Comments by "IIIRattleHeadIII" (@badass6300) on "NVIDIA 4060Ti Details... NVIDIA finally listening to consumers?" video.

Not to defend Nvidia, but considering that 2GB of fast GDDR6 VRAM, 21-24gbps, costs 19-26$ in bulk, that means that 8GB cost 76-104$. Doesn't change that the RTX 4060Ti 8GB should be 270-280$. On the other hand if the 4060Ti is using slow 14-18Gbps GDDR6, then it's literally half the price of the 21-24gbps per 2GB chip and there is no excuse. Also according to JEDEC specs, there should be 1.5GB and 3GB GDDR6 chips, so I'd say they should have used 3GB chips and made the 4060Ti 12GB
9
@fabiusmaximuscunctator7390 Here is how it goes: The rtx 4060Ti is a 50 class card, marketed as a 60Ti class card sold for 70 class prices. The rtx 4070 is a 50Ti class card, marketed as a 70 class card, sold at 80 class prices. The rtx 4070Ti is a 60 class card, marketed as a 70Ti class card, sold at 90 class prices The rtx 4080 is a 70 class card(389mm^2 vs rtx 3070 392mm^2), marketed as an 80 class card sold at insane prices The rtx 4090 is an 80 class card, markted as a 90 class card, sold at ludicrous prices... So here are how prices go: 4060 8GB - 220$ 4060Ti 8GB - 270$ 4070 - 350$ 4070Ti - 420$ 4080 - 600$ 4090 - 900 -1000$ Anything more is unacceptable.
3
@commanderoof4578 I was talking about GDDR6, I don't know the prices of GDDR6X, those aren't publicly available. Fast GDDR6 came out this year, 21Gbps and 24Gbps to be exact. Also to elaborate on the bus width. Each GDDR6(X) chip runs either at 32 bit width or 16 bit width, that means that each chip takes 32 bits of the memory bus, thus to run at full bandwidth a 128 bit card can have 4x GDDR6(X) chips running at 32 bit width, until a couple of months ago, the highest capacity GDDR6(X) chip was 2GB, that means that until recently a 128 bit card could NOT have more than 8GB of VRAM at full bandwidth. If you for example put more than 4x chips on a 128 bit card, for example 8, that means that each of the chips will have to run at 16 bit mode, which would cut bandwidth in half, just like the RTX A4000 ADA which has 20GB of RAM on a 160 bit bus with 10x 2GB Chips, instead of 5x 2GB chips and has half the bandwidth of a 4070. By JEDEC standards there should be 1.5GB and 3GB GDDR6 chips as well, but it seems that they have skipped them. 3GB is a missed opportunity as the jump from 8GB to 16GB is extreme and the cards don't have enough bandwidth to utilize it effectively, while 12GB would have been much better overall for 1440p gaming. 1080p is still fine with even 6GB of VRAM, so 8GB is perfect for 1080p in the next few years.
1
@commanderoof4578 the 3090 has 12 chip MODULES, not 24 chips. They did that because that's how you got 2GB per chip back when it launched, before proper 2GB chips. And if you check AMD's RX 7000 cards have 20-21Gbps GDDR6 chips.
1
@commanderoof4578 AMD can't pick GDDR6X it's co-developed by Nvidia and Nvidia like to keep their tech propriatery. And the 3090 has 12 spots for the VRAM, meaning it is wired for 12 chips and thus has 12 chips/modules.
1
@commanderoof4578 And they are paired and wired into a single module... there are 12 positions, not 24 positions for the chips. Also the founder's card has them only on the front. 3rd party cards had them bundled like that, because when the rtx 3090 came out 2x 1GB chips costed less than 1x 2GB chip.
1