Nvidia might use TSMC's 5nm process node for upcoming GeForce RTX 4000 GPUs

nanoguy

Posts: 1,355   +27
Staff member
Rumor mill: Nvidia’s Ada Lovelace GPUs are still a pretty well-guarded secret, but industry insiders believe the company will rely on TSMC’s N5 node for manufacturing them. These are expected to be monstrous GPUs with 50 percent better gaming performance and higher power consumption than RTX 3000 series GPUs, so moving away from Samsung’s 8 nm node might help on both fronts.

Nvidia’s RTX 4000 graphics cards could break cover as early as next summer, and it’s also possible that we’ll see an RTX 3000 series refresh before that. The next generation of cards from Team Green carry the Ada Lovelace codename internally, but not much is known about the new architecture and what it will bring to the table. The only consensus among industry insiders seems to be that raw performance will double over the current generation, while power consumption will be north of the 400W range.

It’s been rumored that Nvidia will use TSMC’s 5 nm process node, which is believed to be significantly more expensive than its 7 nm process node that AMD uses for its RX 6000 series GPUs. A new report from Digitimes seems to reinforce these expectations, citing industry sources from the supply chain.

The publication also notes Nvidia is planning to leverage TSMC’s chip-on-wafer-on-substrate (CoWoS) packaging technology for its upcoming H100 GPU. This model will be based on Nvidia’s compute-oriented Hopper architecture and is expected to be the company’s first multi-chip GPU module. Ada Lovelace GPUs will most likely use FC-BGA packaging courtesy of ASE Technology, as well as traditional GDDR6X memory.

RTX 3000 series GPUs are manufactured using Samsung’s relatively mature 8 nm process node, so a move to TSMC’s 5 nm node is a risky one as it’s also what other companies are going to use for their upcoming products. Apple currently makes its A14, M1 Pro, and M1 Max chipsets on that same node, while AMD is planning to use it for its Zen 4 CPUs. At the same time, Bitmain has secured some of TSMC’s 5 nm production capacity for mining hardware that will be unveiled in Q1 2022.

It’s possible Nvidia has chosen to go with TSMC for both Hopper and Ada Lovelace, which may have allowed it to secure more capacity. Naturally, the company won’t comment on the news, so we’ll have to wait and see.

Permalink to story.

 
The fact is we won’t ever really know how much capacity any of these companies are getting and for how much. But I’d say it’s safe to say that the bigger clients will do better out of TSMC as they will give them more money.

If only we had more TSMCs in the world!
 
I'm excited.

My 3070 runs everything but max ray tracing well on my ultrawide and my 3080 handles it's 4K living room TV duties with aplomb. But a process node shrink could bring some big advancements. I suppose it depends on if my game of choice next year is controller based or mouse and keyboard. And if it's ray tracing enabled or not.
 
The fact is we won’t ever really know how much capacity any of these companies are getting and for how much. But I’d say it’s safe to say that the bigger clients will do better out of TSMC as they will give them more money.

If only we had more TSMCs in the world!

I always thought TSMC was China's real prize. Taiwan has many significant industries, but none stand out as being indispensable or hard to quickly replace in the world's trade. Bar one. China would have the world by the balls if they take and hold the island.

Of course this has been pointed out before and more so recently. So new plants are being built elsewhere. Fab 21 in the USA. But nothing yet compares to the concentration of manufacturing capacity of the very latest technologies on Taiwan.
 
I always thought TSMC was China's real prize. Taiwan has many significant industries, but none stand out as being indispensable or hard to quickly replace in the world's trade. Bar one. China would have the world by the balls if they take and hold the island.

Of course this has been pointed out before and more so recently. So new plants are being built elsewhere. Fab 21 in the USA. But nothing yet compares to the concentration of manufacturing capacity of the very latest technologies on Taiwan.
Well, many speculated with sleepy Joe at the helm of the USA that the Chinese would make a move. They just effectively took Hong Kong with zero resistance.

But I think the new AUKUS alliance has set the Chinese back a bit and hopefully by the time they could be ready to make a move we will have the new fabs planned for the USA and other parts of the world.
 
My enthusiasm is curbed. Not just because I expect a reasonable purchase opportunity to be much further away than 2H22, but also because I think many fundamental game design decisions will still be anchored around the consoles, meaning that even if the extra performance was obtainable it would mostly go to diminishing returns levels of increased frames and post-processing.

I wonder if they will retain their 7nm process - perhaps this will be the start of a period where we see the past generation remain in full production so they can get the volume from both lines.
 
Well, many speculated with sleepy Joe at the helm of the USA that the Chinese would make a move. They just effectively took Hong Kong with zero resistance.


Tjhey locked-down Hong Kong mostly because they are tired of al the illegal smuggling...but it's still going to take them years to close the huge gaps.

https://en.wikipedia.org/wiki/Hong_Kong–Mainland_China_conflict#Porous_borders_and_smuggling

Taiwan is not anywhere near mainland China, so the effort to take the island would be similar to shipping/supporting the entire D-Day campaign 10x the distance. And they can't really area bomb the island, if TSMC is your ultimate prize. (if it's damaged by fire or bombs, then it will take your months/years to source the equipment + expertise to fix it)

A fab isn't designed to be run in times of war, because any damage takes forever to fix - this is part of the reason why most people acknowledge that China's threats against Taiwan are mostly empty (better to steal the ip through spies/hires , and jump-start SMIC)

But they also need to look powerful to the rest of the world, so the constantly try to come up with exciting new ways mess with people's heads.
 
Last edited:
Tjhey locked-down Hong Kong mostly because they are tired of al the illegal smuggling...but it's still going to take them years to close the huge gaps.

https://en.wikipedia.org/wiki/Hong_Kong–Mainland_China_conflict#Porous_borders_and_smuggling

Taiwan is not anywhere near mainland China, so the effort to take the island would be similar to shipping/supporting the entire D-Day campaign 10x the distance. And they can't really area bomb the island, if TSMC is your ultimate prize. (if it's damaged by fire or bombs, then it will take your months/years to source the equipment + expertise to fix it)

A fab isn't designed to be run in times of war, because any damage takes forever to fix - this is part of the reason why most people acknowledge that China's threats against Taiwan are mostly empty (better to steal the ip through spies , and jump-start SMIC)
Well, we have had battles in the last 100 years between Chinese and American forces over Taiwan. (Google Taiwan strait crisis) And that was before China became a super power. You might be blasé about China never taking Taiwan but personally if I lived in Taiwan I’d want a bit more reassurance than that. And that’s why there are US and now AUKUS forces operating in the area on a permanent basis.

Also, claiming the Chinese only took Hong Kong because of smuggling is laughable. The Chinese have clearly lamented the fact that Hong Kong didn’t belong to them for over a century. I have a lot of friends from Hong Kong, many have now fled to the U.K. and other countries in the west. They would be insulted at your description of events. Overnight millions of people lost a lot of the rights they previously had. Now they could be thrown in jail for speaking out against the Chinese government. Or even tweeting a picture of Winnie the Pooh..
 
Last edited:
You do realise Hong Kong has ALWAYS been Chinese don't you.
Lmao, your attempt to troll is poor. In case you aren’t trolling, read this:


Up until very recently Hong Kong residents could not be tried by Chinese Law, this has prompted a mass exodus as Chinese law is a long way away from the common law deployed in British colonies. Nobody in their right mind would choose to live under Chinese law than British common law…
 
Price projections:

60 Tier - $1k+ (us)

70 Tier - 1500-2k+

80 Tier 2500-3k+

90 - 4k+

And we're not talking scalper prices, we're talking sold by Newegg prices in the shuffle. Conservative estimates on top of that.
 
All these new processes with so little capacity (backlog) to actually manufacture the chips. It's a scalpers paradise until at least 2023.
 
GPU cooling may be more important that CPU cooling going forward. Those things run hot
They are indeed hot, but only at one end of the spectrum

Bus powered (under 75 watts) would still be a massive improvement over internal graphics at the other end of the performance spectrum

We could even start seeing m.2 slot graphics cards in small form factor budget builds with an external power connector before 3nm even arrives (only 7 watts available on m.2)
 
Last edited:
I'm excited.

My 3070 runs everything but max ray tracing well on my ultrawide and my 3080 handles it's 4K living room TV duties with aplomb. But a process node shrink could bring some big advancements. I suppose it depends on if my game of choice next year is controller based or mouse and keyboard. And if it's ray tracing enabled or not.
If this is true, then it will be a significant node improvement. The current Samsung 8nm is basically a 10nm, which is probably just slightly better than TSMC's 12nm (14nm) that it replaces. Having said that, I feel AMD's upcoming chiplet RDNA3 looks more interesting from my perspective. I don't know about performance at this point, but I am more excited to see RDNA3 ' specs and in action to determine how well is first gen chiplet GPU, as oppose to the existing monolithic design.
 
If this is true, then it will be a significant node improvement. The current Samsung 8nm is basically a 10nm, which is probably just slightly better than TSMC's 12nm (14nm) that it replaces. Having said that, I feel AMD's upcoming chiplet RDNA3 looks more interesting from my perspective. I don't know about performance at this point, but I am more excited to see RDNA3 ' specs and in action to determine how well is first gen chiplet GPU, as oppose to the existing monolithic design.
I think the middling Samsung process results make Nvidia's switch to TSMC likely. No sense fighting the competition with one arm tied behind your back even if the manufacturing costs are higher.

Rumors say that Nvidia's 4000 GPUs are going to be even more power-hungry which suggests that AMD's next GPU is going to be quite the contender otherwise Nvidia would lower, or at least keep the same, power levels on a node shrink.

Taken together, the next GPUs from both companies should be a good leap forward. Which is good because prices will likely still be high a year from now so at least we'll get more bang for our many bucks.
 
I always thought TSMC was China's real prize.
I think it's arguable, in the event of an invasion the TSMC fabs in Taiwan would almost certainly lose access to the extreme ultraviolet lithography machines made by ASML Holding N.V. in the Netherlands, thus a Chinese-controlled TSMC would not be able to expand and it would likely lose its technological edge in some years. The real advantage in getting Taiwan is ideological and possibly strategical (getting Taiwan should help them breaking through the first island chain).
 
I think it's arguable, in the event of an invasion the TSMC fabs in Taiwan would almost certainly lose access to the extreme ultraviolet lithography machines made by ASML Holding N.V. in the Netherlands, thus a Chinese-controlled TSMC would not be able to expand and it would likely lose its technological edge in some years. The real advantage in getting Taiwan is ideological and possibly strategical (getting Taiwan should help them breaking through the first island chain).
No doubt it is ideological, I wasn't suggesting they wanted to invade Taiwan to take TSMC. Merely that in terms of the victor the spoils that would be the most significant industry to the rest of the world that would come under their control.
 
Back