Resetera.gif
Well, the GP102 chip is 471mm^2 and has 11.8B transistors, the TU104 chip is 545mm^2 and has 13.6B transistors. So, being in a version of the same process, it's a significantly bigger chip, and has the new and fancy GDDR6 memory.
I did intend to mention die sizes in my post but decided to take it out :P
There are a huge number of asterisks which should be attached to this statement.
For example, you can buy a GTX 1660 right now which will cost you $220 and provide more or less the same performance as a 980Ti which was sold for $650 four years ago.
The fact that some card's name is considered a successor of some other card's name and there's a price hike between them means zero without looking at performance of these cards.
Basically, always check what price/performance comparisons show, not how much a xx60 or xx80 card costs.
Sorry I should have been more clear. I was referring to the direct succeeding generation of GPUs offering more performance at lower price points.
For example, the GTX 970 delivered GTX 780 Ti performance at a significantly lower price, IIRC, in the UK it was selling for around half the price the 780 Ti was, so around £500 down to around £260. This was huge!
The 780 Ti also had a 28nm 561 mm² die while the 970 had a 398 mm² die.
The GTX 1660 offering 980 TI performance 4 years later (3 GPU generations later) isn't really impressive to me, compared to the GTX 970 situation.
I haven't seen anything like it since, hence the reason I made the claim of NVIDIA GPUs moving up in price each generation, although that claim does have it's flaws as there have been performance gains, it's just that their model number segmentation doesn't appear to hold as much relevance as it did with past generation GPUs.
EuroGamer - Nvidia GeForce GTX 970 review
The RTX 2060 appears to be between a GTX 1070 Ti and 1080 and is cheaper than I can recall seeing the 1070 Ti at which is a positive, although it does have 2GB less VRAM.
It's about 10% faster on average and that gap gets bigger in newer titles which Turing architecture is better suited to. It will grow even further if some game will use Turing's capabilities - and that's complex compute (counting FP16 here as well; not async though as Pascal is doing async compute with essentially the same gains as Turing), VRS, mesh shaders. All in all 2080's price/performance is better than that of 1080Ti, especially if we're looking at future titles, but not significantly so. It makes 1080Ti 100% obsolete though.
I say they're similar performers as they're typically within 10% of each other, although in games such as Wolfenstein 2 and Rainbow Six Siege the Turing GPUs really stretch their legs. If more games continue to take advantage of Turing's architectural advantages they'll really shine.
Turing is definitely impressive from an architecture standpoint, and I'm curious to see what will be done with it's Raytracing and mesh shading capabilities, it's just that I'm not very enthusiastic about the price points of the GPUs.
TechPowerUp: ASUS GeForce RTX 2080 Ti Matrix 11 GB Review - Page 22
TechPowerUp: ASUS GeForce RTX 2080 Ti Matrix 11 GB Review - Page 26
Well, yeah, but TBH 2080 doesn't make any sense as an upgrade option for 1080Ti owners so it's not a regression in the sense of a typical user upgrade path. G6 is rather expensive, a year ago especially. Which is coincidentally also why we don't see AMD doing their favorite trick with 5700 this time - providing double the RAM of NV's competition.
I'm curious to see the RX 5700's performance when reviews are published and the card is in the hands of consumers, but so far I'm not impressed with it's price point, it's going against up the RTX 2070 which is a very capable and advanced GPU and doesn't appear to offer much in competition besides allegedly faster performance.
Edit - Corrected spelling mistakes.
Last edited: