Instead of saying to AMD they will be in the rearview mirror, they should have been paranoid. Not do stupid GPUs. and destroy others where it mattered
Hard disagree here x100. Investing in GPUs in the time when Nvidia and AMD started to gouge the market is actually the best decision Intel did in recent times. It's the piece of semiconductor with some of the highest margins in the business and they already own a lot of the patents and IP building blocks to make it happen.
The only stupid thing they did was not getting into GPUs earlier so they would already be on the market during the pandemic GPU shortage and AI boom.
Undercut the big boys with affordable on-prem AI.
Intel should have appointed people form the HN comment section as their CEO, as they clearly know more about running a giant chip design and fabrication company than the guy who worked there for 10+ years.
It is something that is easy to miss if you are just looking at typical business strategy and finances. A high memory consumer GPU would undercut their server GPUs, which are presumably higher margin intended golden geese. It's easy to see them chasing server markets and "gamers" being an afterthought.
However there is huge demand right now for a modern, even a crappy modern, GPU with gobs of memory. Make the card and the open source AI tooling for it will start sprouting in days after it's release.
It's an extremely powerful position to have every at-home AI geek's setup to be bound to using intel cards and intel focused tooling. Nvidia and AMD won't do it because they want to protect their server cards.
So, incredibly small market share while your competitors already have the first-mover advantage and nailed down the ecosystem? With no data backing it up, I think, graphics cards for local LLM needs is not really on demand. Even for gaming it’s probably more attractive, but then again, that’s not even where the real money is.
Exactly. This x100. It was easy for Nvidia to succed in the LLM market by winging it, in the days when there was no LLM market, so they had the greenfield and first mover advantages.
But today, when Nvidia dominates the mature LLM market, Intel winging it the same way Nvidia did, won't provide nearly the same success as Nvidia had.
Ferruccio Lamborghini also built a successful sports car company by building tractors and cars in his garage. Today you won't be able to create a Lamborghini competitor with something you can build in your garage. The market has changed unrecognizably in the mean time.
The people learning how to do local LLMs will be the people directing build out of on-prem transformers for small-midsize companies. The size of the market is irrelevant here, it's who is in that market and the power they will have that is extremely relevant.
AMD has tried this for many of its technologies and I don't think it is working. Granted, they suck at open sourcing, but a shitload of it was open sourced. See TinyGrad voyage into the Red Box driver (streams on youtube).
It's either old slow Tesla cards with 48GB or $2000 nvidia cards with 24GB.
I think you're overestimating what people can and will do.
Nvidia didn't succeed because it just launchend cards and let people write CUDA for them. Nvidia is where it is because it has an army of researchers and SW engineers developing the full stack from research papers, to frameworks, to proofs of concepts, showing customers the value of paying for their pricey HW + SW, most of it proprietary, not community developed.
"People" alone won't be able to get even 10% there. And that's ignoring the fact that Nvidia HW is not FOSS so they'd be working blind.
The current local model/open source model community is literally an army of SWE's and researchers. They already make tons of tooling too.
We can't see the future, but neither can CEOs, no matter how well paid and respected they are.
After all the current CEO is being ousted, so obviously he didn't do the right things despite being a CEO.
I have my thoughts on the matter and cautiously welcomed their move to GPUs ( though admittedly on the basis that we -- consumers -- need more than amd/nvidia duopoly in that space; so I am not exactly unbiased ).
That's just speculation. There's no guarantee that would have happened. Nobody has a crystal ball to guarantee that as the outcome.
It's like saying if someone would have killed Hitler as a baby, that would have prevented WW2.
What do you think has happened so far?
Your mental model of the world may help me understanding the point you are trying to make.
That's just speculations from people online. I don't see any wisdom in that like you do, all I see is just a guessing game from people who think they know an industry when they don't (armchair experts to put it politely).
What made Nvidia dominant was not weak GPUs with a lot of RAM. The puzzle of their success had way more pieces that made the whole package appealing over many years, and a great timing of the market also helped. Intel making underperforming GPUs with a lot of RAM would not guarantee the same outcome at a later time in the market with an already entrenched Nvidia and a completely different landscape.
But since they obviously apply just as well to Intel itself, it is a poor reason to dismiss other’s ideas.
—
> What made Nvidia dominant was not weak GPUs with a lot of RAM.
Intel doesn’t have the luxury of repeating NVidia’s path in GPUs. NVidia didn’t have to compete with an already existing NVidia-like incumbent.
That requires no speculation.
—
Competing with an incumbent via an underserved low end, then moving up market, is called disruption.
It is a very effective strategy since (1) underserved markets may be small but are are immediately profitable, and (2) subsequent upward growth is very hard for the incumbent to defend against. The incumbent would have to lower their margins, and hammer their own market value.
And it would fit with Intel’s need to grow their foundry business from the low end up too.
They should take every low-end underserved market they can find. Those are good cards to play for ambitious startups and comebacks.
And the insane demand for both GPUs and chip making is increasing the number of such markets.
True, it is just speculation. 'Any' seems to be a strong qualifier. One of the reasons I troll landscape of HN is that some of the thoughts and recommendations expressed here ended up being useful in my life. One still has to apply reason and common sense, but I would not dream of saying it has no ( any ) wisdom.
<< What made Nvidia dominant was not weak GPUs with a lot of RAM.
I assume you mean: 'not in isolation'. If so, that statement is true. AMD cards at the very least had parity with nvidia, so it clearly wasn't just a question of ram.
<< The puzzle of their success had way more pieces that made the whole package appealing over many years, and a great timing of the market also helped.
I will be honest. I am biased against nvidia so take the next paragraph for the hate speech that it is.
Nvidia got lucky. CUDA was a big bet that paid off first on crypto and now on ai. Now, we can argue how much of that bet was luck meets preparation, because the bet itself was admittedly a well educated guess.
To your point, without those two waves, nvidia would still likely be battling amd in incremental improvements so the great market timing accounts for majority of its success. I will go as far as to say that we would likely not see a rush to buy 'a100s' and 'AI accellerators' with exception of very niche applications.
<< Intel making underperforming GPUs with a lot of RAM would not guarantee the same outcome at a later time in the market with an already entrenched Nvidia and a completely different landscape.
Underperforming may be the key word here and it is a very broad brush. In what sense are they underperforming and which segment are they intended for? As for ram, it would be kinda silly in current environment to put a new card out with 8gb; I think we can agree on that at least.
<< I'm saying nobody can guarantee the claim of the GP I've replied to,
True, but it is true for just about every aspect of life so as statements go, so it is virtually meaningless as an argument. Best one can do is argue possibilities based on what we do know about the world and the models it tends to follow.
The market for that is just not that large, it wouldn't move the needle on Intels revenue, but then again it could get the enthusiasts onboard and get Intels CUDA alternative moving.
But then you move up, and the incumbents have to choose to keep ceding more of their lower end or lower their margins. And it is very hard and costly for a company succeeding at the high end to do the later.
That would have been a fantastic sign Intel was getting more nimble and strategic.
And been a good fit with a come back in the low end of fabs too.
Some are probably multi millionaires smurfing (and I dont mean cryptobros).
Do you even have a Putnam award?
There are a lot of strategic shifts Intel could do but their timeline to paying off at the scale Intel needs is very long.
What I see is a bunch of endeavours that get killed off too quickly because they were not paying off fast enough and this creates mistrust in the community around Intel’s new initiatives that are not core that make them harder to succeed going forward. It is a bit of a death spiral.
Basically big companies find it hard to learn new tricks when their core offering starts to fail. The time to learn new tricks was a while ago, now it is too late.
If they had focused more on mobile CPUs, GPUs, or especially GPGPUs a decade ago, they could have more product diversity now to hold them over.
Instead, they dipped their toes into a new market every few years and then ran away screaming when they realized how difficult and it would be to gain market share.
If they had any actual vision, they could have a line of ARM CPUs now to compete with the onslaught in all major CPU markets.
They should have listened to their customers and market forces instead of trying to force x86 down everyone’s throats.
Disagree. Selling ARM chips with high profit margins is tough. There's too much competition from the likes of Samsung, MediaTek and until the US ban, Hi Silicon(Huawei). ARM chips are a race to the bottom in terms of price with a market dominated by companies from Asia. There's no guarantee Intel could have had a competitive ARM design that could beat Apple's or Qualcom's.
Yes, without proprietary technology, margins are slim.
But Qualcomm has somewhat succeeded in this area anyhow. I guess they took the ARM base but innovated on top of it in order to command higher margins.
It's wasn't just anyhow. Qualcomm succeed in the mobile SoC space because they also had the best modems in the industry (name comes form Quality Communications after all). And also the best mobile GPU IP they bought from ATI.
Intel and AMD can dominate the x86 market all they want. But x86 has been steadily losing ground every year to cheaper and more power efficient ARM processors. It’s still a big market now, but I don’t think it’s going to be a great business in a decade or two.
ARM was just an example. If Intel spent a decade strengthening their Larrabee GPGPU platform and building AI and crypto ecosystems on it, they may have been well positioned to benefit immensely like Nvidia has over the last 5 years.
Although their latest plan is to turn it into ARMv8 anyway: https://www.intel.com/content/www/us/en/developer/articles/t...
The main problem with x86 is that it has poor security, not poor performance.
With this logic, Apple should have also stayed with making Macs when it had financial troubles in in 1999, since that's its focus, not venture into making stupid stuff like Mp3 players and phones, everyone knows that's the core focus of Sony and Nokia who will rules those markets forever.
You’re off by a few years. Jobs axed any project not related to the Mac when he came back in 1997, so they actually did what you say they did not. The iPod project started around 1999-2000 when Apple was in a massive growth phase after the iMac and the G3 and then G4 desktops.
Also, in the alternate reality where the iPod did not massively succeed, it would very likely have been killed or remain as a hobby like the Apple TV. Apple might not be as huge as they are now, but the Mac was doing great at the time.
The CEO of company X may think that he's as talented as Steve Jobs, but is he really?
Trying to use that the fact Einstein or Muhammad Ali or any other genius in his area could do something or did something is not a counterpoint.
Intel has been "getting into" GPUs for two decades now going back to Larrabee. They are just not good at it.
Engineering chops?
AMD and nVidia already patented too much of the good stuff?
Too much existing code optimized for AMD and nVidia quirks?
A few weeks ago Gelsinger even commented he saw "less need for discrete graphics in the market going forward" - just seemed like a very Intel thing to say
GPUs exist because CPUs just aren't fast enough. Whether or not people are going to start making GPU-only computers is debatable (although there has clearly been a lot of CPU+GPU single chips).
A GPU-only computer would be absolutely horrendous to use. It'd be incredibly slow and unresponsive as GPUs just absolutely suck at running single-threaded code or branchy code.
The overall point is that work is being increasingly done _not_ on the CPU. If your business is CPUs-only then you're going to have rough times as the world moves away from a CPU-centric world. You don't need to predict AI; you just need to see that alternatives are being looked at by competitors and you'll lose an edge if you don't also look at them.
It's not going to matter much if you have a crappy CPU if most of the work is done on the GPU. Its like how iPhones don't advertise themselves as surround sound; phones aren't about calling people anymore so no reason to advertise legacy features.
Eh? GPGPU has been a thing for decades and yet barely made a dent in the demand for CPUs. Heck, CUDA is 17 years old!
The world has not budged from being CPU-centric and it isn't showing any signs of doing so. GPUs remain an accelerator for specialized workloads and are going to continue to be just that.
Of course they also add GPU-like things to the CPU too for the same reason: https://developer.arm.com/documentation/109246/0100/SME-Over...
It's a very useful distinction to make.
EDIT: I personally always liked intel iGPUs because they were always zero bullshit on Linux minus some screen tearing issues and mumbo-jumbo fixes required in X11.
The "stupid" thing with them (maybe) is that they cannot do anything exceptionally good, and that while having compatibility problems. They are cheap yes, but there are many other chips for the same price, and while they are less capable, they are more compatible.
Make A380 cost 30% less or invest way more to the drivers and IMHO it'd been completely different.
Driver stability, less heat and fan noise, battery life is almost assured in the Intel iGPU.
The only area where AMD's discreet GPUs are lagging behind is AI stuff. You get a lot more performance with Nvidia GPUs for the same price. For gaming, though AMD is the clear winner in the price/performance space.
Of course, Nvidia is still a bit of a pain but it's of their own making. You still need to install their proprietary driver (which IMHO isn't that big a deal) but the real issue is that if you upgrade your driver from say, 550 to 555 you have to rebuild all your CUDA stuff. In theory you shouldn't have to do that but in reality I had to blow away my venv and reinstall everything in order to get torch working again.
I am trying to learn this but having difficulty finding good explanations. I know the Wikipedia-level overview, but need more details.
To me, AMD's approach demonstrates planning and buy-in across multiple teams (CPU die, Ryzen IO die, Epyc IO die, etc), and that suggests a high degree of management involvement in these engineering decisions.
Intel's activities seem comparatively chaotic. Which I agree smells of disincentives and disinterested middle management hyperfixated on KPIs.
Their purported VRAM offerings for Battlemage are also lower than hoped for, which is going to be a turnoff to many buying for local inference.
I think we have too many decision-makers gunshy from crypto mining that haven't yet realized that compute isn't just a phase.
Apparently AMD has at least the Sony PS6 contract now.
"Gelsinger said the market will have less demand for dedicated graphics cards in the future."
(from: https://www.techspot.com/news/105407-intel-not-repeat-lunar-...)
Gelsinger apparently agreed with you. However, the market very clearly has enormous demand for discreet GPUs. Specifically for AI workloads (not PC gaming).
If I was on Intel's board I would've fired him for this statement. The demand for GPUs (parallel matrix processors with super fast local memory) is NOT going to go down. Certainly not in the next five to ten years!
I know Intel makes a lot of things, has a lot of IP, and is involved in many different markets but the fact that they're NOT a major player in the GPU/AI space is their biggest failure (in recent times). It's one of those things that should've been painfully obvious at some point in 2022 and here we have Gelsinger saying just a few months ago that somehow demand for AI stuff is just going to disappear (somehow).
It's magic hand waving like this that got him fired.
…Although it looks like Intel tried to go that route without understanding why and got cold feet.
Their dGPU is also pretty promising, I have it on my list to get - even if not for gaming, its possibly the best media encoding / decoding card for the money to get today. The only thing holding it back for entry level or mid level gaming is the drivers - for some games, it wont matter it seems, but for others it has some growing pains but they seem to be diligently working on them with every driver release.
Intel has made vast improvements even within their first generation of dedicated desktop cards. They will likely never compete with cards like a 4080/4090, but they may be great options for people on a budget. Helps put pressure on AMD to be better as well.
No they aren't - much like Boeing, at this point they are considered a national security asset.