Meanwhile, the GPU has a base clock of 1365MHz, with a quoted boost clock of 1680MHz. The RTX 2060 has a total memory bandwidth of 336GB/sec, compared with the RTX 2070’s 448GB/sec. That said, however, the performance will be limited by the limited width of the RTX 2060’s memory interface compared with the RTX 2070.
NVIDIA GEFORCE RTX 2060. PLUS
On the plus side, the 6GB of memory is of the GDDR6 variety, and it’s clocked at the same 14GHz (effective) frequency as the GDDR6 memory on the RTX 2070. It will be interesting to see how the RTX 2060 copes with the demands of future games with this memory allocation, but for the moment it’s enough to handle current titles. That’s a step down from the 8GB found on RTX 2070 cards, and also AMD’s much cheaper Radeon RX 590 cards. The total amount of memory is also, naturally, reduced by 25 per cent, with the RTX 2060 coming with 6GB. As such, the number of ROPs falls from 64 to 48, and the total amount of L2 cache falls from 4MB to 3MB. What’s more, as the Turing architecture links the memory controllers to the ROPs and the shared pool of L2 cache, these parts in turn are also reduced by 25 per cent. That reduces the width of the memory interface from 256 bits in the RTX 2070 to 192 bits in the RTX 2060. The other notable omission from the RTX 2070’s GPU is a pair of memory controllers, reducing the total number from eight to six. Likewise, the number of Tensor cores, which are used to handle Nvidia’s deep learning features, such as its DLSS anti-aliasing mode, have been reduced, from 288 in the RTX 2070 to 240 in the RTX 2060. Accordingly, Nvidia rates the RTX 2060’s ray tracing performance at 5 Giga Rays/sec, compared with 6 Giga Rays/sec on the 2070 (and 8 Giga Rays/sec on the RTX 2080).
The reduction in the number of SMs means that the RTX 2060 has 30 RT cores, compared with the 36 found in the RTX 2070. For starters, Turing has RT cores for handling real-time ray tracing work in games, using a hybrid rendering approach that fuses traditional rasterisation techniques with some mightily impressive real-time ray tracing effects.
This reduction in SMs doesn’t just affect stream processors either, as there are also two more core-types found in the Turing architecture’s SMs than in the Pascal architecture. The end result is a chip with 1,920 stream processors (the same number found in the GeForce GTX 1070), compared with 2,304 in the RTX 2070. Firstly, six of the streaming multiprocessors (SMs) are switched off, giving you a total of 30. The RTX 2060 is based on the same 12nm TU106 Turing GPU found in the RTX 2070 (see Issue 184, p28), with an identical 445mm² die containing 10.8 billion transistors, but with some parts disabled. It’s undoubtedly still expensive, but it’s also a long way from the stratospheric prices of the RTX 20 Ti. We’re hesitant to describe that as a good price, given that this card’s predecessor, the GTX 1060 6GB, could be bought for just £239 (or £279 for the Founders Edition) when it launched. Speaking of which, the prices are now starting to become affordable, with the GeForce RTX 2060 arriving with a £329 inc VAT price tag. We were highly impressed with real-time ray tracing in Battlefield V (see Issue 186, p92) – it just needs some more game support to justify a costly hardware upgrade. Real-time ray tracing how now made its way into Battlefield V, and this game is also promising support for Nvidia’s deep-learning accelerated anti-aliasing system, DLSS, which is also supported by Final Fantasy XV and Anthem.
That situation is thankfully changing now, though (at least a bit, anyway). There were no real games to show off their new abilities, and their prices bordered on the nonsensical. Rather than igniting the graphics landscape with an explosive bang, Nvidia’s first RTX GPUs seemed to wander around their designated landing site looking lost and bewildered.
0 kommentar(er)
