AD104 is 75% the size of GA104. The RTX4080 12GB (AD104) MSRP is $900.
Last edited:
- Status
Well, it's not every generation. It's only a relatively recent trend that it's so expensive and we're going to have to get used to it. Not saying it makes the $900 4080 12GB OK, but it's entirely true that the N5-derived TSMC 4N that Ada uses is far more expensive than Sammy's 8N. Of course, Nvidia is also fleecing us just because they can. They know AMD isn't going to drag them into a pricing war when they have to prioritize their precious N5 wafers for the more lucrative CPU market.
I don't understand your chain of logic. I said costs don't scale at <1x any more (when they used to scale at 0.5x every 2 years), and you conclude that that means costs are scaling at 10x every 2 years? It's not *that* bad! Otherwise the entire industry would have shut down.
Take a look at this chart. A truckload of caveats apply -- it's a rough academic projection, actual costs are different from foundry-to-foundry and depend on vendor-specific volume contracts, which are closely guarded. Also the costs change as a tech node matures and yields improve. But it shows the general trend.
Source: https://www.eetimes.com/fd-soi-benefits-rise-at-14nm/
You can see this has been happening for a while, and if you ignore instantaneous noise from demand spikes you can see how this has been reflected in ASPs over the past few generations that have scaled faster than inflation.
Last edited:
With the smallest being 2.5 slot (which essentially means 3) I wonder why bother even release fan cards.
Anyone getting one would need an upper end case anyway, may as well just come with water block. Probably around same price as the huge coolers.
Make fan aftermarket customizations
Anyone getting one would need an upper end case anyway, may as well just come with water block. Probably around same price as the huge coolers.
Make fan aftermarket customizations
Subtlesnake
Regular
AD104 is likely priced artificially high to help clear out the 3000 stock. But it also isn't going to be able get down to 3060 Ti levels when the 4N process costs 2x+ more per wafer.
With the smallest being 2.5 slot (which essentially means 3) I wonder why bother even release fan cards.
Anyone getting one would need an upper end case anyway, may as well just come with water block. Probably around same price as the huge coolers.
Make fan aftermarket customizations
Honestly with the size and power of gpus, cpus, I think the pc case, motherboard needs a re-design for better thermal performance.
Anyway, 2x cost/mm² and 75% of the mm² would be a 1.5x increase in for the silicon. GDDR6X is also more expensive and adds board complexity though I don't know by how much.
Edit - 3060Ti/3070 is GA104, 3060 is GA106
The point is, since the GTX280 it has been established that cards ending in 80 are high end cards. Even casual gamers know this, and NVIDIA knows they know it. By calling AD104 "RTX4080" they are trying to fool people into thinking it is high end. And it will work, like it always does.
Last edited:
New nodes have always been more complex, so I believe this is indicative of some larger societal problem. I'm not sure exactly what the problem is, but I belive it is mostly not a technical problem. Maybe we aren't producing enough people capable of semiconductor design and manufacturing or something.
Subtlesnake
Regular
The 3070 and 3060 Ti both use GA104. The 3060 uses GA106. AFAIK GA103 was only ever used in laptops.
NVIDIA GeForce RTX 3060 Ti Specs
NVIDIA GA104, 1665 MHz, 4864 Cores, 152 TMUs, 80 ROPs, 8192 MB GDDR6, 1750 MHz, 256 bit
NVIDIA GeForce RTX 3060 12 GB Specs
NVIDIA GA106, 1777 MHz, 3584 Cores, 112 TMUs, 48 ROPs, 12288 MB GDDR6, 1875 MHz, 192 bit
So if you're really comparing with the vanilla 3060, then AD104 is 6.9% bigger. So it would be a 2.14X increase in cost, not a 1.5X increase.
We are, but fighting against Mother Nature is getting exponentially more difficult. So we need more people and more time all the way up and down the chain starting from ASML and ending with board partners. To a first order of approximation the increased cost is just the summation of people*time needed to fight this battle.
That would actually have been consistent with Nvidia's approach in the most recent prior generations. But these decisions are locked-in years in advance. My guess is the 4N decision was made when the GPU demand across all sectors (datacenter, gaming and crypto) was going gangbusters, AMD was catching up in efficiency and Nvidia did not want to take the risk of losing the top-end by choosing an inferior node.
I have no idea how affordable N6 is relative to N5, but it's still probably nowhere close to Samsung 8N. I actually wonder if fabbing the lower-tier Ada chips on Samsung 8N would have been interesting. But again, that decision would have to have been made 2-3 years back. And maybe those tiers will continue to be serviced by Ampere, perhaps with some price cuts and/or rebadging (which used to be fairly common).
They shouldn't have launched the 4080 12GB.
4090 at 1600 was fine (it is a whale card, nobody cares).
4080 16GB over 1K feelsbadman. Should have been $999 max.
4080 12 GB - Don't launch it. Hold until you clear inventory, launch at $749 - $799, when inventory clears. And don't call it a 4080 when the chip is not remotely close to the real one. 1 SM disabled is understandable.... thousands of cc not so much. Honestly, the naming of the 4080 12GB is so moronic and out of touch, it indicates a decision that was made from high up the ladder by somebody who is a real asshole. Only way it doesn't get contested and rejected....
And if they can't manage that financially, then just the 4090 only until inventory clears.
All they accomplished was making their inventory situation worse by generating a bunch of ill will followed up with more damage via tone deaf statements.
4090 at 1600 was fine (it is a whale card, nobody cares).
4080 16GB over 1K feelsbadman. Should have been $999 max.
4080 12 GB - Don't launch it. Hold until you clear inventory, launch at $749 - $799, when inventory clears. And don't call it a 4080 when the chip is not remotely close to the real one. 1 SM disabled is understandable.... thousands of cc not so much. Honestly, the naming of the 4080 12GB is so moronic and out of touch, it indicates a decision that was made from high up the ladder by somebody who is a real asshole. Only way it doesn't get contested and rejected....
And if they can't manage that financially, then just the 4090 only until inventory clears.
All they accomplished was making their inventory situation worse by generating a bunch of ill will followed up with more damage via tone deaf statements.
Last edited:
These are all fair points. I don't particularly care about the brand names but given the specs then I would have guessed the 3 cards to be priced at 1999+, 999 and 699 respectively, i.e., a more or less consistent increase across all tiers. Instead, they shuffled their margins around, possibly to try and help with some of these inventory challenges.
They will have to adjust.
Yes, I was mistaken. Hopefully I've edited my posts with the correct information. The numbers still add up to a midrange card that costs almost a thousand bucks. What a disgrace.The 3070 and 3060 Ti both use GA104. The 3060 uses GA106. AFAIK GA103 was only ever used in laptops.
NVIDIA GeForce RTX 3060 Ti Specs
NVIDIA GA104, 1665 MHz, 4864 Cores, 152 TMUs, 80 ROPs, 8192 MB GDDR6, 1750 MHz, 256 bitwww.techpowerup.com
NVIDIA GeForce RTX 3060 12 GB Specs
NVIDIA GA106, 1777 MHz, 3584 Cores, 112 TMUs, 48 ROPs, 12288 MB GDDR6, 1875 MHz, 192 bit
So if you're really comparing with the vanilla 3060, then AD104 is 6.9% bigger. So it would be a 2.14X increase in cost, not a 1.5X increase.
Last edited:
Everyone seems to have issues with the 4080 12 
www.tomshardware.com
Why Nvidia's RTX 4080, 4090 Cost so Damn Much
And how AMD's RDNA3 chiplets could seriously undercut Nvidia
www.tomshardware.com
Partially because Android is still playing catch-up with A14. Apple has zero incentive to use the latest and greatest on their normal lineup. And yearly release cycle means it won’t take too long to update as well.
- Status
Similar threads
