Nvidia Reportedly in No Rush to Boost RTX 40-Series Output
Nvidia is not scaling up output of Ada Lovelace GPUs.
Nvidia's backend service providers are reportedly not seeing the increased orders for testing and packaging of its latest GeForce RTX 40-series GPUs based on the Ada Lovelace architecture, says DigiTimes. They believe that the company is taking its time and letting its partners sell off any remaining GeForce RTX 30-series inventory before supplying them the latest generation graphics chips.
Nvidia's OSAT providers, Siliconware Precision Industries (SPIL) and King Yuan Electronics (KYEC), have not received any indication from the company about scaling up production for the GeForce RTX 40-series despite the fact that Nvidia has introduced virtually all of its Ada Lovelace GPUs for client applications, including AD102, AD103, AD104, AD106, and AD107, the report says.
To aggressively ramp up output of new graphics processors, Nvidia has to produce them at TSMC and then test and assemble them at SPIL or KYEC. While the company pre-paid TSMC for its silicon manufacturing services in the 2021–2022 timeframe and probably has N4-based Ada Lovelace wafers in stock, it still needs to book OSAT capacity before ramping up orders to increase supply. However, it looks like Nvidia is satisfied with the current GeForce RTX 40-series output and has not shown any intention to increase it.
Normally, sales of GPUs from Nvidia to its partners trend up in the third quarter, so the company should be booking OSAT services now. If the report's information is correct, it looks like Nvidia prefers to wait a bit before further ramping up testing and packaging of its Ada Lovelace silicon.
The company's partners are already shipping GeForce RTX 4070, RTX 4070 Ti, RTX 4080, and RTX 4090 graphics cards for desktops. While these rank among the best graphics cards, demand apparently isn't as high as Nvidia would like. There are also dozens of laptops employing GeForce RTX 4050, RTX 4060, RTX 4070, RTX 4080 and RTX 4090 Laptop GPUs. Nvidia's Ada Lovelace graphics processors introduced so far cover virtually all market segments, and solutions like the recently launched GeForce RTX 4070 and the expected RTX 4060 should become rather popular. That means Nvidia will need more GPU volume and packaging services.
Perhaps Nvidia and its partners already have enough RTX 30- and 40-series GPUs to sell, so there's no need to further ramp up testing and assembly even ahead of alleged GeForce RTX 4050 and RTX 4060 GPU launch in May–June. Certainly it hasn't been trying to drop prices into the sub-$500 space yet. While testing and assembly procedures take time, they're not as lengthy as the production of wafers, so even if Nvidia is behind its normal schedule by several weeks with ordering OSAT services, given the globally slow demand for new PCs, this is probably not going to create any tangible problems for the market.
We've seen plenty of reports of the severe downturn in PC and laptop sales, and after a couple of incredible years (in terms of demand and prices), it looks like many people and businesses are happy to stick with what they have for a while as we deal with economic uncertainties. Still, as this information comes from an unofficial source and lacks many details, take it is rumor rather than fact.
Stay On the Cutting Edge: Get the Tom's Hardware Newsletter
Get Tom's Hardware's best news and in-depth reviews, straight to your inbox.
Anton Shilov is a contributing writer at Tom’s Hardware. Over the past couple of decades, he has covered everything from CPUs and GPUs to supercomputers and from modern process technologies and latest fab tools to high-tech industry trends.
-
bit_user Nvidia's OSAT providers, Siliconware Precision Industries (SPIL) and King Yuan Electronics (KYEC), have not received any indication from the company about scaling up production for the GeForce RTX 40-series
Are these providers also used for their H100 and Grace CPUs? If so, have they said anything about them? I presume Nvidia is happily redirecting its 4N wafer supply to higher-margin datacenter products. -
Eximo BX4096 said:I guess the feeling is mutual. I'm in no rush to buy it.
4070 might be a recommendation as something worth buying for systems with smaller PSUs. The rest of the line-up, not so much.
I'll just keep running the old 3080Ti at 280W until I start running into performance issues or get the itch to swap it out. Maybe 60 series or, fingers crossed, Celestial. -
russell_john A more likely explanation is Nvidia had to take away production time they booked at TSMC for consumer devices and use it instead for their A100 and H100 lines which are in high demand because of the Generative AI explosion that no one saw coming when they booked production time at TSMC over a year ago. Elon Musk/Twitter alone just ordered 10,000 H100 devices which means Nvidia has to make 80,000 of their largest chips and then you have large orders from Microsoft and AWSReply
Everyone is so myopic that they think the tech world revolves around Gamers .... There is only so such time booked at TSMC and Nvidia is going to use it for their most profitable lines and that simply is not low margin consumer markets but their high margin professional lines -
JamesJones44 Eximo said:4070 might be a recommendation as something worth buying for systems with smaller PSUs. The rest of the line-up, not so much.
I'll just keep running the old 3080Ti at 280W until I start running into performance issues or get the itch to swap it out. Maybe 60 series or, fingers crossed, Celestial.
I'm still rocking a 980 Ti. I was going upgrade to a 3000 series, but the whole crypto/pay 3x over MSRP thing turned me off and I decided that I'm just going to wait until the 4000 series comes down in price (which is probably never) or just go with a low end 5000 series when it comes out. -
jaquith If I were Nvidia at this juncture, I would even slow down production to consumers even more. Right now with AI, it is paramount that Nvidia focuses all of their resources for data centers. As a stockholder, I want Nvidia to prioritize profits and by far not even close, it's more profitable in the enterprise side of production.Reply
Frankly, I am so sick and tired of people's belly aching about GPU prices. It's like they forgot how much prices were 10 years ago. The difference in cost between AMD and Nvidia is pennies in comparison. It's like some lunacy has taken over people's thoughts on the subject. Corporations and in particular publicly traded corporations absolutely have to maximize their profits by law. -
bit_user
A100 are made on TSMC N7. None of their gaming GPUs use that node. Only the H100 shares a node with gaming GPUs.russell_john said:A more likely explanation is Nvidia had to take away production time they booked at TSMC for consumer devices and use it instead for their A100 and H100 lines which are in high demand because of the Generative AI explosion
According to this, those were 10k GPUs, not DGX's. So, that's only 10k big dies (after accounting for yield).russell_john said:Elon Musk/Twitter alone just ordered 10,000 H100 devices which means Nvidia has to make 80,000 of their largest chips
https://www.tomshardware.com/news/elon-musk-buys-tens-of-thousands-of-gpus-for-twitter-ai-project
Check their latest quarterly report - gamers still comprise the majority a significant amount of their income. They can't afford to alienate them too much, just yet.jaquith said:If I were Nvidia at this juncture, I would even slow down production to consumers even more.
What if a purpose-built AI chip outperformed Nvidia's best datacenter products? Their datacenter revenue could dry up rather quickly. That's why they need to remain active in all the markets where they play, and not take any of them for granted.jaquith said:it's more profitable in the enterprise side of production.
That's an overstatement. Nobody is going to jail, if a company doesn't screw over consumers by the maximum possible amount. The jeopardy is that they could be sued by shareholders who think the company is not being properly run in a profit-maximizing way. Civil, not criminal. The plaintiffs also have some burden of proof - it can't be merely their opinion.jaquith said:Corporations and in particular publicly traded corporations absolutely have to maximize their profits by law. -
Eximo JamesJones44 said:I'm still rocking a 980 Ti. I was going upgrade to a 3000 series, but the whole crypto/pay 3x over MSRP thing turned me off and I decided that I'm just going to wait until the 4000 series comes down in price (which is probably never) or just go with a low end 5000 series when it comes out.
I had two 980s...then a 1080, and I told myself I would get on the 80Ti next series, then they announced the price for the 2080Ti and I decided to wait. 3080Ti I got for retail, which was still 1400 after taxes, so not ideal, but I had the same GPU for 6 years and I hope to keep this one for at least 5. Hindsight, should have waited for the 4080 or waited for a 3080 12GB when the price was at its bottom. -
bit_user
Same!JamesJones44 said:I'm still rocking a 980 Ti. I was going upgrade to a 3000 series, but the whole crypto/pay 3x over MSRP thing turned me off and I decided that I'm just going to wait until the 4000 series comes down in price