Nvidia To Invest $5 Billion in Intel (ft.com) 49
Nvidia has agreed to invest $5 billion in its struggling rival Intel [non-paywalled source] as part of a deal to develop new chips for PCs and data centres, the latest reordering of the tech industry spurred by AI. From a report: The deal comes a month after the US government agreed to take a 10 per cent stake in Intel, as Donald Trump's administration looks to secure the future of American chip manufacturing.
However, the pair's announcement makes no reference to Nvidia using Intel's foundry to produce its chips. Intel, which has struggled to gain a foothold in the booming AI server market, lost its crown as the world's most valuable chipmaker to Nvidia in 2020. On Thursday Jensen Huang, Nvidia's chief executive, hailed a "historic collaboration" and "a fusion of two world-class platforms," combining its graphics processing units, which dominate the market for AI infrastructure, with Intel's general-purpose chips. Further reading: Intel Weighed $20 Billion Nvidia Takeover in 2005.
However, the pair's announcement makes no reference to Nvidia using Intel's foundry to produce its chips. Intel, which has struggled to gain a foothold in the booming AI server market, lost its crown as the world's most valuable chipmaker to Nvidia in 2020. On Thursday Jensen Huang, Nvidia's chief executive, hailed a "historic collaboration" and "a fusion of two world-class platforms," combining its graphics processing units, which dominate the market for AI infrastructure, with Intel's general-purpose chips. Further reading: Intel Weighed $20 Billion Nvidia Takeover in 2005.
Not really a rival (Score:5, Interesting)
More like a frenemy - Nvidia does not have any sort of market in the datacenter where Intel competes, at all, and would much rather see Intel Xeon chips and motherboards running their AI compute stack than AMD Epycs - because AMD is the *real* rival to where Nvidia makes the bulk of their money (hint: its not to gamers)... Intel compute cards cannot compete on the enterprise level at all with Nvidia and in fact there's a synergy there since they both want the same thing - AMD to lose.
Re: (Score:2)
Nvidia does not have any sort of market in the datacenter where Intel competes
*used to compete, since Intel hasn't had any parts competitive with AMD for years now.
Nvidia is also bringing out their own ARM-based servers, so far the point is to run their GPGPUs cheaper than with amd64 but when the AI bubble collapses they may well have to pivot in that direction to keep up DC sales.
AMD is the *real* rival to where Nvidia makes the bulk of their money
Maybe, except that their AI chips aren't as good at running LLMs as Nvidia's CUDA cards.
Re:Not really a rival (Score:4, Interesting)
You obviously don't work in a datacenter or with enterprise hardware - it doesn't matter what benchmarks you're looking at, large companies are still buying tons of Intel chips because they already have a ton of Intel chips. And you're right, CUDA has more market penetration than ROCm but that doesn't change what I said - AMD is the only *real* rival to the products Nvidia makes.
The more DC start to get AMD, the better their automation teams will get at patching the firmware and performance tuning for Epyc, which again is an entirely different skillset than just "rack and run". Should companies start to *actually buy* MI35x en masse, then Nvidia will feel their biggest profit/margin maker threatened.
This investment is ensuring people don't run *any* AMD anything in the DC.
Re: (Score:2)
You obviously don't work in a datacenter or with enterprise hardware - it doesn't matter what benchmarks you're looking at, large companies are still buying tons of Intel chips
AMD has been outselling Intel in the DC for what, a year now or more?
This investment is ensuring people don't run *any* AMD anything in the DC.
I don't think you know what "ensuring" means
Re:Not really a rival (Score:4, Insightful)
I don't think you know what "ensuring" means
All evidence so far points to nobody knowing.
Re:Not really a rival (Score:4, Informative)
I wasn't trying to be insulting though I can see I hit a sore spot with you. AMD is not outselling Intel in the DC - they had a good year but I don't know of any DC that is buying more AMD than Intel chips, unless its an AMD only company, and there are a few...Giving you the benefit of the doubt (which google searches easily show is wrong), this STILL means the vast majority of hardware, today, in a DC, is STILL Intel. Enterprises aren't just throwing away their servers. Hell my last gig ~25% of our servers were 10 years old (and buying 5k new every quarter)
Nvidia absolutely wants to ensure that Intel remains on top because its in their best interest. On Top doesn't mean who has sold more chips to gamers last year - if you walk into any DC you'll find 75-90% of their silicon is Intel, and if its an AI DC a higher percentage will be Nvidia compute.
For the record, I own lots of AMD and think they have much better products than Intel, but that doesn't change the enterprise reality of momentum, though its changing rapidly and Nvidia sees this...
Re: (Score:2)
I'm more on the fence - I think AMD has a far superior CPU these days, but I still vastly prefer nVidia graphics hardware, even though I try and go back to AMD every few years. There always seems to be something just hideously broken with new features on the AMD side. Keep in mind I don't use DirectX, I use Vulkan and used to use OpenGL because I need to cross-compile. Sure it gets fixed when I file a bug report, but I shouldn't be their product tester, lol. I think AMD tests exclusively on Windows, then us
Re: (Score:2)
I've had the opposite experience with *consumer graphics*, but granted I use linux exclusively. For games and graphics, AMD Radeon *just works* and I have no problems whereas Nvidia does not especially with things like sleep states. Switch that around for ML or compute and I can say ROCm doesn't work at all (again for *consumer graphics*), whereas CUDA *just works* due to AMD's refusal to actually support consumer grade hardware in ROCm. Then again, AMD with vulkan is fine assuming the ML product you're usi
Re:Not really a rival (Score:5, Interesting)
AMD has been outselling Intel in the DC for what, a year now or more?
No, AMD still trails Intel in the data center, both in terms of revenue and unit sales [hothardware.com]. AMD's server share was stagnating at 20% for many years despite an uptick in reputation and positive press. It's only been recently this year that AMD hit around 40% in server market share, and that's based on revenue. In terms of units, AMD's market share is lower at around 32%.
Of course, upward momentum is still with AMD, so it wouldn't be surprising to see AMD claim a majority of server market share in the near future.
Re: (Score:2)
AMD has been outselling Intel in the DC for what, a year now or more?
No, AMD still trails Intel in the data center, both in terms of revenue and unit sales [hothardware.com]. AMD's server share was stagnating at 20% for many years despite an uptick in reputation and positive press. It's only been recently this year that AMD hit around 40% in server market share, and that's based on revenue. In terms of units, AMD's market share is lower at around 32%.
Of course, upward momentum is still with AMD, so it wouldn't be surprising to see AMD claim a majority of server market share in the near future.
What's surprised me in the processor market is AMD making inroads in the laptop space that Intel owned for decades, even when AMD were dominating on the desktop and had very strong server offerings (the Athlon 64/Opteron days, for those who's memories stretch back that far). 15% of laptops are now AMD.
Re: (Score:2)
What's surprised me in the processor market is AMD making inroads in the laptop space that Intel owned for decades
AMD has superior power management in their chip, in terms of being able to shut down functional units when not in use, so now that they are on the same or superior process technology they have an efficiency edge over Intel. Intel was only ahead of AMD in mobile because they had superior process tech. That is now over and there are no signs it's coming back (Intel having failed at getting acceptable yields with two process shrinks in a row now) so AMD has the clear edge for the time being.
Re: (Score:2)
Nvidia makes better shit than AMD across the board, except in market segments that Nvidia doesn't offer products into. Period.
Nvidia GPUs always outperform AMD, and are the industry standard for AI workloads. End of story.
Nvidia Jetson CPUs outperform AMD in the embedded space while bringing more features in a smaller footprint, at less cost in both module price and board design complexity, and less wattage.
It's only a matter of time until the Jetson AGX module starts getting design wins in medium-wattage
Re: (Score:2)
NVIDIA might be absolute performance leader, but when the NVIDIA part is sold out, the AMD guy is can get you a pallet of cards that do the workload at the same number of watts. Making margin is better than waiting on making slightly more margin, we are in the gold rush, and dam strait I want any shovels and Pans and blue jeans to sell to the idiots running for the goldfield. And some times you have to tell an investor you have put x dollars of hardware under a roof, if INTEL and A
Re: (Score:3)
*used to compete, since Intel hasn't had any parts competitive with AMD for years now.
Intel still has a large share of the datacenter market, whether they really deserve it or not. Hence they actually have competition. Versus their flailing around accelerators which both sucked and never got traction. Just like I went looking for a new laptop and in some segments, the vendors only did Intel even as AMD also has everything better in that market too.
Nvidia is also bringing out their own ARM-based servers, so far the point is to run their GPGPUs cheaper than with amd64 but when the AI bubble collapses they may well have to pivot in that direction to keep up DC sales.
Well, they actually have had that available for a bit of time with Grace Hopper. I don't have hard data, but anecdotally it feels like this and r
Re: (Score:2)
their AI chips aren't as good at running LLMs as Nvidia's CUDA cards.
Quite, but AMD is *miles* closer than Intel was to being a realistic threat on this front.
This is very true, and perhaps continuing to close the gap is their strategy for defeating CUDA, since they sure aren't putting enough effort in to do it with ROCm. I'd love to see it, I just don't expect it to work. Happy to be proven wrong, though.
Re: (Score:2)
NVLink haven't had the uptake nVidia presumed.
I'm not sure what this means. NVLink is baked into each GPU, and most customers buy racks rather than single GPUs. These racks have NVLink switches included. So, NVLink has essentially sold just as well as GPUs.
Re: (Score:2)
Have they done rack scale nvlink other than Grace? Usually I see racks of systems with GPUs, but the NVLink terminates within each server, rather than going between servers.
NVLink nowadays even in a single system (as of Blackwell) looks a lot more like infiniband, but still seeing Blackwells usually as GPUs in x86 boxes with NVLink staying inside. Multi-server fabric seems to be RDMA over ethernet as the favored selection for now. Maybe I'm missing some segment, but I can at least say externally switched
Re:Not really a rival (Score:4, Interesting)
Compare the market caps though, Nvidia and Intel are not on the same order of magnitude.
I don't think there is much a of frenemy relationship really to speak of.
My guess is this about two things:
1) Nvidia ensuring they have or could get some access to an x86 license if AMD is somehow able to both make some kind of great leap in MIMD compute space and at the same time is able to deliver some kind of integration advantage with integration in traditional compute in memory architecture with EPYC parts.
2) Being sure they have access to some kind of FAB capacity in the event the excrement hits the fan around TSMC, and with a "partner" to whom they could dictate terms.
Re: (Score:2)
For your point (1) I think AMD will have a say on who else gets a licence for x86, since pretty much x86 64 bits was invented by AMD (and so called AMD64 in the more technical circles).
Unless Nvidia only wants to make a 32bits x86 chip ....
Re: (Score:2)
Compare the market caps though, Nvidia and Intel are not on the same order of magnitude.
I don't think there is much a of frenemy relationship really to speak of.
My guess is this about two things:
1) Nvidia ensuring they have or could get some access to an x86 license if AMD is somehow able to both make some kind of great leap in MIMD compute space and at the same time is able to deliver some kind of integration advantage with integration in traditional compute in memory architecture with EPYC parts.
2) Being sure they have access to some kind of FAB capacity in the event the excrement hits the fan around TSMC, and with a "partner" to whom they could dictate terms.
I think everyone is over thinking this. NVidia need to do something with that money they're drowning in and there's only so much that can be syphoned off to tax havens before everyone starts asking questions. If they kept it, they'd need to pay tax on it and we can't have that.
Re: (Score:2)
You forgot when Intel sued Nvidia for having the audacity to use their licensed Intel IP to create chipsets that support Intel CPUs.
Intel had to pay Nvidia $1.5B and renew a cross-licensing agreement to get out of that self-inflicted GSW to the foot.
Hope (Score:3)
Re: (Score:2)
Re: Hope (Score:2)
There's no reason to kill ARC because it's only taking sales Nvidia didn't want.
Insurance bet (Score:5, Insightful)
Re: (Score:3)
Re: (Score:1)
I've been hearing about China invading Taiwan for decades now.
It's Cold War 1.5 - and a valid threat (Score:1)
I've been hearing about China invading Taiwan for decades now.
Fair criticism, but Ji has explicitly made serious threats and China has been building up it's Navy...and they're getting more desperate due to a demographic collapse as well as a trade war which they're losing (we're not winning...we're losing as well)...but things are dire for China. Their window of a successful invasion keeps shrinking. In fact, it could be a piece of cake if it wasn't for TSMC. So while they've been planning on invading since before I was born, they're actually taking unprecedented s
Re: (Score:2)
How Taiwan is supposed to solve the demographics problem is beyond me though.
Seeing how the Chinese government is killing major industdies like cars and solar over there, having TSMC would be stop-gap at best.
Taiwan, realistically speaking, is nothing but a propaganda stund to keep the party alive a bit longer.
Exactly - it's a stunt to stay in power (Score:1)
How Taiwan is supposed to solve the demographics problem is beyond me though.
Their concern is they won't have the soldiers or economy to invade Taiwan as easily in the future. For them, it's an act now or regret it in the future calculus...and to your point, the only reason I can see for them to want to own Taiwan is as a manipulation of the population and national identity. "Don't question that you property is under water and your job prospects are shrinking and everyone you know is getting poorer...One China!! One China!! One China!!!...support the troops!!!"
Re: (Score:2)
Re: (Score:3)
Re: (Score:3)
Intel can't even make their own chips at a competitive level when it comes to latest and greatest techniques. ;) If there's some problem where TSMC can't provide chips, Nvidia AND AMD AND Intel will all have problems, along with just about every other chip manufacturer.
Re: Insurance bet (Score:2)
Samsung is still an option if TSMC and Intel fail. Not as good as TSMC, but leading edge enough.
Re: Insurance bet (Score:3)
I agree 107% with you. Just a few points:
Samsung is an option for fabinng advanced chips.
TSMC and Samsumg have fabs in the USoA
nVIDIA already worked with samsung (RTX 30x0 and similar AI chips)
nVIDIA invested in Intel because it is cheeeeeeap right now AAAAAND because, in the current political climate, investing in a non-USoAn company is frowned upon.
Re: Insurance bet (Score:2)
"TSMC and Samsumg have fabs in the USoA"
How long before the tangerine terror fucks that up?
Re: (Score:2)
"TSMC and Samsumg have fabs in the USoA"
How long before the tangerine terror fucks that up?
Your question is valid. The answer is simple. the "Tangerine Terror" can only stop new fabs getting here. The fabs that TSMC and Samsung already have in USoAn soil will remain there AND be upgraded.
The sole process of building a fab that can sustain the cleanliness needed for chipmaking (significantly more clean particle wise than an Operating Room, although less asceptic) is super-Expensive. even before you deploy the Litho Equipment. Also, the load bearing on the ceiling for the waffer transport pods is
Re: (Score:2)
Nvidia is still heavily dependent on x86 CPUs to pump data into their gpus. That is changing a bit with arm but it's still a thing.
It also helps Nvidia when the European governments come sniffing around about antitrust violations. I say European because God knows American government's not going to do anything. We might shake them down for some bribes for the current adm
This is nothing more than... (Score:2)
This is nothing more than what Microsoft did with Apple in the 90's. They're worried about getting hit with anti trust laws because they dominate so much of the market so they're trying to prop up failing competition.
Re: (Score:2)
Lol, this is just Trump strong-arming yet again. He's told Nvidia to put some money in the pot so things won't go badly for Nvidia. Of course, like any wannabe gangster, Trump changes his mind on a dime.
Re: (Score:2)
Re: (Score:2)
Loongson Technology (Score:2)
Re: Loongson Technology (Score:2)
Loongson is a very capable MIPS derived design. Its claim to fame was a set of extensions that accelerated emulation of X86 VMs via QEMU.
nVIDIA,meanwhile, is NOT interested in X86 (or MIPS, for that matter). They are interested on ARM for the short and medium term, and on RISC-V for the long term.
Try the same idea agai, but with an ARM or RISC-V company as the acquisition target.
And about the geopolitical implications of acquiring a chinese processor developer, I will not talk.
digging the hole for Intel to fall into (Score:2)
and then buying what's left of Intel for a few bucks.
Clever strategy!
reminds me of microsof investing on apple (Score:2)
And please remember that nVIDIA also makes networking gear (mellanox aquisition) and miscelaneous chips. So, it would not be out of the questions to throw intel a bone, and fab that (and not AI chips) on intel.
Reads: Nvidia Invests in AMD's Competitor (Score:2)
Re: (Score:2)
What people dont like, instead of tree hugging Northern California Hippies with eastern p