Thursday, December 2nd 2021
NVIDIA GeForce RTX 2060 12GB Has CUDA Core Count Rivaling RTX 2060 SUPER
NVIDIA's surprise launch of the GeForce RTX 2060 12 GB graphics card could stir things up in the 1080p mainstream graphics segment. Apparently, there's more to this card than just a doubling in memory amount. Specifications put out by NVIDIA point to the card featuring 2,176 CUDA cores, compared to 1,920 on the original RTX 2060 (6 GB). 2,176 is the same number of CUDA cores that the RTX 2060 SUPER was endowed with. What sets the two cards apart is the memory configuration.
While the RTX 2060 maxed out the "TU106" silicon, the RTX 2060 12 GB is likely based on the larger "TU104," in order to achieve its CUDA core count. The RTX 2060 SUPER features 8 GB of memory across a 256-bit wide memory bus, however, the RTX 2060 12 GB uses a narrower 192-bit wide bus, disabling 1/4th of the bus width of the "TU104." The memory data-rate on both SKUs is the same—14 Gbps. The segmentation between the two in the area of GPU clock speeds appears negligible. The original RTX 2060 ticks at 1680 MHz boost, while the new RTX 2060 12 GB does 1650 MHz boost. The typical board power is increased to 185 W compared to 160 W of the original RTX 2060, and 175 W of the RTX 2060 SUPER.
Update 15:32 UTC: NVIDIA has updated their website to remove the "Founders Edition" part from their specs page (3rd screenshot below). We confirmed with NVIDIA that there will be no RTX 2060 12 GB Founders Edition, only custom designs by their various board partners.NVIDIA is getting its add-in card partners to come up with several custom-design products based on the new SKU, which should occupy price-points below those of the RTX 3060 "Ampere." This could be an answer to AMD's Radeon RX 6600 (non-XT), which beats the RTX 2060 SUPER by 3% and the original RTX 2060 by 13%, at 1080p, in our testing. Technologically, the older "Turing" architecture won't find itself obsolete in the current market, as it maintains full DirectX 12 Ultimate compatibility.
Source:
VideoCardz
While the RTX 2060 maxed out the "TU106" silicon, the RTX 2060 12 GB is likely based on the larger "TU104," in order to achieve its CUDA core count. The RTX 2060 SUPER features 8 GB of memory across a 256-bit wide memory bus, however, the RTX 2060 12 GB uses a narrower 192-bit wide bus, disabling 1/4th of the bus width of the "TU104." The memory data-rate on both SKUs is the same—14 Gbps. The segmentation between the two in the area of GPU clock speeds appears negligible. The original RTX 2060 ticks at 1680 MHz boost, while the new RTX 2060 12 GB does 1650 MHz boost. The typical board power is increased to 185 W compared to 160 W of the original RTX 2060, and 175 W of the RTX 2060 SUPER.
Update 15:32 UTC: NVIDIA has updated their website to remove the "Founders Edition" part from their specs page (3rd screenshot below). We confirmed with NVIDIA that there will be no RTX 2060 12 GB Founders Edition, only custom designs by their various board partners.NVIDIA is getting its add-in card partners to come up with several custom-design products based on the new SKU, which should occupy price-points below those of the RTX 3060 "Ampere." This could be an answer to AMD's Radeon RX 6600 (non-XT), which beats the RTX 2060 SUPER by 3% and the original RTX 2060 by 13%, at 1080p, in our testing. Technologically, the older "Turing" architecture won't find itself obsolete in the current market, as it maintains full DirectX 12 Ultimate compatibility.
99 Comments on NVIDIA GeForce RTX 2060 12GB Has CUDA Core Count Rivaling RTX 2060 SUPER
Sadge man. Why couldn't they just revived the 2060S instead of making up this abomination w/ VRAM no1 will ever make use of.
Oof.
It is common for Nvidia to make new SKUs of surplus GPUs from previous generations. And this is a positive thing, as it improves availability (slightly) and reduces waste. This is most likely lower unused bins of higher GPUs.
If they started production of a new batch of a particular GPU, then they would have a complete selection of bins of that chip. Why wouldn't it be? Geforce RTX 20 series is supported on Windows 7.
If priced well, it could provide entry into the RT world for GTX users, but I doubt it will be. I'm sure it will be scalped as F, like every other graphics card nowadays.
The 4GB ones seem to be significantly cheaper because as far as I know that amount of VRAM is insufficient for mining. This price gap alone is enough to show the 8GB ones are still bought by miners.
First, motherboard + PSU and PCI extenders + RAM / SSD and CPU will run ballpark $1000 to run 5 mining GPUs. Yes we can tinker with these numbers but it's not going to make much difference.
Electricity in the US is 14.5c/KWH average.
So lets assume RX 580 is $700 as you say, 5 of them will cost $3500.
We're in $4500 not including monitors and so on - we'll assume we already have that stuff.
So using Kryptex mining calculator and that electrical cost:
Using a 580 top mining target is ETH at 8.68c/day after electrical cost using. This means it will take 518 days 24/7 no interruptions to recoup your investment. This is assuming nothing breaks.
I can get a 1660 at Stockx right now for $525. 5 of those is $2625 + $1000 base setup and it will yield about $7.40/day and require 489 days to recoup the investment.
Both of those suck. You'll have multiple failures in the 18+ months to get your capital back. You'll probably lose money.
Lets move on to the 2060. 2060 can be had for $650 for $3250 + $1000 base and will yield $8.68 \ day on ETH. This is 489 days to break even. This still sucks.
The sweet spot:
3060 - can be had for $725 x5 =3625 + 1000 = 4625 initial investment. Yield is $10.64/day on eth giving you 434 days until in the green. Knocks 55 days off the 1660 and 2060.
3070 - can be had for 1050 x 5 = 5250 + 1000 = 6250 initial investment. Yield is $18.60 / day on eth giving you 336 days to ROI, 98 days quicker than the 3060.
3080 - can be had for 1600 x 5 = 8000 + 1000 base = 9000 initial. Yield is 28.84/day on ETH giving you 312 days to ROI.
3090 - can be had for 2800 x 5 = 14,000 + 1000 base = 15000 initial. Yield is 35.37 / day on ETH giving 424 days to ROI. We went backwards, probably not miners driving this one up.
Anyway, I feel we have gone a full circle and back to TSMC 12nm again. In my opinion, the initial GPU shortage was because Nvidia went ahead to drop TSMC 12nm based Turing way too early, while at the same time, their Samsung 8nm based Ampere was still struggling with supply. Do you have an AGP board to run it? ;)
The fourth digit is totally useless isn't it? They'd rather add more digit and letters than changing the fourth one...
trog