NVIDIA recently released the GeForce GTX 1660 Ti gaming graphics card based on the new TU116 GPU, but Turing's move to budget devices is not over yet. With the GTX 1660 Ti, the company replaced the GeForce GTX 1070 with a fresher, more affordable, lower power model, but the new GeForce GTX 1660 has another challenge: to close the gap in the NVIDIA catalog that still exists between the GeForce GTX 1060 and GTX 1070. Last fall, the Radeon RX 590 settled in this gap, and the Radeon RX 580, as a result of driver optimization and the transition of games to Direct3D 12, became at least a worthy alternative to the GeForce GTX 1060. But with the release of the GTX 1660, the "red" GPUs have a serious rival in the mass category of consumer video cards, because the new product is both cheaper than the Radeon RX 590 and has a higher performance potential.
Specifications, price
At the heart of the GeForce GTX 1660 is the TU116 GPU with partially deactivated compute units. The difference in GPU configuration between the GTX 1660 and GTX 1660 Ti boils down to two streaming multiprocessors (SMs), which in total contain 128 32-bit CUDA cores and 8 texture mappers. Thus, the throughput of the GeForce GTX 1660 suffered only 8,3% in floating point operations and texel fillrate without adjusting for GPU clock speeds. And the frequencies, by the way, have only grown in the younger model: NVIDIA increased the base frequency by 30 MHz, and the Boost Clock by 15 MHz.
But such careful changes would not be enough to differentiate between the GeForce GTX 1660 and the GTX 1660 Ti. The main feature separating the two models was the type of RAM. While the Ti modification is equipped with GDDR6 chips with a bandwidth of 12 Gb / s per pin, the GeForce GTX 1660 returned to the GDDR5 standard. Moreover, the GTX 1660 is equipped with chips with a bandwidth of 8 Gb / s, which means that in terms of total memory bandwidth, the new video card is fully consistent with the starting specifications of the GeForce GTX 1060 with 6 GB of RAM, and the later versions of the GTX 1060 with 9 Gb / s RAM even win the GTX 1660 has this parameter. However, the TU116 graphics processor, thanks to improved color compression, works with RAM more efficiently.
Since neither the clock speeds nor the GPU configuration of the GeForce GTX 1660 and GTX 1660 Ti differ significantly, and the younger model also carries RAM chips with higher power consumption (compared to GDDR6), the two younger accelerators of the Turing family are characterized by the same power reserve - 120 W .
We have already discussed other characteristics of the TU116 chip in comparison with full-fledged representatives of the Turing family (TU106, TU104 and TU102) in the GeForce GTX 1660 Ti review, but it is worth focusing on several key features that make TU116 related to its older counterparts or, conversely, draw between them an insurmountable boundary. In total, the TU116 has all the innovations that NVIDIA has implemented in the Turing architecture, with the exception of cores that perform Ray Tracing, and tensor cores that perform FMA (Fused-Multiply Add) calculations on half-precision real number matrices (FP16). The latter are used primarily in machine learning tasks, when the GPU passes data through a neural network previously formed locally or on a remote farm. Thus, the GeForce GTX 1660 and GTX 1660 Ti simultaneously lost compatibility with both DXR (Direct3D 12 extension for ray tracing) and DLSS technology, which allows the GPU to render at a reduced resolution, followed by frame scaling by the neural network.
Instead of tensor blocks, NVIDIA equipped the TU116 with a separate array of 16-bit CUDA cores - they are not fast enough for DLSS to work efficiently, but there are already games that use half-precision operations in shader calculations (for example, Wolfenstein II: The New Colossus), for due to which the performance of suitable GPUs is significantly increased (now these are Vega and Turing chips). Otherwise, again, the TU116 differs from the older chips of its family only in a quantitative way, it has all the pipeline optimizations inherent in the Turing architecture, and supports proprietary rendering functions such as VRS (Variable Rate Shading).
| Manufacturer | NVIDIA | |||||
|---|---|---|---|---|---|---|
| Model | GeForce GTX 1060 3 GB | GeForce GTX 1060 6 GB | GeForce GTX 1660 | GeForce GTX 1660 Ti | GeForce RTX 2060 | GeForce RTX 2070 |
| Graphics Processor | ||||||
| Name | GP106 | GP106 | TU116 | TU116 | TU106 | TU106 |
| Microarchitecture | Pascal | Pascal | Turing | Turing | Turing | Turing |
| Process technology, nm | 16nm FinFET | 16nm FinFET | 12 nm FFN | 12 nm FFN | 12 nm FFN | 12 nm FFN |
| Number of transistors, million | 4 400 | 4400 | 6 600 | 6 600 | 10 800 | 10 800 |
| Clock frequency, MHz: Base Clock / Boost Clock | 1506/1708 | 1506/1708 | 1530/1785 | 1500/1770 | 1365/1680 | 1 / 410 (Founders Edition: 1 / 620) |
| Number of shader ALUs | 1152 | 1280 | 1408 | 1536 | 1920 | 2304 |
| Number of texture overlays | 72 | 80 | 88 | 96 | 120 | 144 |
| ROP number | 48 | 48 | 48 | 48 | 48 | 64 |
| Number of tensor cores | No | No | No | No | 240 | 288 |
| Number of RT cores | No | No | No | No | 30 | 36 |
| RAM | ||||||
| Bus width, bit | 192 | 192 | 192 | 192 | 192 | 256 |
| Chip type | GDDR5SDRAM | GDDR5SDRAM | GDDR5SDRAM | GDDR6SDRAM | GDDR6SDRAM | GDDR6SDRAM |
| Clock frequency, MHz (bandwidth per contact, Mbps) | 2000 (8000) 2250 (9000) | 2000 (8000) 2250 (9000) | 2000 (8000) | 1 500 (12 000) | 1 750 (14 000) | 1 750 (14 000) |
| Volume, MB | 3 096 | 6 144 | 6 144 | 6 144 | 6 144 | 8 192 |
| I/O bus | PCI Express 3.0 x16 | PCI Express 3.0 x16 | PCI Express 3.0 x16 | PCI Express 3.0 x16 | PCI Express 3.0 x16 | PCI Express 3.0 x16 |
| Performance | ||||||
| Peak performance FP32, GFLOPS (based on maximum specified frequency) | 3935 | 4372 | 5027 | 5437 | 6451 | 7 / 465 (Founders Edition) |
| Performance FP32/FP64 | 1/32 | 1/32 | 1/32 | 1/32 | 1/32 | 1/32 |
| Performance FP32/FP16 | 1/128 | 1/128 | 2/1 | 2/1 | 2/1 | 2/1 |
| RAM bandwidth, GB/s | 192/216 | 192/216 | 192 | 288 | 336 | 448 |
| Image output | ||||||
| Image output interfaces | DL DVI-D, DisplayPort 1.3/1.4, HDMI 2.0b | DL DVI-D, DisplayPort 1.3/1.4, HDMI 2.0b | DL DVI-D, DisplayPort 1.4a, HDMI 2.0b | DL DVI-D, DisplayPort 1.4a, HDMI 2.0b | DL DVI-D, DisplayPort 1.4a, HDMI 2.0b | DL DVI-D, DisplayPort 1.4a, HDMI 2.0b |
| TBP/TDP, W | 120 | 120 | 120 | 160 | 175/185 (Founders Edition) | |
| Retail price (USA, without tax), $ | 199 (recommended at the time of release) | 249 (recommended at launch) / 299 (Founders Edition, nvidia.com) | 229 (recommended) | 279 (recommended) | 349 (recommended) / 349 (Founders Edition, nvidia.com) | 499 (recommended) / 599 (Founders Edition, nvidia.com) |
| Retail price (Russia), rub. | ND | ND (recommended at launch) / 22 (Founders Edition, nvidia.ru) | 17 990 (recommended) | 22 990 (recommended) | ND (recommended) / 31 (Founders Edition, nvidia.ru) | ND (recommended) / 47 (Founders Edition, nvidia.ru) |
The GeForce GTX 1660 can already be called the third successor (after the RTX 2060 and GTX 1660 Ti) of the main mid-range graphics card in the Pascal family - the GeForce GTX 1060. version of the GeForce GTX 1060 with 3 GB of RAM. Compared to the latest GTX 1660, not only does it have twice the frame buffer, but it also outperforms the old model by 27% in theoretical shader throughput, just as the GTX 1660 Ti outperforms the full GTX 24 with 1060 GB of RAM by 6%. At the same time, NVIDIA does not abandon the pricing policy set by the GeForce RTX 20 family of video cards, in which all new devices cost the buyer more than their direct counterparts by model numbers from the previous generation. So the GeForce GTX 1660 went on sale at a recommended price of $229, although the GeForce GTX 1060 with 3 GB of RAM started at $199.
Looking at the price tag of the novelty, one could once again resent NVIDIA's greed, if it were not for the weakness of AMD's current offerings, which, with the advent of the Turing architecture, spread from the upper to the middle price segment. So, the most affordable modifications of the Radeon RX 590 (at a price of $ 240 on newegg.com) are currently more expensive than the GeForce GTX 1660, and in the Russian market NVIDIA's recommendation (17 rubles) placed the GTX 990 in the lower part of the range, which occupies an AMD product (from 1660 rubles according to market.yandex.ru).
Unlike other Turing-based accelerators, including the GeForce GTX 1660 Ti, the GTX 1660 has no direct rivals in its own camp. The closest 10-series models in terms of specifications and speed - GeForce GTX 1060 6 GB and GeForce GTX 1070 - are far from the new product in terms of price, although the first one (and the GTX 1060 is now sold at prices starting at $ 209 or 14 rubles) will still delay on some potential buyers until stocks of old NVIDIA hardware accumulated during the cryptocurrency boom run out.
GIGABYTE GeForce GTX 1660 OC: construction
The first impression of a new graphics card of the lower and middle price category (and, by the way, about expensive models too) is best made using a simple modification as an example, because these are the most in demand - unlike "premium" versions based on the same GPU , which at a price often penetrate the range of the nearest of the older models. In this sense, we are lucky again, because the GeForce GTX 1660 represents a GIGABYTE device released outside the well-known WINDFORCE and AORUS series. You can't go wrong with the photos showing the same graphics card we tested three weeks earlier in our GeForce GTX 1660 Ti review - using the same PCB and cooling, but with a different GPU and GDDR5 chips instead of GDDR6.
The GPU on the GIGABYTE GeForce GTX 1660 OC is pre-overclocked. Although we won't find out the exact data about its passport frequencies until the publication of the article, when the manufacturer posts a description of the new products on its own website, it is already clear from the modest cooling system that overclocking here is purely symbolic. Yes, and the older model GIGABYTE overclocked by only 30 MHz.
The design of the GIGABYTE GeForce GTX 1660 OC shows signs of savings everywhere. The video card is devoid of even the simplest backlight, not to mention RGB LEDs with a customizable shade and the ability to connect LED strips. The casing, which is entirely made of plastic, covers the PCB on three sides, hiding the compact dimensions of the PCB.
The cooling system is also extremely simple: the heat of the GPU and RAM chips is dissipated by an aluminum radiator, and the only copper part is a heat pipe passed through its base. However, the GIGABYTE GeForce GTX 1660 OC cooler is not without individual improvements. So, the radiator has a protrusion in contact with field-effect transistors and voltage regulator chokes, and two fans with a diameter of 87 mm rotate in opposite directions - thus reducing the turbulence of the air flow.
|
|
|
The GIGABYTE GeForce GTX 1660 OC bundle is as ascetic as it is possible in principle: in addition to the video card itself, the box contains only paper instructions and a software CD.
GIGABYTE GeForce GTX 1660 OC PCB
Based on the PCB used in the GeForce GTX 1660, GIGABYTE has already produced a host of other devices, from the GeForce GTX 1660 Ti to the GeForce RTX 2070. In this range of models, various GPUs (TU116, TU106) and two types of RAM chips (GDDR5 and GDDR6) are electrically compatible, and the small dimensions of the PCB made it possible to produce both standard-sized devices and compact Mini ITX form factor video cards.
This PCB can accept components for eight phases of the voltage regulator, but the power consumption of devices based on the TU116 and TU106 ranges from 120 to 175 W (according to the reference specifications), so the junior accelerator is content with a six-phase VRM: four phases serve the GPU and two - microcircuits random access memory. Due to its relationship with the older models of the Turing family, the novelty is equipped with field-effect transistors with an integrated driver (the so-called DrMOS or βpower stagesβ - power stages), which provide high efficiency and allow the VRM PWM controller to record the voltage at the transistor drain with high accuracy.
Although the TU116's display controller is DVI compatible, GIGABYTE opted for three DisplayPort connectors and a single HDMI output. But the USB 3.1 Gen 2 interface with support for the DisplayLink protocol and the GeForce GTX 1660 and GTX 1660 Ti are fundamentally deprived. Contact pads for monitoring voltages and a hardware voltmode, a backup BIOS chip and other similar luxury, of course, are also absent here.
Source: 3dnews.ru