-Slot

Indeed for me that's the only standout from this release, 48gb not too overpriced.
I’ve filed a detailed AMD Windows driver crash bug months ago that is getting totally ignored because all their devs have been moved to working on AI.
> (I miss the old PC era where the world at large was benefiting in tandem from new things happening (or falling behind from not adapting))

Paper launches aren't anything new. It's always been a thing especially in hardware.

What is kind of new is that hyperscalers sometimes get chips 1-2 years before the rest. If the new chip is 3x-4x faster that looks pretty unfair.
> What is kind of new is that hyperscalers sometimes get chips 1-2 years before the rest.

That's not new. Replace hyperscalers with crypto miners and we've had a similar situation for a while.

Crypto miners no longer buy up large quantities of GPUs. That market died off when ETH switched from PoW to PoS.
> Crypto miners no longer buy up large quantities of GPUs.

That's news to crypto miners.

I no longer have the 150,000 GPUs I was mining with.

The only ones left standing are the small ones and most of them are just using what they already purchased.

There is no liquidity in the shitcoins to be able to support daily large dumps of their tokens and unless you have almost free power, there is zero profitability.

> No, you still won't be able to buy a MI300X in Q4 2024

Why not? Because they’re sold out to hyperscalers?

They are not sold out. It is just a lot more work to support retail on a novel new product, so they are focused on hyperscalers and CSP's. Don't forget that high end GPUs are US export controlled as well. They are considered weapons by the government. [See 88 Fed. Reg. 73458 (Oct. 25, 2023) and the Export administration Regulations (EAR)].
> No, you still won't be able to buy a MI300X in Q4 2024.

they're 15k - who exactly is disappointed they won't be able to buy one?

People on reddit.com/r/localllama build even more expensive rigs sometimes. Everyone wants to run llama3 70B and eventually the 405B version.
As someone who's buying them, that is not the price. Nor can you just buy one at a time. They are on a OAM/UBB, which is 8.
I’d buy a few at that price. But I can’t.
I mean it's the price of 2 * Apple Mac Pro's. It isn't exactly unheard of.

Just a few weeks ago I spoke to someone who shelled out $10k for running LLM's locally. I've seen more expensive builds as well.

> who exactly is disappointed they won't be able to buy one?

HPC centers and research clusters.