GeForce RTX 2060 12GB quietly released. More memory costs more

RTX 2060 in name only, performance closer to the 2060 Super?

GeForce RTX 2060 with 12GB of memory is officially here. Earlier leaks had it coming on December 7, which is the date of store availability. In the end Nvidia already “revealed” the card last week. But extremely quietly, the company just published the specs of the “new” on its website. The return of the RTX 2060, or arguably RTX 2060 Super is now a real thing. But unfortunately there are some questions around the price.

The release of the card actually happened without any press releases or announcements whatsoever, which probably has something to do with the fact that it was a comeback of a card under a name of product that has already been sold for years. But in reality, this card doesn’t match the original GeForce RTX 2060, even if we ignore the increased memory. The memory increase from 6GB is naturally the most obvious change, which will also be represented in the product name – Nvidia has listed the card as GeForce RTX 2060 (12GB) on their site.

But alongside this comes another unexpected change: the card will also have increased compute power (and power draw). To some extent, this graphics card is actually closer to the GeForce RTX 2060 Super, so the benchmarks might turn out to be more interesting than expected.

The GPU used is a 12nm TU106 die based on the 2018 Turing architecture (not the current Ampere), with 34 SM blocks and thus 2176 shaders (“Cuda cores”), 34 RT cores and 136 Tensor cores. Frequency is quoted as 1470 MHz base and 1650 MHz boost. The original RTX 2060 card had only 30 SM blocks (1920 shaders) at 1365-1680 MHz. If the performance matches these clock speeds, the 12GB version could have up to 11-22% higher raw compute performance (but it will come down to what the real game frequencies are, which may differ from the claimed boosts).

This increase in raw power doesn’t come for free, Nvidia bumped up the power draw from 160W on the GeForce RTX 2060 6GB to 185W on the GeForce RTX 2060 12GB (a 15% increase). The GeForce RTX 3000 generation has also frequently increased power draw against the predecessor RTX 2000 cards, so in a sense this move is like bringing old Turing closer to today’s Ampere cards. The cards are still supposed to be powered by a single eight-pin connector according to the reference specs, but the recommended power supply capacity has gone up from 500 to 550W.

GeForce RTX 2060 12GB specs (Source: Nvidia)

12GB memory but worse bandwidth than 2060 Super

In terms of raw performance, the GPU configuration in this card seems to be virtually identical to the GeForce RTX 2060 Super – same number of units and clock speeds, though the TDP is even higher. Of course, then there’s the second part, where the card is only on par with the original RTX 2060. The latter had a 192-bit memory bus with 14.0GHz GDDR6 memory and 336GB/s bandwidth, while the RTX 2060 Super used a full 256-bit bus (carrying 8GB of memory) and had 448GB/s bandwidth.

Since 12GB memory requires a 192-bit bus, the first option was chosen and therefore the GeForce RTX 2060 12GB has a bandwidth of only 336GB/s. Therefore, performance in games can be expected to end up being worse than the GeForce RTX 2060 Super – somewhere between the Super and the original weaker vanilla RTX 2060 6GB. The card will probably perform best at lower resolutions and might fall behind a bit at higher resolutions.

Nvidia GeForce RTX 2060 Founders Edition (Source: Nvidia)

Only non-reference cards, price will be higher than regular RTX 2060

When the cards do go on sale, they will be exclusively non-reference cards with “custom” cooling designs from various manufacturers. At first, it seemed that Nvidia might even offer a Founders Edition card (see image above). This would be interesting in that Nvidia sells these cards from its own eshop, where prices are not inflated (if the cards can be ordered). So if the cards were sold there and you managed to snag one, you could avoid the scalpers’/cryptomoguls’ markups. Eventually, though, Nvidia removed the mention of the Founders Edition from its website and confirmed that this reference version will not exist. Thus, your only choice is to comb through regular online stores as usual.

Card sales have started on December 7. Unfortunately, the suggested price has not been communicated. Nvidia has only communicated that they consider the larger memory to be (logically) extra value, the 12GB model is “premium”, and therefore the price of these cards will realistically be higher than the 6GB version, which is officially priced at $299. However, nothing has been said about how much the difference roughly is, if there is any internal recommended amount at all. It shouldn’t be much more, as the RTX 3060 is officially supposed to cost $329 and has premium 12GB memory as well. Nvidia has only officially communicated that it leaves pricing up to the card manufacturers.

It is probably also a certain sort of resignation to the current market situation, where the real selling prices of graphics for end users are up to about twice what the cards should cost according to the recommended amounts. And some such markup over and above the (not even disclosed) official price will undoubtedly be charged for these cards as well. First news after the sale start reported cards going for as much as 600 to 700 €.

GeForce RTX 2060 12GB specs, extended specs (Source: Nvidia)

Why the return of Turing?

It may seem strange that a GeForce RTX 2000 (Turing) generation card should be released as a new market entry now, when the RTX 3000 series with the latest improved Ampere architecture has been on sale for over a year. The reason for this is lies almost certainly in the chronic shortage of GPUs we are experiencing. Ampere GPUs are made on Samsung’s 8nm process, and it’s possible Nvidia/Samsung can’t ramp up production further on this node. However, the Turing chips use TSMC’s cheaper 12nm process, which might have more readily available free capacity now. It’s likely that by restarting sales of these graphics cards, Nvidia will be able to supply more graphics overall thanks to the combination of Samsung and TSMC resources. Turing is therefore returning to the market as a complement to the GeForce RTX 3000 line of graphics.

Turing generation GPU (Image, Source: Nvidia)

Since Turing GPUs have support for ray tracing effects (RT cores) and DLSS or Nvidia Broadcast (Tensor cores), this card will fit into the current range well despite using an older GPU. It should probably be positioned between the upcoming GeForce RTX 3050 and the already released GeForce RTX 3060 in terms of performance.

However, this graphics will lack something compared to the current generation: for example, the card will not have hardware video acceleration of AV1 format, will not support PCI Express 4.0 (the interface is PCIe 3.0 ×16) and will probably not be able to use PCIe Resizable BAR. It also has a bit higher power draw than the GeForce RTX 3060, even though it should be slower.

Sources: Nvidia, VideoCardz

English translation and edit by Jozef Dudáš, original text by Jan Olšan, editor for Cnews.cz



  •  
  •  
  •  
Flattr this!

Leave a Reply

Your email address will not be published. Required fields are marked *