Nvidia shows signs in [2023]

[neckthrough]

Yes but even with this we're looking at GM falling over the course of Lovelace launch, not increasing or being at historical heights. Current Nv's margins are about on par with the period between 2011 and 2016 - which is Kepler/Maxwell/Pascal gens generally lauded as one of the best in perf/price gains. The rise after that is linked to the first crypto mining boom, then covid, then the second boom. I'm struggling to find any indication of Lovelace bringing unusually high margins in any of the data on hand yet people continue saying this as if it's a proven fact at this point.
Thus the question (multitude of them really) remain.

Right now is definitely boom time. I believe GM guidance for next Q is an all-time high. But it's all ChatGPT/LLM driven. Everyone and their mother is buying datacenter GPUs.

But who knows, maybe it's all 4060s .

 

[DegustatoR]

Right now is definitely boom time. I believe GM guidance for next Q is an all-time high. But it's all ChatGPT/LLM driven. Everyone and their mother is buying datacenter GPUs.

Yes, this is already reflected in the results of the last quarter where margins started to rise - it's easily attributable to high margin DC product sales. I fully expect the usual suspects to start painting this as if it's all coming from incredible sales of 4060s.

 

[digitalwanderer]

Yes, this is already reflected in the results of the last quarter where margins started to rise - it's easily attributable to high margin DC product sales. I fully expect the usual suspects to start painting this as if it's all coming from incredible sales of 4060s.

Just wanted to mention I didn't mean the laughing emoji insultingly, I really thought it was a funny comment.

 

[Father_Murphy]

It's a little of both. The price is where about 3/4 of the previous market can bear it. That's roughly how far down shipments of discrete GPUs are. The price is also where the market leader sets it. They could lower it and sell more consumer discrete GPUs, but they have no reason to when they could instead sell more AI silicon.

So, that allows them to maintain their high margins and even increase them because there's no downside to the consumer market shrinking due to inflated pricing (they're still sticking to the inflated covid/crypto mining bubble pricing levels). Basically, any shrinkage in the consumer dGPU market due to high prices is more than offset by the increased demand by the even higher margin burgeoning AI market.

The competition also isn't interested or able to significantly challenge that as they also would prefer to shift silicon allotments to more lucrative AI buyers and their console partners.

This is a succinct, well-articulated analysis and echoes my thoughts as well! It would seem that the market and its players are still adjusting to find the sweet spot in pricing, volume, margin, and opportunity costs. From what I've seen, prices are dropping slowly on both sides of the aisle, which suggests Nvidia/AMD are still feeling out the market's and its customers' tolerances. A few examples of the latest stories from Videocardz on it (4080 down to $1,000 --- RX7600 and RTX 4060ti/4060 prices falling --- 4060 price falling), and I recall reading how the 7900 series and the 4090 have fallen in price here and there. To me, this suggests that the discrete GPU market is mostly not "broken," as it is adjusting to market realities, and that neither vendor is willing to allow it to wither on the vine too much.

But, short of real competition on price and across all the premier gaming features (e.g., ray tracing, upscaling and frame generation, driver quality, etc.), I do not believe we'll see a return to the generational price/performance gains of yore. Alas!

 

[Bondrewd]

I do not believe we'll see a return to the generational price/performance gains of yore

Those were driven by shrinks anyway.
Party's over.

 

[Father_Murphy]

Those were driven by shrinks anyway.
Party's over.

That's fair enough. I should have said, "Short of real competition on price and across all the premier gaming features (e.g., ray tracing, upscaling and frame generation, driver quality, etc.), I do not believe we'll get better price/performance gains than we're getting these days. Alas!" And, now, having said this twice, I realize it was probably so obvious a statement that it needn't have been said at all....

 

[DavidGraham]

According to The Next Platform, Microsoft is building a 25,000 NVIDIA GPU cluster to train GPT-5.

Now, let’s take a stab at what the rumored 25,000 GPU cluster that Microsoft is building for OpenAI to train GPT-5. Historically, as Nidhi Chappell, general manager of Azure HPC and AI at Microsoft, explained to us back in March, Azure uses PCI-Express versions of Nvidia accelerators to build its HPC and AI clusters, and it uses InfiniBand networking to link them together. We assume this rumored cluster uses Nvidia H100 PCI-Express cards

Lining Up The “El Capitan” Supercomputer Against The AI Upstarts - The Next Platform

The question is no longer whether or not the “El Capitan” supercomputer that has been in the process of being installed at Lawrence Livermore National Laboratory for the past week – with photographic evidence to prove it – will be the most powerful system in the world. The question is how long...

www.nextplatform.com

 

[Father_Murphy]

An interesting video about Nvidia research into ray tracing simulation via light waves rather than, well, "rays."

Two Minute Papers

 

[SlmDnk]

Why not link to the actual paper instead?

A Generalized Ray Formulation For Wave-Optics Rendering

ssteinberg.xyz

https://www.youtube.com/@ShlomiSteinberg/videos

 

[Deleted member 2197]

Why not link to the actual paper instead?

A Generalized Ray Formulation For Wave-Optics Rendering

ssteinberg.xyz

https://www.youtube.com/@ShlomiSteinberg/videos

The Two Minute videos are extremely knowledgeable and provide just the right amount of information without having to immerse completely on a topic.
Corresponding research links are normally provided with their videos.

 

[DavidGraham]

NVIDIA enjoys a 94% market share in the DIY GPU space, in South Korea for the 1H2023.

Now coming to the GPU side of things, here NVIDIA leads as the most popular brand with a 94% market share hold within the DIY segment followed by a 5% share for AMD Radeon GPUs and just a 1.0% market share for Intel Arc GPUs. Amongst the NVIDIA GPUs, the GeForce GTX 16 series has a 15% market share, the GeForce RTX 40 series has a 22.12% market share and the RTX 30 series has a 62.73% market share.

NVIDIA Dominates South Korean DIY Market With 94% GPU Market Share, AMD & Intel Fighting For CPU Market Share

NVIDIA is the most popular GPU brand within the South Korean DIY market while AMD & Intel are fighting for CPU market share.

wccftech.com

 

[Deleted member 2197]

Placed here since the ARMS Execution thread is closed.

Report: NVIDIA in Talks to Become Arm Anchor Investor, Intel May Join in - High-Performance Computing News Analysis | insideHPC

UK chip design company Arm is in negotiations with NVIDIA to be an anchor investor in Arm’s initial public offering, The Financial Times reported last week. The news comes nearly 18 months after NVIDIA ended its attempted acquisition of Arm from Japanese investment company SoftBank due to...

insidehpc.com

UK chip design company Arm is in negotiations with NVIDIA to be an anchor investor in Arm’s initial public offering, The Financial Times reported last week. The news comes nearly 18 months after NVIDIA ended its attempted acquisition of Arm from Japanese investment company SoftBank due to regulatory hurdles in several countries and Europe.

The FT story includes speculation that Intel may also become an Arm anchor investor, with Apple and Microsoft potentially become involved.

According to the story in the FT, NVIDIA wants a share price that would place the value of Arm at about $40 billion, whereas Arm seeks a valuation of about twice that amount. NVIDIA’s total investment is likely to be “in the low hundreds of millions of dollars.”

According to another story in The Guardian, “Anchor investors typically buy shares at an agreed price before an IPO begins, to enhance the stability of an offering and reassure potential investors.”

 

[Deleted member 2197]

Nvidia Is Dominant in AI Chips. Forget About the Competition, Citi Says. | Barron's (barrons.com)
July 17, 2023

The analysts are backing the stock despite some risks, including tighter U.S. restrictions on exports to China and future competition.

Nvidia has emerged as the unquestioned leader in AI chips due to its specialism in graphics-processing units, or GPUs, but rivals such as Advanced Micro Devices AMD +2.05% (AMD) are also aiming to take advantage of a multibillion-dollar opportunity in providing chips to customers looking for potential alternatives.

Nvidia’s dominance is set to continue based on the advantage it has in optimizing software for GPU computations, its networking portfolio and superior hardware, according to Citi.

The Citi analysts said they expect Nvidia to maintain a substantial advantage over AMD in AI performance and keep a market share of around 90% in the market for AI-related graphics-processing units for the next two-to-three years, which they expect to hit $150 billion in 2027.

 

[Kaotik]

NVIDIAs GeForce RTX 4060 Ti 16GB review embargo ended today. There's no reviews because no-one sent review samples. No-one knows when they'll be actually available for purchase either. Only thing known is that there won't be FE model at all.

NVIDIA GeForce RTX 4060 Ti 16GB Review—Not

NVIDIA today launched the 16 GB version of the GeForce RTX 4060 Ti "Ada" graphics card to no fanfare. Essentially a memory variant of the RTX 4060 Ti with no other changes in the specs sheet, the RTX 4060 Ti 16 GB commands a $100 premium (25%) over the regular RTX 4060 Ti for double the memory...

www.techpowerup.com

 

[trinibwoy]

What a pointless card.

 

[DegustatoR]

I'd wager that it would actually be the card for those who want to buy one and just use it for like a dozen of years.

 

[arandomguy]

It's going to be in demand for content creation for those who prefer (if not outright require) Nvidia.

 

[Deleted member 2197]

Nvidia looks to bolster AI earnings with stake in Lambda Labs - KitGuru

Lambda Labs is a smaller cloud company with plenty of ambition, looking to carve out a space in the

www.kitguru.net

Lambda Labs is a smaller cloud company with plenty of ambition, looking to carve out a space in the market with AI training services and powerful data centres. Back in March following an investment round that generated over $44 million, it was announced that Lambda would be adding thousands of Nvidia H100 GPUs to its servers. Now a few months later, it appears that an even bigger deal is in the works between Nvidia and Lambda.

Today, The Information reported that Nvidia and Lambda Labs are nearing an equity deal. The deal would bring in around $300 million in capital for Lambda and would value the company at $1 billion. Nvidia also bought a similar stake in another cloud company, CoreWeave. Naturally, Nvidia has been supplying GPUs to both. While not officially confirmed by Nvidia or Lambda Labs, the report cites sources familiar with “knowledge of the situation”.

 

[Opeth]

Brilliant

RTX 4060 Ti 16GB Marginally Slower than 8GB Model: MSI Testing

MSI Insider Weekly tested an NVIDIA GeForce RTX 4060 Ti 16 GB graphics card as part of a Live Benchmarking event, where it is shown to be in fact slower than the RTX 4060 Ti 8 GB. This begins to explain why NVIDIA or its board partners weren't too keen on sampling the RTX 4060 Ti 16 GB for...

www.techpowerup.com

 

[Father_Murphy]

The whole thing makes me think that Nvidia released the card really for only two reasons: (1) to quiet the critics of the 8GB version and show that this level of card does not warrant 16GB in the vast majority of cases, and (2) so Nvidia will have it in the market for comparison purposes next generation. I suppose it also has its edge case uses in creator workloads, but the cost/benefit/performance advantages over the 12GB 4070/4070ti don't seem compelling.