Wednesday, September 14th 2022
NVIDIA RTX 4080 12GB and 16GB Based on Different Chips, Vastly Different Shader Counts
When we first got news about NVIDIA's upcoming GeForce RTX 4080 "Ada" coming in 12 GB and 16 GB variants, we knew there was more setting the two apart than just memory size and memory bus-width. Turns out there's a lot more. According to detailed specifications leaked to the web, while the 16 GB variant of the RTX 4080 is based on the same AD103, the second largest chip after the AD102; the 12 GB RTX 4080 is based on the smaller AD104 chip which has a physically narrower memory bus.
It looks like NVIDIA is debuting the RTX 40-series with at least three models—RTX 4090 24 GB, RTX 4080 16 GB, and RTX 4080 12 GB. The RTX 4090 is the top-dog part, with the ASIC code "AD102-300-xx." It's endowed with 16,384 CUDA cores, a boost frequency of up to 2.52 GHz, 24 GB of 21 Gbps GDDR6X memory, and a typical graphics power (TGP) of 450 W, which is "configurable" up to 600 W. The RTX 4080 16 GB is based on the AD103-300-xx" comes with 9,728 CUDA cores, a boost frequency of 2.50 GHz, and 16 GB of 23 Gbps GDDR6X memory across a narrower memory bus than the one the RTX 4090 comes with. This card reportedly has a 340 W TGP configurable up to 516 W.
The GeForce RTX 4090 12 GB is positioned a notch below its 16 GB namesake, but is based on the smaller AD104 chip, with 7,680 CUDA cores running at speeds of up to 2.61 GHz, 12 GB of 21 Gbps GDDR6X memory, and a TGP of 285 W that's configurable up to 366 W. It's interesting how the leak includes not just TGP, but also maximum configurable TGP. The various board partners will utilize the latter as their power limits to achieve overclocked speeds. Even the NVIDIA Founders Edition board is technically "custom design," and so it could feature higher-than-stock TGP.
Source:
VideoCardz
It looks like NVIDIA is debuting the RTX 40-series with at least three models—RTX 4090 24 GB, RTX 4080 16 GB, and RTX 4080 12 GB. The RTX 4090 is the top-dog part, with the ASIC code "AD102-300-xx." It's endowed with 16,384 CUDA cores, a boost frequency of up to 2.52 GHz, 24 GB of 21 Gbps GDDR6X memory, and a typical graphics power (TGP) of 450 W, which is "configurable" up to 600 W. The RTX 4080 16 GB is based on the AD103-300-xx" comes with 9,728 CUDA cores, a boost frequency of 2.50 GHz, and 16 GB of 23 Gbps GDDR6X memory across a narrower memory bus than the one the RTX 4090 comes with. This card reportedly has a 340 W TGP configurable up to 516 W.
66 Comments on NVIDIA RTX 4080 12GB and 16GB Based on Different Chips, Vastly Different Shader Counts
Never been a huge fan of two different skus carrying the same name though.
RTX 3080 used the top 102 chip, if the 4080 12GB uses AD104 instead of 103 or 102, you are actually paying the same money for 2 tiers lower performance in the stack compared to last gen.
The 600, 700, 1000, 2000 non ti 80 series cards all used the 104 die. Ampere was the first time since the GTX 580 that we got above that for a 80 non ti tiered card.
Back in the day, these "4080"'s would have been a 4070. Now they want to charge more without saying they've once again upped the price on the tiers, so they're disguising it by calling it a 4080. Far from "who cares if it performs," I'd care because they're charging you more for the performance tier they would have sold you in prior generations for less because it would have been designated a lower tier. This way, they get to reduce tier relative performance because if THIS is considered a 4080 then imagine the 4070 below it. Or the 4060 below that.
They are already doing too many SKU's but to then also have different versions of cards with the same name.
And yes I know both AMD and Nvidia have done this in the past (which I hated then as well), this needs to stop honestly.
Reviewers now have a lot more work telling people how each version of teh same gpu performs....
Don't like this personally and they're is likely to be a performance disparity between the 12/16GB parts, how could there not be?!.
So from 3070 successor ($499) we went to 3080 successor ($699) skipping 3070Ti successor ($599) price level entirely.
So I don't know what pricing level this rumor suggests, maybe something like the below scenario:
$999-$799 16GB cut-down AD103 based
$799-$649 12GB Full AD104 based
$649-$499 10GB cut-down AD104 based
If true, pricing seems to be getting worst by the day!
In less than one week, we will probably have the real deal from Nvidia themselves at GTC.
I don't know why people buy them at this price. £250 maximum is a sensible amount to spend on a GPU in mh opinion.
If, however, you do the smallest amount of due diligence and read a review before buying, then you know exactly what performance you're buying. Internal organization of the GPU is pretty much irrelevant to the average buyer (and even for some enthusiasts).
I'm not a fan of using the same moniker for essentially different GPUs, but that's not the end of the world.
End-all of your wallet. :laugh: