NVIDIA's NVLink Fusion Stays Proprietary, Third Parties Can Only Work Around It

[Rover4444]

Someone's code for sure, not a corporation..


I am not sure if I can get behind that one. I guess it would depend.

Corporations also release open source code. NVIDIA themselves have open kernel modules available for some of their cards.

 

[Patriot]

Corporations also release open source code. NVIDIA themselves have open kernel modules available for some of their cards.

XD, not by choice...

 

[LastDudeALive]

On topic, Nvidia doesn't have a performance lead. They have a proprietary lock-in technology scheme going. Their hardware performs nominally to AMD's for instance but customers are locked into CUDA and other proprietary Nvidia products. This is very bad for everyone.

Okay, in the datacenter, professional, and consumer markets, this is just silly. Everybody whines about how Nvidia has ~90% marketshare, but nobody takes a second to think of WHY 90% of people and companies are using Nvidia's products.

Datacenter: Nvidia offers high powered compute, networking, and ease of use (CUDA), all in a single package. Dealing with multiple vendors and systems is a IT department's WORST nightmare. If your datacenter goes down, you don't want to have to wrangle representatives from the GPU company, the CPU company, the networking company, and the software vendor into a conference call at 3 AM to figure out who's fault it is. If you get everything from Nvidia, you have 1 person to call. This is HUGE for businesses, and it absolutely is a competitive advantage. Even if the base hardware isn't 10x better, their ability to deliver value is 100x better than their competitors, which is why they are such a huge player in the HPC market. Only companies big enough to handle all that on their own (like AWS, Google, etc) can afford to not use them.

Professional: Again, Nvidia has a virtual monopoly here because they have invested the time into making their products easy to use (CUDA). Why are customers "locked in" to CUDA? Maybe it's because CUDA is the best way to write their software. It's very telling that AMD has imitated CUDA with ROCm, and it's also very telling that people continue to use CUDA, because AMD is just worse at offering support (e.g. ROCm didn't support RDNA 4 at launch).

Consumer: Nvidia absolutely does have the performance lead. AMD was 2 generations behind with ray tracing and AI. You may weep and gnash your teeth, but RT and AI sells GPUs now. This is what consumers want, and they're voting with their wallets. It's not "lock-in," the DirectX API is open. AMD just sucks at giving consumers what they want. Even Alchemist beat RDNA 3 in RT and AI.

 

[JustBenching]

Streamline is open source but amd doesn't want to play ball. Amd only supports open source when it benefits them. Disgusting company.

 

[Patriot]

Streamline is open source but amd doesn't want to play ball. Amd only supports open source when it benefits them. Disgusting company.

WTF is streamline, never heard of them. Sounds like Tinycorp making a Radeon based server and offering it for sale before testing it and expecting AMD to bail them out.
Maybe they are referring to Nvidia-streamline released in 2023, allowing other vendors to use Nvidia's framework for integrating DLSS into games. Competing with the existing GPUOpen frameworks that already allowed this.... Following a hack that threatened to release source code for dlss and nvidia drivers.... prompting them to release a few opensource things... which competed with existing opensource projects.

Ah yes, they should definitely stop the opensource things they have been doing for a decade and start supporting this new thing the competitor begrudgingly released immediately.

 

[Hecate91]

Amd only supports open source when it benefits them. Disgusting company.

As opposed to the company of this thread topic not supporting open source, lol

 

[Daven]

Okay, in the datacenter, professional, and consumer markets, this is just silly. Everybody whines about how Nvidia has ~90% marketshare, but nobody takes a second to think of WHY 90% of people and companies are using Nvidia's products.

Datacenter: Nvidia offers high powered compute, networking, and ease of use (CUDA), all in a single package. Dealing with multiple vendors and systems is a IT department's WORST nightmare. If your datacenter goes down, you don't want to have to wrangle representatives from the GPU company, the CPU company, the networking company, and the software vendor into a conference call at 3 AM to figure out who's fault it is. If you get everything from Nvidia, you have 1 person to call. This is HUGE for businesses, and it absolutely is a competitive advantage. Even if the base hardware isn't 10x better, their ability to deliver value is 100x better than their competitors, which is why they are such a huge player in the HPC market. Only companies big enough to handle all that on their own (like AWS, Google, etc) can afford to not use them.

Professional: Again, Nvidia has a virtual monopoly here because they have invested the time into making their products easy to use (CUDA). Why are customers "locked in" to CUDA? Maybe it's because CUDA is the best way to write their software. It's very telling that AMD has imitated CUDA with ROCm, and it's also very telling that people continue to use CUDA, because AMD is just worse at offering support (e.g. ROCm didn't support RDNA 4 at launch).

Consumer: Nvidia absolutely does have the performance lead. AMD was 2 generations behind with ray tracing and AI. You may weep and gnash your teeth, but RT and AI sells GPUs now. This is what consumers want, and they're voting with their wallets. It's not "lock-in," the DirectX API is open. AMD just sucks at giving consumers what they want. Even Alchemist beat RDNA 3 in RT and AI.

The performance leads you mention are nowhere near enough to garner 90% of the market especially at near 80% margins!!! Nvidia can sell more just not that much more for that much higher cost. When Intel had 90% plus of the market, AMD performance was abysmal. Like, calculator abacus different. It was bad at that time.

No, Nvidia has 90% marketshare because of lock-in. Being 10% better in gen ras, 20% better in RT and similar in data center, doesn't allow a company to charge FIVE times more than the competition and have that much marketshare.

 

[Patriot]

The performance leads you mention are nowhere near enough to garner 90% of the market especially at near 80% margins!!! Nvidia can sell more just not that much more for that much higher cost. When Intel had 90% plus of the market, AMD performance was abysmal. Like, calculator abacus different. It was bad at that time.

No, Nvidia has 90% marketshare because of lock-in. Being 10% better in gen ras, 20% better in RT and similar in data center, doesn't allow a company to charge FIVE times more than the competition and have that much marketshare.

It's both, they have an army of software engineers developing the ecosystem and an army of ruthless sales engineers pushing their solutions.
They have the most mature ecosystem that is easy to use... everyone else has hacked together solutions.

AMD competes in raw performance but, the software isn't there except in certain workloads, and because of that, AMD is gaining momentum in HPC and cloud vendors. They are working on that but they are miles behind in general usefulness. As a partner to both companies... I was telling them in 2018 that they needed to work on the ecosystem and until kids could run ROCm on their college laptops they wouldn't have an easy path for gaining new developers.

Have you ever looked at Nvidia's enterprise software? It is baller... and I am saying that with the full passion that I despise the companies operating parameters.
I do not believe it is necessary, they could let the hardware and software speak for itself... there is no need for their behavior, they act scared of UALink, and competitive offerings. They have the best ecosystem, but they arent acting like it.

 

[Daven]

It's both, they have an army of software engineers developing the ecosystem and an army of ruthless sales engineers pushing their solutions.
They have the most mature ecosystem that is easy to use... everyone else has hacked together solutions.

AMD competes in raw performance but, the software isn't there except in certain workloads, and because of that, AMD is gaining momentum in HPC and cloud vendors. They are working on that but they are miles behind in general usefulness. As a partner to both companies... I was telling them in 2018 that they needed to work on the ecosystem and until kids could run ROCm on their college laptops they wouldn't have an easy path for gaining new developers.

Have you ever looked at Nvidia's enterprise software? It is baller... and I am saying that with the full passion that I despise the companies operating parameters.
I do not believe it is necessary, they could let the hardware and software speak for itself... there is no need for their behavior, they act scared of UALink, and competitive offerings. They have the best ecosystem, but they arent acting like it.

So this I can agree with. The amount of work Nvidia put into the proprietary software department was a brilliant move. So much enterprise solutions are now locked into using it that it is extremely expensive to jump ship. Both AMD and Nvidia has good hardware. The hardware differences are nowhere near enough to justify the marketshare and margins but the proprietary software, OMG, Nvidia planned that move so carefully and it really paid off.

By the way, talking about hardware, I would like to propose an analogy for those who think Nvidia hardware is so much better. Are there any car companies that have 90% of the market and charge 80% margins? I mean cars and trucks are a huge, integral part of our society. Companies and individuals need automobiles basically to survive at this point. One could argue that Ford had a monopoly at one point but they INVENTED the modern day automobile industry. There were no other players. Nvidia didn't invent the GPU and they never had a GPU that was so much better than any other company. But to go back to the analogy, if Ford had made their cars in such a way that only Ford cars could work on a road designed by Ford and all roads were Ford built then that would give them a monopoly. This is how Nvidia is winning. They built the entire infrastructure while other manufacturers just made a chip with some drivers.

 

[freeagent]

Corporations also release open source code. NVIDIA themselves have open kernel modules available for some of their cards.

I understand, but why does this one hit differently?

 

[Vayra86]

Ahhh yes, another of those "the more you buy, the more you save" charts...

View attachment 404362

Everything stays proprietary.... until it no longer is financially viable (or forced to by some FRAND thing)...
At this point the only people sufferring from any potential price gouging is those who are willing to keep paying money to Nvidia for AI products...

The revenue even glows brightly. I mean, wow!

 

[LastDudeALive]

The performance leads you mention are nowhere near enough to garner 90% of the market especially at near 80% margins!!! Nvidia can sell more just not that much more for that much higher cost.

For enterprise customers, yes, it is. Nvidia's ecosystem is absolutely worth the premium. If your business will lose millions of $ every minute your datacenter is down, you will pay whatever Nvidia charges to guarantee everything works together. You obviously don't understand the priorities and mindset of large corporations that now make up ~90% of Nvidia's revenue.

When Intel had 90% plus of the market, AMD performance was abysmal. Like, calculator abacus different. It was bad at that time.

Look at it this way. Maybe since Nvidia has 90% of the GPU market, that means AMD's performance is actually abysmal (which again, in AI and RT for RDNA 1, 2, and 3, it was). You obviously can see that Intel's market dominance was because AMD sucked. Why is it so hard to accept that Nvidia's market dominance in the consumer GPU market actually means AMD sucks there as well? That's the most reasonable conclusion.

 

[Daven]

For enterprise customers, yes, it is. Nvidia's ecosystem is absolutely worth the premium. If your business will lose millions of $ every minute your datacenter is down, you will pay whatever Nvidia charges to guarantee everything works together. You obviously don't understand the priorities and mindset of large corporations that now make up ~90% of Nvidia's revenue.


Look at it this way. Maybe since Nvidia has 90% of the GPU market, that means AMD's performance is actually abysmal (which again, in AI and RT for RDNA 1, 2, and 3, it was). You obviously can see that Intel's market dominance was because AMD sucked. Why is it so hard to accept that Nvidia's market dominance in the consumer GPU market actually means AMD sucks there as well? That's the most reasonable conclusion.

Corporations can save millions by going AMD because of Nvidia high prices. The reason they cannot is due to anti-competitive behavior. What you are describing as smart business decisions is the exact reason we have anti-trust laws.

You can’t hold a gun to someone and say its a healthy living decision to pay you not to fire. This is very easy stuff to understand.

 

[Rover4444]

XD, not by choice...

Yes, by choice. Who do you think did it, Intel? Speaking of, there's plenty of open source software that both AMD and Intel have released.

I understand, but why does this one hit differently?

It's all in your head. This sort of thing has already happened with USB4 versus Thunderbolt, no reason to shit on people for liking the open implementation more.

 

[LastDudeALive]

Corporations can save millions by going AMD because of Nvidia high prices. The reason they cannot is due to anti-competitive behavior. What you are describing as smart business decisions is the exact reason we have anti-trust laws.

It's not anti-competitive to offer a better product, buddy. Nvidia offers a full product stack they've spent billions of $ developing (CPU, GPU, networking, and software), and that is incredibly valuable to companies. AMD has not. They don't have networking IP, their software stack is trash compared to CUDA, they dropped the ball on AI. If you can't comprehend why a Fortune 500 company would prefer to have a unified solution for their data centers instead of dealing with 10 different vendors, you shouldn't be commenting on this.

You can’t hold a gun to someone and say its a healthy living decision to pay you not to fire. This is very easy stuff to understand.

Nvidia is not holding a gun to a billion $ corporation. Nvidia is offering an entire car instead of selling you the individual parts and making you build the car yourself. The choice is obvious.

 

[freeagent]

It's all in your head.

No it isn't. I honestly do not care one iota. I just wanted to know why this was bugging people.

 

[Rover4444]

I just wanted to know why this was bugging people.

This is what I'm talking about.

 

[Visible Noise]

Maybe it's finally the time to ditch Nvidia for ever?
My next setup in 2026 will be fully based on Intel.

Yeah, that’ll teach Nvidia not to use their proprietary interconnect on their data center GPUs.

Lightmatter with help of Ayar Labs will annihilate Nvidia in AI very soon.

I wish I could laugh at this more than once.

Your point would be relevant if there was no such thing as open source. Open source is also someone's code but that someone freely shares it. In a better world, open source centric companies should be on top and closed, proprietary companies would be on bottom or not exist at all.

Why?

This is the dumb self referencing argument of open source is better because it’s open source.

Next up: The more eyeballs fallacy.

 

[wolf]

People are addicted to performance and top of the line graphics, that's why they buy Nvidia. Nobody cares about the company, we care about the products.

Well I wouldn't say nobody cares, there's plenty of non-Nvidia users on these forums that are effectively obsessed with Nvidia, and above all - Jensen.

I rarely, if ever, hear the owners of Nvidia products talk about Jensen, but those folks can't seem to stop talking about him or his company.

The icing on that cake is the attempts at projecting those feelings onto Nv users and claiming that they're the ones with the obsession.

 

[Visible Noise]

Corporations can save millions by going AMD

Corporations can spend millions by going AMD for all the development costs.

The cost of hardware is a small fraction of the cost of running a data center scale compute platform. Nvidia’s software is plug and play. AMD’s software is fix it for them before it will even begin to execute.

 

[TheinsanegamerN]

Your point would be relevant if there was no such thing as open source. Open source is also someone's code but that someone freely shares it. In a better world, open source centric companies should be on top and closed, proprietary companies would be on bottom or not exist at all.

In this better world, are people working for free? Because that is the implication I'm getting here, which is the FOSS line of "everything is open and free as in beer" which works great until you have to feed your family. These closed proprietary systems do suck for the end consumer, but they also benefit the end consumer in the form of strong support from the market which is a necessity in corporate environments.

AMD has open source alternatives for a lot of Nvidia software. None of it is developed to anywhere near the same degree, and if you need help, you are SOL.

The performance leads you mention are nowhere near enough to garner 90% of the market especially at near 80% margins!!! Nvidia can sell more just not that much more for that much higher cost. When Intel had 90% plus of the market, AMD performance was abysmal. Like, calculator abacus different. It was bad at that time.

No, Nvidia has 90% marketshare because of lock-in. Being 10% better in gen ras, 20% better in RT and similar in data center, doesn't allow a company to charge FIVE times more than the competition and have that much marketshare.

Nowhere near? For the last 4 generations in the data center, there has been NOTHING opposing them. AMDs token efforts are like a lemonade stands effect on the local McDonald's.

For gaming, AMD hasn't had a truly competitive flagship since Hawaii. In 2013. That was 12 years ago. Everything since then has been a massive compromise or simply doesn't compete at the high end.

That "lead" you talk about is so huge that many don't even consider AMD an option.

 

[Rover4444]

This is the dumb self referencing argument of open source is better because it’s open source.

The only reason why open source doesn't dominate every single aspect of computing is because of binary blobs and vendor lock-in. Mesa shits on proprietary Radeon drivers but NVK will probably never be good.

In this better world, are people working for free? Because that is the implication I'm getting here, which is the FOSS line of "everything is open and free as in beer" which works great until you have to feed your family.

Somebody's not paying attention to the literature. FOSS (now FLOSS because of people like you) is free as in freedom, NOT free as in free beer. I guess you should tell Red Hat their business model isn't sustainable either.

These closed proprietary systems do suck for the end consumer, but they also benefit the end consumer in the form of strong support from the market which is a necessity in corporate environments.

No they don't. Strong market support isn't a guarantee for any system, proprietary or not.

 

[JustBenching]

The performance leads you mention are nowhere near enough to garner 90% of the market especially at near 80% margins!!! Nvidia can sell more just not that much more for that much higher cost. When Intel had 90% plus of the market, AMD performance was abysmal. Like, calculator abacus different. It was bad at that time.

No, Nvidia has 90% marketshare because of lock-in. Being 10% better in gen ras, 20% better in RT and similar in data center, doesn't allow a company to charge FIVE times more than the competition and have that much marketshare.

They were not 20% better in RT though. In games that you wanted to use it, the difference was ~50% at the same segment. Yes, as big as the difference between intels chips and bulldozer at the time

 

[Rightness_1]

This is why it's so good nGreedia didn't get the console wins.

 

[Vya Domus]

This is why it's so good nGreedia didn't get the console wins.

They priced themselves out of that segment, no one would pay what they would ask for.