Turing Discussion Thread

[SteelCrysis]

https://www.techpowerup.com/257012/nvidi...a-fab-11nm

During our disassembly of the GeForce RTX 2060 Super, we noticed a shocking detail. The 12 nm "TU106" GPU on which it is based, has the marking "Korea." We know for a fact that TSMC does not have any fabs there. The only Korean semiconductor manufacturer capable of contract-manufacturing a piece of silicon as complex as a GPU, for a designer with the energy-efficiency OCD as NVIDIA, is Samsung.

What makes this interesting is that Samsung does not officially have a 12 nm FinFET process. It has 14 nm, and the 11LPP, a 11 nm nodelet, which the company designed to compete with TSMC 12 nm. It would hence be really interesting to hear from NVIDIA on whether they've scaled out the "TU106" to 14LPP, or down to 11LPP at Samsung. It's interesting to note that the shrink in transistor sizes in these nodelets doesn't affect die-sizes. We hence see no die-size difference between these Korea-marked chips, and those marked "Taiwan." We've reached out to NVIDIA for comment.

https://www.tomshardware.com/reviews/nvi...207-8.html

GeForce RTX 2070 Super is an attempt to improve Turing’s standing among gamers who turned their noses up at GeForce RTX 2070 last year. The Founders Edition model we tested is almost 13% faster than its predecessor at a more attractive $500 price point. Nvidia’s partners probably aren’t pleased that they’re now battling a beefy reference design. But gamers benefit, which is what we want to see.

Inserted right between the $350 GeForce RTX 2060 and $500 RTX 2070 Super, we can’t imagine that anyone actually asked for a $400 GeForce RTX 2060 Super. However, if it’s able to outperform Radeon RX 5700 when the vanilla 2060 would have lost, then you know the game Nvidia is playing. Sandwich AMD’s card between a slightly slower and a slightly faster GeForce, then use “but ours has ray tracing” as the coup de grâce to dissuade potential customers. That’s a tough argument to beat, except with a lower price. We’ll have to see how AMD responds.
...
Cramming today’s launch in between AMD’s Radeon RX 5700-series announcement and performance review means that we can’t give the GeForce RTX 2060 Super or 2070 Super a recommendation one way or the other, though. In a few short days, the results from both Navi-based boards will go live. At that point we’ll have a more complete picture of high-end graphics performance in 2019. Count on us to declare a winner once the smoke clears.

https://www.techpowerup.com/review/nvida...er/34.html
The RTX 2070 Super is performing similarly to the Radeon VII at a lower price point.

Results from our new graphics card test suite with all the latest games and a new Core i9-9900K paired with an EVGA Z390 DARK motherboard, show a solid 14% performance improvement over RTX 2070. Last generation's flagship, the GeForce GTX 1080 Ti is 2% slower, so RTX 2070 Super can be considered to deliver equal performance. AMD's fastest, the Radeon VII is 3% behind, and we expect the Navi-based Radeon RX 5700 XT to be around 10% slower than RTX 2070 Super, too. The next step up is the GeForce RTX 2080, which is 7% faster, not that much, especially when you consider the price difference. With those performance numbers, we can recommend RTX 2070 Super for highest detail gaming at 1440p.
...
Releasing on July 9th, priced at $500, the RTX 2070 Super is not cheap. It comes at the exact same price point as RTX 2070 — but with more performance, so effectively price/performance does go up. We reached out to NVIDIA whether the MSRP of RTX 2070 Non-Super will go down now, and they responded that they haven't made any changes to the RTX 2070 MSRP, but naturally its price will go down, following the changed market conditions. In order to sweeten the deal, NVIDIA will include a two-game bundle with all RTX Super cards: "Wolfenstein Youngblood" and "Control." Both titles come with support for NVIDIA RTX raytracing, to show off the capabilities of their new technology. The new RTX Super lineup will definitely turn up the heat on AMD, who are pricing their RX 5700 XT at $450 (for now). RX 5700 XT will definitely not be able to match RTX 2070 Super performance, and we're having doubts whether thermals or noise will end up being comparable, which could make the $50 price increase to Super very justifiable for potential customers, or force AMD to reduce their pricing.

https://www.techpowerup.com/review/nvidi...er/34.html

With these hardware changes, the RTX 2060 Super is a significantly faster card than the original RTX 2060, without beating the original RTX 2070. It gets danger-close, though. At 1080p, the card is 10% faster than the original RTX 2060, and 12% faster at 1440p. The gap increases at 4K to 13%, and although not its forte, quite a few games in our bench are playable with this card. It's hard to say just how much of a dividend the 33% faster and larger memory setup pays, but it could certainly do wonders for the card's future-proofing, especially when you take into account memory sizes on future consoles. Across the competitive landscape, the RX Vega 64 is the closest AMD card to the RTX 2060 Super, and trails by 11% across all resolutions. This landscape will probably change when AMD formally launches the RX 5700 and RX 5700 XT next week.
...
Overclocking the RTX 2060 Super is a breeze, with a massive 29 percent GPU overclock (+200 MHz) without tinkering with the voltages, and an easy-peasy memory overclock to 16 Gbps. It's astonishingly easy to get this card up to RTX 2070 performance-levels, which is probably why we hear rumors of NVIDIA retiring the RTX 2070 from its product stack. What the RTX 2060 Super doesn't do, however, is replace the RTX 2060 at $349. It instead comes at a $50 premium, which is probably NVIDIA trying to offset the cost of more memory chips and stronger VRM solutions. Unless NVIDIA cuts prices of the original RTX 2060, it risks the RTX 2060 Super cannibalizing not just the RTX 2070, but also the original RTX 2060. We would definitely spend the extra $50 on the Super not just for the added performance, but also the future-proofing the added memory brings to the table.