NVIDIA discussion [2024]

[Lurkmass]

There’s no barrier for entry for nvidia to enter the console space. And nvidia owns 80% of the GPU space on PC.

They don’t bid for deals because the price requirement for silicon in the console space is so low, margins aren’t worth it and I don’t even know if they have chips that can go so low in margin even if they event want to. I don’t see them supporting anything but Nintendo since they have a bunch of Jetson style chips that would fit that market well.

For AAA centric game console vendors, their only requirement this time around was high graphics perf/area ...

 

[DegustatoR]

For AAA centric game console vendors, their only requirement this time around was high graphics perf/area ...

This wasn't the only requirement which is easy to see if you look at what specs both console GPUs have ended up with.

 

[tunafish]

Nvidia is not "a monopolist" though and their market "power" comes from their products advantage. Saying that you should "play under different rules" here is the same as saying "hey why don't you make your products worse for the sake of competition".

What? No. The rules, as defined in antitrust law, basically say that when you have market power (also, as defined in antitrust law), you can no longer do things like deciding who you sell to based on what benefits you, but you will instead have to start treating all customers the same. There's nothing in it about having to make products worse. As I put in my posts above twice already, there is nothing wrong with having market power, it's just that the rules of what you can do change when you have it. Refusing to ship product to a customer who also buys from your competitor is perfectly legal to do in a fully competitive market. It is not legal to do if you hold market power.

The definition of monopoly doesn’t include “more popular because better at doing the same thing as competing products”.

In the past, at times market share alone has been used to determine market power. Looking at earnings in the sector, Nvidia seems to trivially qualify.

 

[Deleted member 2197]

What? No. The rules, as defined in antitrust law, basically say that when you have market power (also, as defined in antitrust law), you can no longer do things like deciding who you sell to based on what benefits you, but you will instead have to start treating all customers the same. There's nothing in it about having to make products worse. As I put in my posts above twice already, there is nothing wrong with having market power, it's just that the rules of what you can do change when you have it. Refusing to ship product to a customer who also buys from your competitor is perfectly legal to do in a fully competitive market. It is not legal to do if you hold market power.



In the past, at times market share alone has been used to determine market power. Looking at earnings in the sector, Nvidia seems to trivially qualify.

Anti-trust typically applies to larger dominant companies but can also apply to smaller nondominant firms. Rules don't change all that much for nondominant firms.

Although the risks of violating antitrust laws are far smaller for small businesses than for larger businesses (in fact, small businesses are more often the victim than the perpetrator), you still can run afoul of the laws.

Antitrust law is about market power — either yours individually or, more often, in conjunction with that of your competitors. What you have to keep in mind is that not every market is nationwide; a county, a city, or even a smaller area may be found to be a relevant market within which the implications of antitrust can be assessed. Also, you do not have to follow through on a course of action in order to be found liable; it is the conspiracy that most antitrust laws forbid. If you follow through, you're in even bigger trouble.

So, how do you avoid getting involved in activities that might expose you to potential antitrust problems? Don't do any of the following:

• Discuss prices with competitors.
• Discuss contract bid terms with competitors.
• Discuss territories with competitors.
• Join competitors in boycotting another competitor or supplier.
• Require customers to purchase unwanted items in order to get desired items.
• Use your market power unfairly to drive competitors out of business.
• If you are the only or dominant market member in a local area, use unfair methods to keep competitors from entering the market.

 

[Deleted member 2197]

Nvidia bans using translation layers for CUDA software — previously the prohibition was only listed in the online EULA, now included in installed files [Updated]

Translators in the crosshairs.

www.tomshardware.com

Nvidia has banned running CUDA-based software on other hardware platforms using translation layers in its licensing terms listed online since 2021, but the warning previously wasn't included in the documentation placed on a host system during the installation process. This language has been added to the EULA that's included when installing CUDA 11.6 and newer versions.

The restriction appears to be designed to prevent initiatives like ZLUDA, which both Intel and AMD have recently participated, and, perhaps more critically, some Chinese GPU makers from utilizing CUDA code with translation layers. We've pinged Nvidia for comment and will update you with additional details or clarifications when we get a response.
...
The restriction appears to be designed to prevent initiatives like ZLUDA, which both Intel and AMD have recently participated, and, perhaps more critically, some Chinese GPU makers from utilizing CUDA code with translation layers.
...
The clause was absent in the EULA documentation that's installed with the CUDA 11.4 and 11.5 release, and presumably with all versions before that. However, it is present in the installed documentation with version 11.6 and newer.
...
Recompiling existing CUDA programs remains perfectly legal. To simplify this, both AMD and Intel have tools to port CUDA programs to their ROCm (1) and OpenAPI platforms, respectively.

 

[Deleted member 2197]

March 6, 2024

SK Hynix Sends Cutting-Edge 12-Layer HBM3E Memory Samples To NVIDIA For Qualification Tests

The industry-leading HBM manufacturer SK Hynix has reportedly sent samples of its 12-layer HBM3E process to NVIDIA for qualification testing.

wccftech.com

The report comes from ZDNet Korea, which has disclosed that SK Hynix has decided to take the next leap in the HBM3E segment as the firm reaches the testing stages of its 12-layer HBM3E type.
...
For those unaware, the 12-layer type within the standard is much superior, offering much higher capacity, running 36GB per stack compared to the 24GB of the 8-layer HBM3E. Moreover, it is said to be a more efficient process, but as far as we know, 12-layer HBM3E hasn't been implemented in the industry yet. If SK Hynix passes its qualification tests, it might debut with NVIDIA's upcoming H200 AI GPU.
...
Just recently, SK Hynix revealed that they have witnessed outrageous demand from the HBM sector, claiming that this year's supply is already sold out, and SK Hynix is already gearing up for a dominant FY 2025. This isn't surprising at all because AI is still witnessing a huge boom moving into 2024, and with firms such as NVIDIA and AMD preparing for next-gen solutions, it is evident that the demand for HBM will be huge as well.

 

[Cyan]

NVIDIA: no longer a graphics company

CEO sent out an email on Friday evening saying that everything is moving to deep learning and We're no longer a graphics company. Nvidia, known for its longstanding role in video card manufacturing, has traditionally served both consumer and professional markets, maintaining a strong presence in...

www.guru3d.com

 

[pjbliverpool]

NVIDIA: no longer a graphics company

CEO sent out an email on Friday evening saying that everything is moving to deep learning and We're no longer a graphics company. Nvidia, known for its longstanding role in video card manufacturing, has traditionally served both consumer and professional markets, maintaining a strong presence in...

www.guru3d.com

Although arguably neither are AMD or Intel.

 

[DegustatoR]

Nvidia has been trying to be more than "a graphics company" since the launch of G80 and CUDA in 2008. This doesn't mean that they'll abandon graphics anytime soon as it is still very much the basis of everything they are doing.
Also this post on Guru3D is from two months ago.

 

[Bondrewd]

Although arguably neither are AMD or Intel.

was Intel ever a graphics company?

Nvidia has been trying to be more than "a graphics company" since the launch of G80 and CUDA in 2008.

ehhh.
Since Fermi.

 

[Silent_Buddha]

Similar to AMD's dominance and influence over partners in the game console market?

It depends on if AMD is using that dominance to coerce partners into potentially unfavorable contracts. IE - are they withholding or delaying shipments to Sony and/or MS if they don't meet "unwritten" AMD guidelines WRT how much they are being paid? The unwritten part is the contention being levied against NV. Unwritten, because the accused behavior would certainly see them going to court if they were written down in the contract due to their market position.

The question becomes, is this really happening? Even if the accusations are true, it'd be hard to prove in a court of law becase it's not written in a contract. They would need to find evidence in a "paper trail" (any correspondence would suffice, even inter office e-mails and memos). Without that a company can continue to get away with the behavior as long as they leave no evidence that they are doing it "on pupose" or in court it may be termed "in malice" or "premeditated". I'm not saying NV are or are not engaging in this behavior, BTW.

As for AMD, all evidence points to the console makers having more power in negotiations as AMD need the console makers more than the console makers need AMD. This is due to their semi-custom buisiness being overly reliant on Sony and MS without other significant major contractees. Thus, AMD doesn't have significant leverage and thus have to offer beneficial to the customer contracts in order to get their business.

IE - if Sony or MS pull out then AMD suffers immensely. Conversely, if MSI or EVGA (oops one happened already) pulls out of buying GPUs from NV, there are many many other contractees willing and able to (and happily so) absorb any GPUs that may have been allocated to the company that drops out. That gives NV more power than the contractees because it's no loss to them if a customer or two or three pulls out. Perhaps that changes in the future for AMD (if their semi-custom business gets more high volume customers), perhaps it doesn't (Sony and MS remain the largest customers by a long shot).

Regards,
SB

 

[Davros]

ehhh.
Since Fermi.

You could argue before remember nforce motherboards at roughly the same time as the Geforce 2 / 3

 

[Bondrewd]

You could argue before remember nforce motherboards at roughly the same time as the Geforce 2 / 3

Yeah but the genuine GPGPU push started with Fermi and either way didn't really materialize until GK110 won Titan.
Then it got real.

 

[Remij]

Never forget where you came from.

 

[Deleted member 2197]

Will Nvidia GTC Mark The Peak Of AI?

In these explosive times and stock market surges, how to find the nuggets in the most important AI show in the world.

www.forbes.com

Software is one aspect that doesn't get much attention but has the potential to become a major growth component.

Nvidia last quarter indicated the size of its software business for the first time: it has grown to a $1B run-rate business. This includes Enterprise AI, Nvidia’s all-inclusive platform of AI models and tools, as well as Nvidia Omniverse Enterprise. NVIDIA AI Enterprise is available as a perpetual license at $3,595 per CPU socket with Enterprise Business Standard Support at $899 annually per license. So nearly $5K per user!

This stuff is super sticky. And incredibly valuable. Mercedes, for example, is using Omniverse to create a digital twin of a new factory, and will continue to use Omniverse to manage and evolve that factory over time. While Meta has some toys and avatars in its metaverse world, Nvidia has real engineers and creators collaborating and simulating (with real physics) digital worlds that will become reality. This is some of the most exciting stuff Nvidia is working on.

Given the massive success of Nvidia AI hardware, further monetizing that installed base will be a major opportunity for Nvidia going forward. $1B is just the starting point.

 

[DavidGraham]

CEO of NVIDIA on why some competitors choose to focus on Inference.

https://twitter.com/x/status/1766343225407848790

 

[Arun]

CEO of NVIDIA on why some competitors choose to focus on Inference.

https://twitter.com/x/status/1766343225407848790

Not sure I agree, he's right that training is a systems problem rather than just a chips problem, which makes it more difficult technically (i.e. you need to solve 2 separate-but-related problems at the same time). And he's right that their install base is an advantage to get new models deployed on them first.

But once you have a successful model (e.g. Copilot or ChatGPT at MS/OpenAI) or you're about to deploy an improved version of the same thing (e.g. GPT4 Turbo), and you know you know you've got huge scale, *and* you still need your existing NVIDIA GPUs to train the model *after* that, your only focus is going to be cost, cost, and cost (although it's full system cost including engineering salaries & power/cooling/etc.) - that cost includes the cost of highly paid engineers optimising the model for another architecture, and also even designing new hardware themselves in Microsoft's case... so again, it only works with scale, but once you're at that point I don't see why the install base is relevant at all.

So it kinda depends how much of the inference market is for "extremely successful large scale deployment models" vs "random startup which made a thing". I would bet on the former to be >>50% of inference flops, so I don't really buy Jen-Hsun's argument in general although it does apply to part of the market.

 

[trinibwoy]

His answer seems focused on cloud inference which of course Nvidia has a first mover advantage in. The edge inference market is still wide open and Intel/AMD/Qualcomm/Apple will be mixing it up there.

 

[Bondrewd]

The edge inference market

Doesn't really exist?
Phones have had GEMM blobs since Kirin 970 we're still somewhere there where it all began.

 

[trinibwoy]

Hence wide open.