As time goes on I don’t see how you break the CUDA moat even if you had all of nvidia Al’s engineers.
CUDA means you need everyone in AI to target your new (hopefully open) platform and that platform is faster than CUDA is. Given how most frameworks of the last 10 years have been optimized for CUDA you would need to turn around a global sized cruise ship.
If Intel’s GPUs are only 3% faster, will that be enough to rewrite my entire software stack for something not CUDA? If intel opts for a translation layer, could they ever match nvidia’s performance?
If Intel is 10% slower but 50% cheaper and the open source stack you use has been heavily updated to work well with Intel drivers would that not be an enticing product?
Yes, and in fact both AMD and Intel have libraries. You can run Stable Diffusion and suchlike on AMD GPUs today, apparently. And you can export models from most ML frameworks to run in the browser, on phones and suchlike.
> If Intel is 10% slower but 50% cheaper [...] would that not be an enticing product?
Sometimes, yes. Some of the largest models apparently cost $600,000 in compute time to train [1], so halving that would be pretty appealing.
However, part of the reason for nvidia's dominance is that if you're hiring an ML engineer for $160,000/year spending $1,600 to give them an RTX 4090 is chump change.
[1] https://twitter.com/emostaque/status/1563870674111832066
Lol... Intel is famously stingy when it comes to salaries.
Principals usually broke into the three digit range from what I hear.
Probably Intel's best bet now would be to try to be the fab for NVDA.
No.
For different people, the timespan between now and when they may want or need to sell it is different, and thus different people will arrive at different conclusions.
And note that “worth more” above simply means growth in market capitalization, so as long as someone is willing to buy the shares at a price supporting that increased market cap, then it does not matter if Tesla is still selling fewer cars than the others combined.
And as somebody with a significant short position, I would add - "The market can stay irrational longer than you can stay solvent" !
Their numbers for this and next Q are absolutely amazing. It's also quite "refreshing" - a company with great product, almost without competition (so far - it will come real quick). And fun part being their main advantage is probably CUDA and not even the chips itself (which by the way they don't manufacture - they "only" do the design).
But still - even with those numbers, and even with this pace of growth (both being absolutely not sustainable, and will probably reverse hard next year) - the valuation doesn't make any sense, especially given the current interest rates.
Raptor Lake is at 7nm and incredibly competitive there (~2700 single core on geekbench, taken with a pinch of salt). They’re still planning on being on 1.8nm/18A within 2 years, while at the same time ramping up their GPU efforts (albeit using TSMC for 4nm). Nvidia is very much in the lead, but this is just the beginning.
tldr; I ain’t hear no bell.
1. They don't pay - Nvidia/Google/Apple easily pays 1.5-2x Intel before appreciattion.
2. They're cheap/beaurcratic. The office sucks, your laptop sucks.
3. They suck at software. https://pharr.org/matt/blog/2018/04/18/ispc-origins
4. They can't develop/retain talent. Half the ML-HW/FW teams at AMD/Google/Nvidia/Apple are ex-Intel.
(Retention notwithstanding :P)
So, if intel 7 is a 10nm class node then so is tsmc N7. But the whole industry cooked their node names a decade ago and intel is allowed to do the same terminology. To do anything otherwise would be a needless marketing disadvantage. If someone’s cheating on marketing, everyone has to - and the reality was everyone but intel was already cheating.
i.e. This would mean Intel will be getting a four-node-scale jump in performance (10nm->7nm->5nm->3nm->1.8nm) from where they are with Raptor Lake by 2025 if they can stay on track.
I don't have any insight into this but I would hope Pat Geslinger is righting the salary, perks and incentive structure as a matter of priority to stop the shedding of talent.
Absolutely not and this is one of the strongest headwinds intel faces. They just did a big round of layoffs and then everyone who didn’t get cut got a surprise 20% pay cut after, and retention bonuses for seniors was typically low-3-digits to high-2-digits. Principle engineers were solidly 3 digits though. And then they did stock buybacks that exceeded the size of the layoffs and pay cuts.
Intel is fucking broke and they’re gonna have to cut every cost they can to focus on their key markets, and do it with personnel that were already underpaid and have better opportunities elsewhere. Not that labor is the straw that’s breaking intel’s back here but the situation is really AMD-in-2012 level dire, they have an incredibly long and expensive road to return to profitability and they will have to do it with whatever they can scrape up out of a staff that was already abused even in the “good times”.
Oh and they want you to move from desirable west-coast locations (well, except Arizona) to Ohio. And they’re massively behind on their tech stack (fighting uphill is not what everyone wants) in a failing org with a ton of middle-management rot and neglect. And they’re still failing to execute a huge amount of the time, causing massive delays, etc.
It’s really really dire and just because intel has staunched the worst of the financial bleeding doesn’t mean there’s much light at the end of the tunnel. Like Boeing I don’t think they will be allowed to go under, they’re strategically too important, but they’re gonna have a bad time for a while and probably legitimately do need a bunch of subsidy and “pity contracts” for national labs HPC computers to stay on their feet - just like AMD did.
(oh and to be clear by "2 to 3 digits" I don't mean 10-999% bonus... I mean they gave senior engineers a $75-125 retention bonus and principals might get 2 or 3 hundred. Those are the people who will build Intel's next foundational architectures and the nodes and the packaging technologies that lead themselves and their foundry customers to success. Have fun with that.)
I worked at Sears corporate when Amazon was getting big, about 25 years ago.
Always made me chuckle when armchair quarterbacks on TV would wonder why Sears couldn't do what Amazon did.
Bezos took tremendous risks in the late 90s and early 00s, while Sears was trying to figure out how to wring a few more pennies out of their stores. Sears Corporate was 110% focused on taking the existing business and maximizing profits, not on innovation of any kind whatsoever.