AMD's AI Assault: Unveiling the Challenger to Nvidia's Market Reign

0:00Today on the podcast we're going to be covering some big news in the AI chipmaking space that actually is probably going to start cutting down the costs of training and running different AI models. So the news is that AMD has announced a new AI chip that is going to challenge NVIDIA's current chip. And the thing that I think is so important and interesting with this is just the fact that they're new, it's called the MI300X, but it's going to start shipping to some customers later this year. So I think that's important because we have seen a lot of other companies announcing chips meta's making a chip but meta says their chip won't be ready till 2025 so i think um obviously when you have a company like amd that already is making a lot of this silicone a lot of this tech uh these gpus and all of a sudden they're able to use their current factories and their current manufacturing to get something out they obviously have a leg up on other people that are going to have to start from scratch on that and that being said i do think that in the next couple years we're going to be seeing some major we'll see some major cost cutting in this area as we see people like meta and a lot of other companies coming to market with chips now that they see how lucrative this is in the meantime nvidia is going to be able to make a lot of money i mean we're still you know we're still forecasting till for this amd chip the mi300x to not be coming out till the end of the year so nvidia i would say has a pretty good run till then but I think we're going to start seeing costs cut down quite a bit because currently NVIDIA actually dominates the market for AI chips and they have over 80 % market share according to some analysts so they've obviously cornered that to a pretty high degree and as you know GPUs are really important so this new GPU chip is going to be a big deal for AMD and GPU chips are currently used by you know like open ai to run their chat gpt models um if amd's ai chips which they currently call accelerators are actually embraced by developers um and servers they possibly could substitute nvidia products with this and i think this could be a really big untapped market for uh amd because they are actually just traditionally known for their computer processors so the ceo of amd which by the way this is something i just learned recently but did you know that the CEO of AMD and the CEO of Nvidia, they're like second cousins?

2:29This is what they got this like, they got this like chip dynasty going on. That's just a fun rabbit hole to go down. But in any case, the CEO of AMD recently told investors and analysts in San Francisco on Tuesday that AI is the company's quote, largest and most strategic long term growth opportunity. And I think she's spot on with this. I think that this is where we're going to be seeing the most growth for a lot of these companies. And if you look at it in a in an economic perspective, we're actually seeing a decline at the moment with a lot of these computer companies selling just traditional computer hardware.

3:05And so I think that AI is kind of a, you know, is a bright spot on the horizon where we're going to see a lot of growth. This is not an area I believe is going to slow down anytime soon. And so I think investing resources and energy into this area is probably the most beneficial things some of these companies can do. So Lisa Su also said, we think about the data center AI accelerator market growing from something like 30 billion this year at over 50 % compound annual growth rate to over 150 billion in 2027. So this is obviously massive. If we're looking at a market that they're forecasting to grow to 50 % compounded annual growth rate for the next number of years, this is a really big opportunity.

3:48So while AMD, they didn't actually say what the price of this new chip is going to be. I think this is going to put a little bit of pressure on Nvidia for their GPUs because currently their H100, that can cost$30 ,000 or more. So I think this is going to drive down the prices and I think this is a good thing. I think lower GPU prices are going to start cutting down the cost of, you know, just doing all of this generative AI technology. And I think it's really cool, you know, we've seen out of OpenAI as they've released new AI models like GPT-4, they've cut down the cost of their older models. And I think that, you know, if they're starting to get chips that are a lot cheaper, we might see those costs go down significantly, which could be quite incredible.

4:36I think also last month, the CEO, Lisa Su, she said on their earnings call that while the MI300X is going to be available for sampling this fall it would start shipping in greater volumes next year so to be fair they're not about to you know just dominate nvidia at this exact moment amd did say that its new mi300x chip and its cdna architecture were designed for larger language models and other cutting edge ai models they said at the center of this are gpus gpus are enabling generative ai so what's interesting is this new GPU can actually use up to 192 gigabytes of memory, which means that it can fit a lot of the larger AI models and a lot of bigger AI models than other chips.

5:26So NVIDIA is essentially the rival here at NVIDIA is the H100, which only supports 120 gigabytes of memory. So 192 gigs is a significant boost over that. A lot of LLMs for generative AI use quite a bit of memory because they run an increasing number of calculations so amd when they did their their demo of this new gpu they did it running a 40 billion parameter model called falcon and what's interesting to note is just by comparison openai's gpt3 model has 175 billion parameters so they ran one ai model on one gpu and inevitably you know you're gonna have to stack multiple of these together but as Sue said model sizes are getting much larger and you actually need multiple GPUs to run the latest LLMs but with the extra memory that they have on these GPUs developers aren't going to need to use as many in order to run stuff so I think that it is another interesting play AMD said that they're going to be offering what they're calling an infinity architecture so essentially it combines eight of its M1300X accelerators in one system and this is similar to what NVIDIA and Google have both done.

6:46They have different packages that essentially they combine eight or more GPUs into a single box for AI applications and so I think the reason that the reason that AMD might not have a breakaway thing just quite yet is the fact that developers really do prefer nvidia at the moment they they prefer these chips um because one of the strongest things in addition to just the fact they have good chips is the fact that they have a really well-developed software package called cuda and essentially that helps them to access chips core hardware features so amd on tuesday did say that they wanted to come out with their or they have their own software for these AI chips they call it Rock M so they said now while this is a journey we've made really great progress in building a powerful software stack that works with the open ecosystem of models libraries frameworks and tools that was Victor Peng who is their AMD president I think this is gonna be a really interesting space to watch this is gonna have some pretty massive implications for the AI industry as a whole but if AMD can you know managed to pull out an impressive GPU that is getting some mass adoption NVIDIA might be getting a little bit of a run for their money although we should note that that isn't going to be slated till really ramping up till next year.