Nvidia are one of the very few conformant hardware vendors in regards to modern D3D graphics API support so it absolutely does make sense why they'd want to partner up with them in comparison to ARM /w Mali or IMG /w PowerVR ...
NVIDIA discussion [2024]
- Thread starter pharma
- Start date
Nvidia doesn't have much interest in producing a PC SoC, the market is too low margins for them. MTK on the other hand is interested, and using Nv's GPU IP in such a SoC is a nice differentiator for them.
This ^^
In fact, the partnership goes even beyond that. Mediatek is also working on a mobile SoC for handled gaming with a NV GPU chiplet as Jensen is frustrated by Nintendo and he sees a good potential in this market.
PS: and it's different than the new Nvidia semi custom division that tries to sell ARM SoCs to console vendors (with already interested customers in China)
cheapchips
Veteran
Source for either of those things?
What's Jensen's frustration actually supposed to be? The Switch is likely going to end up as the best selling console of all time and all he did was flog Nintendo spare SoCs he had gathering dust in warehouse.
Source is within NVDA like all the stuff I share for over a decadeSource for either of those things?
What's Jensen's frustration actually supposed to be? The Switch is likely going to end up as the best selling console of all time and all he did was flog Nintendo spare SoCs he had gathering dust in warehouse.
Frustration is Nintendo choosing a tech multiple gen behind the latest and greatest. It slows down new NV tech adoption, thus advantage. I have a story to share about the first switch deal for which Jensen travelled to Japan. He offered a custom SoC updated with Pascal for less than $1 extra aver the TegraX but Nintendo didn't care. And now for Switch2, it's the same story. Could be Lovelace but nah, Nintendo doesn't care again. Even worst, Switch will be launched more than 2 years after the SoC design completion
cheapchips
Veteran
Hehe. He was wrong though. It wouldn't have made any meaningful difference to sales and Nintendo have netted an extra $141m so far.
There's cheap and then there's Nintendo cheap.
Yeah I completely get why Nvidia would want to showcase more of their tech but it has to align with Nintendo’s goals and the types of games they want to make. Making plans based on a 2 year old SoC is also a lot safer than waiting for Jensen to build some new thing that only exists on PowerPoint.
Switch 2 launching with a 10nm based SoC with Ampere GPU in 2025 will be quite a bit more than "2 years old SoC" really.
When was the hardware config locked down though? Latest rumors on launch date indicate it’s being driven by software/game dev timelines.
in 2025, nearly 3 years prior to launch time
Edit: Nvidia has been waiting for roughly a year before starting to produce the damn thing
Would they want to?
Wouldn´t they ?
What would be the point? Get the struggling factories which are eating all of Intel profits right now and the CPU business which can't compete? They certainly are not interested in Intel's GPUs or the AI stuff. So why would they want to buy Intel?
Some people seem to think that every company on the planet wants to compete in the x86 market for some reason while we've seen companies leaving this space or struggling in it for the last 20 years suggesting that this market isn't really interesting to pursue as an expansion of your business. In other words anyone suggesting things like Nvidia suddenly buying Intel or AMD don't really get the current dynamics in these markets and is just basically feeding into the baseless dreams of various fanboys.
Also Nvidia tried to buy Arm (which was a much much better fit for them really) and that didn't end up well so why would buying Intel be any different?
They wouldn't want to buy the foundry, i'd imagine
Do you know if they were waiting on improved node at costs they are happy with to make it viable as a handheld? Rumour has it that the Switch successor is potentially using 12GB LPDDR5X memory which is a step up over what Jetson AGX Orin uses. If they can use a new node like Samsung 4LPP+ with FoWLP technology, it would make a huge difference in terms of efficiency and clocks as well as die size for mass production.
There's no appreciable cost savings to be gained either way from using the latest process technologies. It only opens up the possibility for the architects to design more complex HW for smaller die sizes ... (the trade off between logic complexity/cost are similar as before to the previous process nodes)
I guess the costs of producing a 10nm SoC are lower now than they were a couple of years ago. So if they were waiting on costs to decrease then this is the only way it may have happened.
The performance/power advantages of Pascal and 16nm would have given them significantly better performance/battery life and certainly would have made a lot of sense for a handheld device. Especially if the cost difference was marginal. But yeah it is Nintendo after all.
Wouldn't it be 8nm? It is a 10nm class process of course, closer to 10nm than 7nm.
The fab equipment (most likely installed pre 2020) would almost completely be depreciated by now and hence costs should be significantly cheaper indeed.
