Nvidia has been producing their own ARM cores since at least 2014:

https://en.wikipedia.org/wiki/Project_Denver

The ARM core performance hasn't been spectacular vs the latest and greatest from Samsung and Apple but combined with CUDA and tensor cores it can provide pretty incredible results considering the cost, form factor, and power consumption. Carmel is also at least two years old at this point...

Their project Denver announcement from 2011 discusses plans for server and HPC applications:

https://blogs.nvidia.com/blog/2011/01/05/project-denver-proc...

Yes, its worth saying for context: The Nintendo Switch runs on an Nvidia ARM core.

They're capable of pretty great performance in the right contexts, and the industry definitely needs more consumer-accessible HPC-focused ARM competition what with Apple leading the charge on "everything to ARM". Microsoft has also shown interest, with the SPX, but we need processors more capable than what Qualcomm is producing; I'm hopeful Nvidia can help fill that gap in the next five years.

Very succinct summary - with the contrast (from an investor perspective) magnified by the fact that Nvidia has stable management that has delivered over a number of years, compared to Intel's untested CEO and recent history of significant mis-steps.

>> Nvidia has stable management

I'd expand this further to say Nvidia has an enthusiastic, highly-charismatic CEO who literally has groupies (e.g., when you see him at conferences.) For a big-time CEO, he is also surprisingly nice and approachable -- I've now interacted with him on three separate occasions. It makes me like him, his company, what he does.

I realize that is touchy-feely but I wonder how common this sentiment is.

I realise I did Nvidia a dis-service as Jen-Hsun and team are obviously much, much more than just 'stable'.

Never met him, but seems very impressive and approachable - how many CEOs would give a talk to students sitting on the steps of a university entrance? [1]

Just to add though that having groupies is not necessarily a positive for investors!

[1] https://www.youtube.com/watch?v=hqiBzmuqPNY

He definitely seems approachable. Someone I knew from school interned at nvidia and supposedly he hosted a party at his house for the interns (possibly just PhD interns) and let people rev one of his fancy cars

Oracle also threw us a party as new hires at Mark Hurd's "house" (party villa) in Atherton. HR and all of the top dogs were there except for Larry and the Board of course. It was a nice gesture but he had a lot of security combing the house at all times and we weren't allowed to bring our phones even.

Definitely a well intentioned gesture but as always in SV corporate world, just becomes weirder by the minute :)

was an intern at nvidia. He's very approachable and all interns are invited to his huge ass california house. I saw came out of a meeting and just said hi to him. He's pretty much an engineer just like everyone except he's a CEO and he doesn't have that douchebag vibe like most tech CEOs are.

As someone trying to start a company of my own, I really enjoyed that talk he gave on the steps, where he talked about how nvidia was born and grew up. Highly motivating. In fact I curate such entrepreneur talks.

Any playlist do you like to share?

Yeah every year he throws a party for the interns. Not a weird sketchy one either, but HR is involved, drinks are limited and that whole thing.

He gives me creepy vibes. But every CEO does that, so I’m not sure it’s a bad thing. At least I actually liked him.

Those are his personalities, but much more important I think is that Jensen is actually a product CEO. A Product person would care about your product beyond it is just selling well. Not something could be said about Intel CEOs.

Dr Lisa Su is a product CEO as well. She had a much tougher time with AMD's debt and lack of resources, but still manage to pull things off. And small Fun Fact. Jensen and Dr Lisa Su are actually family relatives.

I think they are both interesting in being both product focused and having a deep understanding of the technology - which is a fairly unusual and a very powerful combination.

Agreed on the management. Regarding Intel’s CEO, Bob Swan is a finance guy with no history or background in semiconductors. It remains to be seen whether someone with no real background in technology or in the semiconductor industry can turn things around at Intel.

Apparently his current strategy is to just cut costs like craz-, I mean, like every CEO who was once a CFO. I know driver developers who work there and are stuck with old broken laptops of previous generations because there's no budget (they call it BTI) to upgrade. No dogfoodin for them. Even the previously free fruits are now just one per day per employee.

Source: I know people who work there.

Meanwhile the cost of monitoring the free fruit situation will cost more than the couple of crates of apple savings

perks are very very hard to take away. THe lower morale, and in many cases cause the best employees to leave, not because of the actual perk, but because of what that means - That the company doesn't value its employees.

I agree, I've seen many companies cut perks and yet not save any money, they just reduce morale.

> but because of what that means - That the company doesn't value its employees.

reducing perks is also a danger sign that a company is in a death spiral. If the company is to the point where it's worrying about $200 a week in fruit across N thousands employees then that can be indicative of other internal problems

It's also very indicative at this time, during pandemic about those companies chopping regular employees pay versus not doing anything to top line C suite.

Many companies reduced pay by 10%-20% or stopped 401k match or even pay raise when C suite still made profits from recent stock moves.

People get used to how a company is.

If the lunchroom is like the orphanage in Oliver Twist, and you have 12 inch VGA monitors, well, that's just the way it is.

At another company, if one day you notice that the coffee in the break room is no longer organic, even if it's of good quality, and the string cheese is only restocked twice a day instead of three times, that's a bad sign. Somebody is trying to save money. That may not mean the company is in trouble but it's a sign something isn't as good as it once was.

> not because of the actual perk, but because of what that means - That the company doesn't value its employees.

and/or that money is tight and things are likely to get worse

Anyone know how the AMD workplace is? Learning about work life at different companies is really interesting to me

I've always personally envisioned AMD's graphics driver team as a bunch of underdogs pulling off serious wizardry on a shoestring budget and still being criticized by the public for having any bugs at all.

> I am still waiting for Nvidia to make their play in CPU Market. May be a move in ARM Server CPU space.

What makes you think they are better prepared for this than, say, Qualcomm... or even Intel?

Despite Apple's decision to switch to ARM, I'm still not convinced it's the right architecture for the servers of the future. It more likely indicates Apple's convergence to a single platform, away from the classic desktop.

What makes you think it's not the right architecture for servers?

If anything AWS' Graviton2s are competitive on multithreaded workloads and it looks like the Neoverse cores are only going to improve there.

Personally I think servers are hard to get into because you rely on scoring big deals against Intel/AMD and the shipment volume is rather small compared to other markets.

However, being able to provide decent ML to cars, cell towers, buildings etc at a reasonable price/power budget has the potential for much larger volume albeit at a smaller profit margin. NVIDIA is largely uncontested in ML acceleration so they have a unique selling point in an embedded market that's waiting for big disruption.

It's no secret that NVIDIA is involved with RISC V. And don't forget Intel v Ampere (the company, not the NVIDIA chip). Intel is facing incredible competition. Once a crack appears in the server market and Arm rushes in, Intel will be in a very, very bad situation. Especially with 100,000+ employees compared to NVIDIA's 13,000.

EDIT: Downvotes? Really?

I'm personally not terribly bullish on RISC V currently. All of the advantages have to be weighed against "it's not x86_64". ARM processors at least have tons of smartphone people working out the bugs and random weirdness. With RISC V you're the one on the bleeding edge, and will be for a long time.

Plus if you're looking at the medium-long horizon you have to consider the possibility that the chips will stagnate or die out entirely and you'll be forced to abandon them anyway.

Everybody wants to wait until there have been some people working with the technology at scale and working out he bugs, finding problems, fixing the docs, populating StackExchange, etc... In order to get people to make the leap you need to offer them something significantly better than the current offerings so they'll be willing to do the extra work. If you are offering marginal power efficiency improvements that is not going to cut it.

A typical GPU has a bunch of small microcontrollers on it, usually Cortex-M’s. Those need to be licensed. I wonder if Nvidia’s interest in RISC-V is just to replace those microcontrollers?

Indeed, RISC-V has mainly been as replacement for Falcon cores in GPUs so far.

They evaluated vs Cortex A-53 through rather than Cortex-M.

https://riscv.org/wp-content/uploads/2016/07/Tue1100_Nvidia_...

Their gain in risc-v may be more about making life hard for Intel (anti-competitive) and avoiding patents etc in gpu, than not actually wanting to go head to head on CPU revenue. Similar but weaker to why Google did Android (Apple phone monopoly might hurt selling ads).

If they produce competitive CPU on RISC-V that'd be very interesting. I'd like to see a competition between ARM and RISC-V chips, so that ARM is not the monopoly in that niche market (high-end non-x86 chips). I think it's very unlikely though, if not impossible. Seems more logical to produce ARM.

RISC-V chips have tended to run a bit cooler than ARM given comparable performance class, microarchitecture and fabrication device node. However it remains to be seen if this advantage will be preserved, especially as more RISC-V instruction extensions get standardized (vector instructions, packed SIMD, bit-level ops are all on schedule, plus a grab-bag extension for JIT hardware acceleration) and as ARM64 implementations improve.

Citation?

> EDIT: Downvotes? Really?

I was so close to actually upvoting you, and then I had to go and downvote... and by this point you were back in the black, so to speak.

It's baffling to me how smart people like yourself, knowing the HN rules or at least being able to read them, still complain about downvotes. I really don't get it.

Downvotes!? Really!?

> AMD was never really in the Notebook market segment in its history

AM386 SX was a fully static cpu with a much lower power consumption, way better for laptops that Intel offerings at the time.

Tracks with my experience; I remember having an amd 486 notebook, and it was about 50/50 back then.

Wow, Intel has dominated the market for something like 40 years now?

Thing is, the competition is even worse regarding the fruit to fix.

It would be interesting to put a few dozen ARM cores on a high end GPU card. Why not just run your complete application directly on the card?

Intel tried that with Intel Xeon Phi, basically.

That was just a lot of Intel cores. I'm talking about an NVIDIA GPU that happens to have some ARM cores along with the CUDA cores.

You're essentially talking about a Jetson:

https://elinux.org/Jetson_AGX_Xavier

Eight Nvidia ARMv8 cores, 512 Volta cores, 64 tensor cores, and 2 DLAs - all connected to 32 GB of shared RAM.

It was announced over two years ago and the price has fallen to $800. For another $200 you can add a USB-C to DP cable (for a second display) and 1TB of NVMe storage.

If I had a $1,000 compute budget it's what I'd get. You'd have a more than functional dual 4K daily driver for dev work, browsing, etc plus the ability to run ML workloads. Oh yeah, and it tops out at less than 50 watts from the wall with all but the heaviest tasks coming in at 25-50% of that.

I love the Jetson -- we have a bunch of them here -- but that's a SBC. What I want is an NVIDIA 2080Ti or 3080Ti card with some ARM cores that I can run apps that use CUDA+ARM directly on!

I love my Jetsons too!

One has to wonder if at some point Nvidia will produce a PCIe carrier board or something for souped up Jetson modules or some variant. I have no idea how the connectivity (PCIe device?) would work out but it's definitely within the realm of possibility.

The Xavier generation supports behaving as a PCIe device (PCIe device mode), so it's maybe not too outlandish. :-)

At that point how is the "GPU" card not just a small motherboard/SoM? I mean, I guess having it run off of a host system still helps with density, but it's 80% of a full computer already at that point.

I keep waiting for Intel's FPGA investments to bear fruit, but so far it seems like it ain't happening.

I am not surprised and pretty much expected. I would love to be wrong and corrected but there is absolutely nothing that suggest Intel's FPGA ( Altera ) are doing any better than before they were acquired, and no signs to suggest future growth either. If anything they are actually doing worst all while Xilinx is storming past them.

As a matter of fact I have yet to see a single Intel Acquisition in its history that has created synergy or additional value.

I am on Xilinx camp, but I am worried what happens with the Altera’s remains. People aren’t crazy about FPGAs paired with Xeon.

Well that entirely depended on Intel's so-called process advantages, didn't it?

> ... I just dont see how they could grow its current record revenue in the next few years.

They'll move into shadier tactics like subscriptions to unlock cores/overclocking, etc.

CUDA is nowhere near the de facto in the ML industry. Beyond some specific application domains like CV where DL makes sense, most ML still runs on commodity hardware, and I happen to believe this will be the case for a long time once the Gartner Cycle of Hype settles on the Deep Learning craze and everyone runs out of VC money.

Despite what cloud providers and NVIDIA want you to believe, no you most likely don't need a lot of compute power for your industry problem. Basic knowledge of statistics would help though.

OP was likely talking about training where NVIDIA has virtually no competition apart from Google's TPUs.

In inferencing a lot more models are deployed on GPUs than before and this trend is likely to continue. NVIDIA is alone there.

Ha I see - I've been training ML models with millions of data points and millions of features on CPUs for a decade now, I didn't know I needed a GPU - I guess I've been doing it all wrong all this time!

>I've been training ML models with millions of data points and millions of features on CPUs for a decade now

That is why it is taking a decade to train your models. If you upgraded to a GPU, you could train them much faster :)

> didn't know I needed a GPU - I guess I've been doing it all wrong all this time

No, not the whole time, just these last couple of years :)

I would be very surprised if a ML pipeline didn't benefit from some GPU acceleration. What kind of algorithms are you using?