Thanks to a recent driver update, GeForce 10 and 16 series graphics cards have learned how to perform ray tracing in DXR-enabled games. On и (and any other NVIDIA models, with the exception of GeForce RTX and older accelerators based on Pascal chips) this function works in demo mode rather than has any practical value. But even if real-time ray tracing is not up to the GeForce GTX 1660, this does not detract from other advantages of the youngest member of the Turing family. The GeForce GTX 1060 lost its former appeal, and the balance of power among mid-range graphics cards was shifted in favor of the Radeon RX 580 and , however, the novelties on the TU116 chip instantly brought NVIDIA back to the leading position.
In reviews of high-performance video cards, the attention of the press is usually focused on expensive models with powerful cooling systems, spectacular design and a lot of additional features. But when it comes to mass hardware, it is more important to study the lower positions of the price list. The low power consumption of the GeForce GTX 1660 makes it possible to greatly simplify the binding of the video card, while at the same time maintaining a large margin of safety for overclocking. To what extent the simplification retains overclocking capabilities - we will study using the Inno3D GeForce GTX 1660 Twin X2 as an example.
Specifications, scope of delivery, prices
Unlike products from more famous manufacturers, the catalog of modern Inno3D offers is not diverse, and the GeForce GTX 1660 and GTX 1660 Ti are only available as Twin X2 models. Under this brand, Inno3D has released several accelerators on all GPUs of the Turing architecture - up to the GeForce RTX 2080 Ti. Each of them is remarkable for its affordable price in its market niche and operates at strictly reference clock frequencies. Accordingly, the base GPU frequency of the Inno3D GeForce GTX 1660 Twin X2 is 1530 MHz, and the estimated value at typical load (Boost Clock) is 1785 MHz. However, NVIDIA underestimates the potential of its video cards, because even in the most demanding games, GeForce 10, 16 and 20 series devices usually overclock more. However, we will find out for sure at the testing stage. Like all variants of the GeForce GTX 1660, the Inno3D board carries 6 GB of GDDR5 RAM, which again features a reference bandwidth of 8 Gb / s per bus pin.
| Manufacturer | NVIDIA | Inno3D |
|---|---|---|
| Model | GeForce GTX 1660 | GeForce GTX 1660 Twin X2 |
| Graphics Processor | ||
| Name | TU116 | TU116 |
| Microarchitecture | Turing | Turing |
| Process technology, nm | 12 nm FFN | 12 nm FFN |
| Number of transistors, million | 6 600 | 6 600 |
| Clock frequency, MHz: Base Clock / Boost Clock | 1530/1785 | 1530/1785 |
| Number of shader ALUs | 1408 | 1408 |
| Number of texture overlays | 88 | 88 |
| ROP number | 48 | 48 |
| Number of tensor cores | No | No |
| Number of RT cores | No | No |
| RAM | ||
| Bus width, bit | 192 | 192 |
| Chip type | GDDR5SDRAM | GDDR5SDRAM |
| Clock frequency, MHz (bandwidth per contact, Mbps) | 2000 (8000) | 2000 (8000) |
| Volume, MB | 6 144 | 6 144 |
| I/O bus | PCI Express 3.0 x16 | PCI Express 3.0 x16 |
| Performance | ||
| Peak performance FP32, GFLOPS (based on maximum specified frequency) | 5027 | 5027 |
| Performance FP32/FP64 | 1/32 | 1/32 |
| Performance FP32/FP16 | 2/1 | 2/1 |
| RAM bandwidth, GB/s | 192 | 192 |
| Image output | ||
| Image output interfaces | DL DVI-D, DisplayPort 1.4a, HDMI 2.0b | DL DVI-D, DisplayPort 1.4a, HDMI 2.0b |
| TBP/TDP, W | 120 | 120 |
| Retail price (USA, without tax), $ | 229 (recommended) | ND |
| Retail price (Russia), rub. | 17 990 (recommended) | From 15 658 (market.yandex.ru) |
In the Russian market, NVIDIA initially priced the GeForce GTX 1660 at 17 rubles, but a month after the release of the new product, the prices of its various modifications moved away in both directions from the recommended value. The “premium” versions of the GTX 990, according to Yandex Market, reach 1660 rubles, and the most “popular” model can be bought for only 20 rubles. - this is the Inno240D GeForce GTX 15 Twin X975. But we did not find it on foreign trading floors, so private importers will have to consider other ways to save money.
Despite its excellent price-to-performance ratio, the Inno3D GeForce GTX 1660 Twin X2 has strong competition in the sub-$250 category. AMD met the debut of "green" new products with price cuts for the Radeon RX 580 and Radeon RX 590. The flagship accelerators on Polaris chips are now available at a price of 16 rubles, and the Radeon RX 261 in a configuration with 580 GB of RAM - from 8 rubles. These numbers are quite in line with how the competition was distributed in our first review of the GeForce GTX 12. However, AMD allows you to reduce upgrade costs even more with the Radeon RX 517 models with 1660 GB of memory, which start at 580 rubles.
Despite the cheapness of the Inno3D GeForce GTX 1660 Twin X2, something valuable was found in the package bundle of the video card - the keys to the paid versions of the 3DMark and VRMark test packages. But there are no other accessories, besides the software disk, in the box.
Design
Exterior Inno3D GeForce GTX 1660 Twin X2 does not need a detailed description. In appearance, the video card differs little from many other budget modifications of the GeForce GTX 10th and 16th series. Due to its modest size, you can decide that it fits the dimensions of Mini-ITX form factor systems, but no - the device is 1,96 cm longer than the prescribed 17 cm. Of course, there is no LED backlighting and other optional decorations. Even the protective plate on the back of the PCB is missing.
The insides of the Inno3D GeForce GTX 1660 Twin X2 are blown by two 72mm fans. Due to the fact that the manufacturer did not stretch the casing of the cooling system beyond the dimensions of the PCB, as is often done with compact video cards, there was no longer enough space for wider impellers. At a die temperature of 45°C, the cooling system operates in a strange semi-passive mode: the fans are stationary almost all the time, but every few seconds they come to life to make a few revolutions.
With low power consumption (120W based on GeForce GTX 1660 reference specs), the TU116 GPU is efficiently cooled by a compact, simple heatsink. However, Inno3D engineers have spared no expense for three 4mm thick heatpipes that are in direct contact with the GPU die. GIGABYTE, for example, in their budget versions of the GeForce GTX 1660 and GTX 1660 Ti cost one tube.
Printed circuit board
Since the TU116 and TU106 chips are interchangeable in terms of the physical interface, the mass of variants of the GeForce GTX 1660, 1660 Ti, and с are the same printed circuit boards. Manufacturers change only the GPU itself, the type and number of memory chips, and the configuration of the voltage regulator. Judging by external signs, all Inno3D Twin X2 models up to the GeForce RTX 2070 are also based on the same PCB. Curiously, it bears the NVIDIA label: it looks like Inno3D is purchasing Turing chips for this series of devices, along with reference boards.
The textolite has contact pads for the components of six GPU power phases and two power phases of RAM chips, but in the GeForce GTX 1660 only four and one phases are soldered, respectively. A modest configuration, but since the GeForce GTX 1060 version of the Founders Edition was enough for a total of four phases, then the younger model on the TU116 chip should be quite enough for five. Due to its kinship with the GeForce RTX series video cards, the Inno3D GeForce GTX 1660 Twin X2 voltage regulator contains field-effect transistors with an integrated driver (the so-called DrMOS, or “power stages” - power stages). Compared to discrete driver-transistor pairs, the integrated solution provides increased efficiency and allows the PWM controller to control the output voltage with high accuracy.
Additional power is supplied to the video card through a single eight-pin connector, and a set of external interfaces consists of three DisplayPort outputs and a single HDMI. Judging by the pad layout, the reference PCB allows you to replace two DisplayPort connectors with a DVI header, which would be quite appropriate for an inexpensive accelerator, but Inno3D did not take advantage of this opportunity.
Source: 3dnews.ru