Friday, September 23rd 2022
NVIDIA AD103 and AD104 Chips Powering RTX 4080 Series Detailed
Here's our first look at the "AD103" and "AD104" chips powering the GeForce RTX 4080 16 GB and RTX 4080 12 GB, respectively, thanks to Ryan Smith from Anandtech. These are the second- and third-largest implementations of the GeForce "Ada" graphics architecture, with the "AD102" powering the RTX 4090 being the largest. Both chips are built on the same TSMC 4N (4 nm EUV) silicon fabrication process as the AD102, but are significantly distant from it in specifications. For example, the AD102 has a staggering 80 percent more number-crunching machinery than the AD103, and a 50 percent wider memory interface. The sheer numbers at play here, enable NVIDIA to carve out dozens of SKUs based on the three chips alone, before we're shown the mid-range "AD106" in the future.
The AD103 die measures 378.6 mm², significantly smaller than the 608 mm² of the AD102, and it reflects in a much lower transistor count of 45.9 billion. The chip physically features 80 streaming multiprocessors (SM), which work out to 10,240 CUDA cores, 320 Tensor cores, 80 RT cores, and 320 TMUs. The chip is endowed with a healthy ROP count of 112, and has a 256-bit wide GDDR6X memory interface. The AD104 is smaller still, with a die-size of 294.5 mm², a transistor count of 35.8 billion, 60 SM, 7,680 CUDA cores, 240 Tensor cores, 60 RT cores, 240 TMUs, and 80 ROPs. Ryan Smith says that the RTX 4080 12 GB maxes out the AD104, which means its memory interface is physically just 192-bit wide.
Sources:
Ryan Smith (Twitter), VideoCardz
The AD103 die measures 378.6 mm², significantly smaller than the 608 mm² of the AD102, and it reflects in a much lower transistor count of 45.9 billion. The chip physically features 80 streaming multiprocessors (SM), which work out to 10,240 CUDA cores, 320 Tensor cores, 80 RT cores, and 320 TMUs. The chip is endowed with a healthy ROP count of 112, and has a 256-bit wide GDDR6X memory interface. The AD104 is smaller still, with a die-size of 294.5 mm², a transistor count of 35.8 billion, 60 SM, 7,680 CUDA cores, 240 Tensor cores, 60 RT cores, 240 TMUs, and 80 ROPs. Ryan Smith says that the RTX 4080 12 GB maxes out the AD104, which means its memory interface is physically just 192-bit wide.
152 Comments on NVIDIA AD103 and AD104 Chips Powering RTX 4080 Series Detailed
It is closest to 12 nm Ground Rules:
10 nm process - Wikipedia
You can see that they are very similar. Perhaps both should be called 14 nm and that would be fair as they are closer to Intel's 14 nm.
10 nm lithography process - WikiChip
But if it's not similarly superior to the 3080ti, then I'll consider the 4080 12GB to be a ludicrous proposition.
The new kid on the block here is the fact there is a 103-SKU in between the 102 and 104. And even the 102 is a spin off from Titan's 110.
The fact is we're looking at the exact opposite situation: they can't place a 104 that high in the stack anymore. They need TWO bigger SKUs to cater to their top end of the stack, and 104 is upper midrange but at best - it no longer 'stands in as early high end'. It used to only get succeeded by a 102 later in gen, now they have to place it at the front of the gen to make even a tiny bit of impact compared to the last. ADA's current positioning is the best example: they're not even removing the 3xxx cards below it; we're looking at ONLY 102 dies from gen to gen populating half their new stack for a while. These are all signs Nvidia's running into new territory wrt their wiggle room: we're paying their wiggle room now on the Geforce stack; the 102/103 SKUs are simply new tiers also in terms of price, and they need every single piece of it to survive against the competition.
Back in the HD-days, they could make do with a 104 and then destroy AMD with a 102 either cut down or full later on. Which they have been doing up until they started pushing RT. Ever since, the changes happened, prices soared and VRAM magically went poof. The price of RT... again... ;) We still laughing about this fantastic tech?
This will become worse when an actual RTX 4070 (or not RTX 4060) come out.
Press's opinions will be shared on the naming in their own content pieces.
They actually did this with the previous Ampere generation. The 3080 Ti uses the same GPU die as the 3090 and 3090 Ti.
Most PC hardware reviewers errorneously compared the 3080 Ti to the 3080. The 3080 Ti was essentially a slightly binned 3090 with half the VRAM.
historically, the x80 (which had always been an anemic idiot choice tbh - not to be confused w/ the x80ti) had always been the 104 die. maxwell, pascal turing - you name it.
but those were also the days of the x80ti being like twice as powerful and the gap between the x80 and the x80ti larger than between x60 and x80
'twas not until ampere that nv decided to change their tiering and abolish the x80ti as the halo card.
It's not wise to compare NVIDIA's graphics card generations by comparing model numbers since they aren't consistent with what each model number represents. It's really just a numerical slot relative to a given card's placement within that generation's product stack.
Clearly NVIDIA's current strategy is to use binning to maximize gross margin. They're not going to sell Joe Gamer an $800 graphics card with a GPU that can be overclocked +30%. Those days are over. They're going to keep those binned chips, relabel them, and sell them at a higher price like $1200.
Anyhow, the way NVIDIA binned their GA102 GPUs, the 3080 Ti still ended up much closer to the 3090 (CUDA, RT, Tensor) than the 3080.
The main point is regardless of the given die, NVIDIA is going to bin the foundry's output and differentiate GPUs to maximize gross margin regardless of what final model number they slap on the chip.
One thing that is becoming increasingly evident is that their best silicon is destined for datacenter systems not consumer gaming PCs.
www.techpowerup.com/review/nvidia-geforce-gtx-980/
www.techpowerup.com/review/nvidia-geforce-gtx-1080/
www.techpowerup.com/review/nvidia-geforce-rtx-2080-founders-edition/
The only post-Fermi x80s NOT based on a 104 were the 780 and 3080.
If the product suit your needs and in your budget frame and it's the best price\preformance in it's segments then get it.
Simple as that.
If there is a large difference between SKUs die size wise (and by extension manufacturer cost) that would indicate to the customer that Nvidia is likely to release products to fill that gap or AMD will do it for them. In addition, comparing the die size of the 3080 and 4080 shows you that you are getting less than half the die area. Even accounting for inflation and the cost increases of smaller nodes, it does not even come close to making a die less than 300mm2 in size worth $900 USD, especially when you compare it to last gen products.
I think just about any customer would be mad to know they are getting much less relative to the 4090 compared to the prior GPU generation while also being charged $200 more.
Your criteria for what product to buy is simply far too naive. You advise customers to just blindly buy without considering factors that could net then massive savings or a better end product.