After more than two years with Pascal, Nvidia finally introduced a new architecture called Turing, where the company attaches great importance to what has long been seen as a distant dream in computer graphics – ray tracing. This of course means new graphics cards, where partner models are flanked by Nvidia “Founder’s Edition”.
► Nvidia unveils Geforce RTX 2000 series with “Turing” architecture
Nvidia does not use the word reference design for Founder’s Edition, but in practice it is. With Turing, they are taking new steps and introducing, for the first time in years, a completely new cooling solution and more expensive circuit boards than before, something that will provide lower temperatures and sound levels, but also more room for overclocking.
The most distinctive feature is that Nvidia has abolished the single radial fan in favor of a design with dual axial fans. These blow against a solid cooling flange in aluminum without heat conduction pipes and instead are connected to a large steam chamber that covers large parts of the circuit board. The back of all models has a solid cover plate.
Specifications: Nvidia Geforce RTX 2080 Ti, RTX 2080 and RTX 2070
RTX 2080 Ti | GTX 1080 Ti | RTX 2080 | RTX 2070 | |
---|---|---|---|---|
Technical | 12 nm TSMC | 16 nm TSMC | 12 nm TSMC | 12 nm TSMC |
Circuit | TU102 | GP102 | TU104 | TU106 |
Circuit surface | 754 mm2 | 471 mm2 | 545 mm2 | 445 mm2 |
Transistors | 18.6 billion | 12 billion | 13.6 billion | 10.6 billion |
Architecture | Turing | Pascal | Turing | Turing |
CUDA cores | 4 352 st. | 3 584 st. | 2 944 st. | 2 304 st. |
Texture units | 272 st. | 224 st. | 184 st. | 144 st. |
Raster units | 88 st. | 88 st. | 64 st. | 64 st. |
Clock frequency | 1 350 MHz | 1 480 MHz | 1 515 MHz | 1 410 MHz |
GPU Boost | 1 545 MHz | 1 582 MHz | 1 710 MHz | 1 620 MHz |
Computational power | 13 448 GFLOPS | 11 334 GFLOPS | 10 068 GFLOPS | 7 465 GFLOPS |
Memory bus | 352-bit | 352-bit | 256-bit | 256-bit |
Memory amount | 11 GB GDDR6 | 11 GB GDDR5X | 8 GB GDDR6 | 8 GB GDDR6 |
Memory frequency | 14 000 MHz | 11 000 MHz | 14 000 MHz | 14 000 MHz |
Memory bandwidth | 616 GB/s | 484 GB/s | 448 GB/s | 448 GB/s |
Power supply | 8+8-pin | 8+6-pin | 8+6-pin | 8-pin |
SLI connection | NVLink 2.0 x16 | SLI HB Bridge | NVLink 2.0 x8 | – |
Outputs | 1 st. HDMI 2.0b | 1 st. DL-DVI | 1 st. HDMI 2.0b | 1 st. HDMI 2.0b |
TDP | 250 W | 250 W | 215 W | 175 W |
Rec. Award | 999 USD | SEK 7,500 | 699 USD | 499 USD |
* Values apply to Nvidia’s Founder’s Edition of each model, which comes with 90 MHz overclocking. In addition to slightly higher performance, this in all cases leads to an increase in TDP of 10 W.
Nvidia Geforce RTX 2080 Ti, RTX 2080 and RTX 2070 each come with completely different circuit boards, with 13, 8 and 6 phase power supply, respectively. For power supply, the top card has 8 + 8-pin PCI Express, while RTX 2080 goes down to 8 + 6 pins and RTX 2070 goes down to a single 8-pin.
The power supply is well above what is required in terms of each graphics card’s specified power consumption. While Nvidia specifies 250 W for RTX 2080 Ti, 215 W for RTX 2080 and 175 W for RTX 2070 as “reference”, Founder’s Edition steps up to 260 W, 225 W and 185 W respectively.
► Nvidia “Turing” Geforce RTX 2080 performs 50 percent better than GTX 1080
The reason why Founder’s Edition has higher power consumption is attributed to the fact that the trio has a factory overclocking of 90 MHz, which theoretically corresponds to a little more than five percent higher performance than the reference frequencies. However, the models’ GDDR6 memory is left untouched at 14,000 MHz.
Aesthetically, it is noticeable that the trio belongs to the same family, where above all Geforce RTX 2080 Ti and RTX 2080 are the same as berries and like the previous Founder’s Edition, these measure in at 26.7 centimeters. The sibling RTX 2070 is distinguished by being shorter, only 22.9 centimeters, but it is worth noting that the input for power supply is on the short side and in practice adds a few more centimeters. All take up two expansion sites.
Common to the trio are also the connections, where three Displayport 1.4 and an HDMI 2.0b are flanked by a USB Type-C with Virtuallink. This is a new standard that delivers both Displayport HBR3 channels (32.4 Gbps) as well as USB 3.1 (10 Gbps) and power supply of up to 27 watts, enabling the future VR headset through a single cable to the computer.
Another thing that stands out with Founder’s Edition are the prices, which are $ 200 higher for the GTX 2080 Ti and $ 100 higher for the GTX 2080 and GTX 2070. The models are sold on Nvidia’s own website for $ 1,199, 799 and 599, which on the company’s The Swedish side corresponds to SEK 13,999, 9,199 and SEK 6,899, respectively.
Over to something else, the Turing architecture marks the end of traditional SLI bridges. The reason is that the standard is difficult to keep up with in high resolutions and limits communication between graphics circuits. Nvidia therefore instead introduces connection according to the NVLink standard, which for a few years has been used exclusively in the server market.
Nvidia’s NVLink means as much as 50 times higher bandwidth than today’s SLI bridges can deliver. This of course means new bridges and Nvidia offers two variants where the distance between the connections is three and four expansion places. No bridges are offered that support more than two graphics cards in SLI.
With the current generation of graphics cards, the Geforce GTX 1000 series, big headlines were created about Nvidia blocking SLI for GTX 1060 and below. Based on the company’s Founder’s Edition card, Nvidia takes it a step further with the architecture Turing, where not even an RTX 2070 supports SLI. This is thus limited to the top models RTX 2080 Ti and RTX 2080.
Nvidia is not alone in launching the Geforce RTX 2080 Ti and RTX 2080 with its Founder’s Edition on September 20, but these will also be joined by graphics cards from all of the major partner manufacturers. Anyone looking forward to RTX 2070 will have to wait until sometime in October.