GeForce GTX 1660 SUPER Video Card Review: Titan Fighter

От Yraa #1080p, #1200W, #120W, #1440p, #3DMark, #8K resolution, #ability, #access, #action, #activate, #active, #Adobe Premiere Pro, #afterburner, #ahead, #air, #Alpha, #Alpha and Omega, #alternative, #Amazon, #AMD, #AMD Radeon Software, #Ammeter, #among, #announced, #announcement, #API, #app, #appeared, #application, #ARC, #arrived, #Ashes of the Singularity, #Assassin, #Assassin's Creed Odyssey, #Asus, #attempt, #backlight, #balanced, #Ban, #Band, #Base, #based, #battle, #Battlefield V, #beautiful, #Bee, #Beta, #bit, #Black, #board, #body, #bought, #box, #brand, #brands, #Budget, #built, #buy, #Byte, #called, #card, #cards, #case, #cat, #category, #cell, #CES, #characteristics, #chart, #cheap, #cheaper, #check, #chip, #Choose, #class, #clock, #closed, #closer, #closes, #Codec, #Coffee, #Coming, #Compact, #company, #Comparison, #compete, #complete, #conclusions, #configuration, #connect, #connectors, #cons, #Control, #Cool, #cooler, #Cooling, #Core, #Cores, #couple, #created, #Crysis 3, #CUDA, #Curve, #Cute, #Data, #date, #day, #deal, #debut, #Deco, #delay, #DERE, #design, #developed, #Developer, #Device, #devices, #differences, #Direct3D, #DirectX, #Dirt Rally 2.0, #Disco, #discount, #display, #DisplayPort, #Dollar gap, #drive, #dual, #Ear, #early, #ease, #Easy, #economical, #Edge, #Edition, #efficiency, #ELE, #electric, #Elf, #Energy, #engine, #enthusiast, #entire, #EVE, #EVO, #Excellent, #Execution unit, #expected, #expensive, #experience, #Express, #Extra, #Extreme, #face, #factory, #fall, #Far Cry 5, #fast, #faster, #features, #FFmpeg, #Fighter, #finally, #Finding, #Fire, #flagship, #FLOPS, #formula, #Forza Horizon 4, #Frame, #Frame rate, #frames, #Free, #Full, #fully, #function, #functionality, #future, #G.Skill, #game, #gamers, #games, #gaming, #GDDR5 SDRAM, #GDDR6 SDRAM, #GeForce, #GeForce 16 series, #GeForce 20 series, #Gen, #GENE, #generation, #GHz, #Giga, #GIGABYTE, #good, #Grand Theft Auto V, #Graphic, #graphics, #Graphics card, #Graphics pipeline, #ground, #GTH, #GTX, #hack, #hands, #hard, #hardware, #HDMI, #heat, #Heat pipe, #Heat sink, #hero, #high, #Highest, #honest, #Honor, #Hot, #house, #huge, #ideal, #image, #Impeller, #Initially, #Inno3D, #Integrated, #Intel, #Intel Core, #interesting, #interface, #interfaces, #Introduced, #IPS, #king, #Laptop, #large, #latest, #launch, #lead, #LED, #level, #library, #light, #lighting, #limit, #line, #Lineup, #list, #lit, #Live, #long, #longer, #LSS, #LTE, #luxury, #main, #man, #manufacturers, #Mark, #market, #Master, #Mate, #Matrix multiplication, #Max, #MAXIMUS, #measure, #Media, #Meet, #memory, #Metro Exodus, #micro, #Microarchitecture, #Microsoft, #Mind, #mini, #mode, #model, #models, #modern, #modes, #Monitor, #monitoring, #month, #months, #MOSFET, #Motherboard, #move, #MSI, #N-body simulation, #NEC, #Neo, #Neural network, #Newegg, #News, #NEX, #Noise, #Note, #Nova, #novelty, #number, #NVIDIA, #obvious, #offer, #Official, #officially, #Open, #OpenCL, #operating, #Operating system, #OPPO, #option, #order, #Overclocking, #Pack, #Pass, #pay, #PCI, #PCI Express, #Pen, #Perfect, #Performa, #performance, #phases, #place, #plans, #platform, #play, #playing, #power, #Power supply, #powerful, #premiere, #preparing, #present, #previous, #price, #prices, #Printed circuit board, #PRO, #processor, #processors, #product, #production, #productive, #products, #project, #projects, #Puget Systems, #Pure, #quality, #Quest, #quiet, #Radeon, #RAID, #RAM, #rate, #rated, #rating, #ray, #read, #reader, #readers, #real, #Reality, #receive, #received, #record, #Recording, #Red, #reference, #Refresh, #release, #releases, #remaining, #representative, #rest, #Results, #review, #RGB, #rice, #rid, #ring, #river, #ROG, #roll, #RTX, #Ruble, #Russia, #Russian, #Sale, #SAPPHIRE, #save, #screen, #segment, #Semiconductor, #series, #set, #Set-top box, #Setting, #settings, #severe, #Shader, #Shadow, #Shadow of the Tomb Raider, #shipments, #showed, #significantly, #silence, #SIM, #simultaneous, #single, #slot, #software, #solution, #source, #special, #specifications, #speed, #ssd, #stage, #stand, #store, #story, #strange, #Strange Brigade, #strong, #study, #style, #success, #successful, #suitable, #Super, #supply, #support, #Surface, #survive, #system, #Tab, #tag, #Taiwan, #tasks, #technology, #temperature, #terms, #test, #testing, #tests, #Thermocouple, #thin, #time, #Titan, #Titanium, #Tom Clancy's The Division, #Tom Clancy's The Division 2, #Tomb Raider, #Top, #Total war, #train, #Transcoding, #Transistor, #turn, #turns, #Twin, #UEFI, #Ultra, #unpretentious, #update, #updated, #upgrade, #User, #users, #Vega, #version, #Versions, #video, #Video codec, #View, #virtual, #visible, #vision, #Volta, #Voltage regulator, #VP9, #Vulkan, #wait, #waiting, #warm, #watts, #ways, #Wind, #window, #windows, #Windows 10, #work, #working, #Yandex, #year, #Years, #Zen


In the fourteen months that have passed since the introduction of the first GeForce RTX family of graphics cards, NVIDIA’s engineering and marketing team has been more prolific than ever before. If we put together all the mainstream models under the GeForce 16 brand and the proposals for gamers-enthusiasts of the GeForce 20 series, but do not take into account their laptop varieties (and among the latter there is a division into Max-Q and versions unofficially called Max-P), then a total of thirteen devices. Of course, the company managed to withdraw some of them from production without waiting for the upcoming silicon base upgrade, and nevertheless, the average rate of release of new products is approaching a record level of one per month (NVIDIA, push it!).

The previous explosion in the GeForce discrete graphics card catalog came with the RTX 2060, RTX 2070, and RTX 2080 SUPER, which was ultimately fueled by rival company’s successful releases, the Radeon RX 5700 and RX 5700 XT. But NVIDIA is sure that the purpose of the SUPER set-top box has not yet been fully fulfilled. It’s time to refresh another branch of accelerators based on Turing chips — the GeForce GTX 1650 and GTX 1660, because AMD has already announced the specifications of the upcoming mainstream model called the Radeon RX 5500, and NVIDIA prefers to act proactively. The SUPER version of the GTX 1650 will not be available for sale until November 22 — then we will publish its review, but the GTX 1660 SUPER is already in our hands.

⇡#Specifications, prices

Although this article focuses on the older of the two new products, you should first pay attention to the GeForce GTX 1650 SUPER, so as not to be distracted and focus on the updated GTX 1660. The original version of the GeForce GTX 1650, whatever one may say, became the weak link in the entire chain of Turing family for some time slowed down the positive trend in the growth of the ratio of performance and price of NVIDIA discrete GPUs. Despite the excellent energy efficiency of the GeForce GTX 1650, it hopelessly lags behind its closest red rival, the Radeon RX 570, in terms of speed, not to mention the modest amount of 4 GB of RAM. It seems that NVIDIA was preparing the GeForce GTX 1650 more for OEM customers than for retail. But now that AMD is preparing to launch a promising replacement for the Radeon RX 570/580 in the face of the Radeon RX 5500, NVIDIA’s position in the budget price segment will not survive due to energy efficiency alone.

This is how the GeForce GTX 1650 SUPER was born, based on the TU116 GPU (instead of the TU117), but with an extremely stripped-down configuration of functional blocks, including a reduced ROP set and a 128-bit RAM bus. As a result, the GTX 1650 SUPER will still not get rid of the 4 GB RAM limit, but otherwise, theoretical performance estimates allow us to count on an increase of 53% compared to the original version of the accelerator. Moreover, instead of GDDR5 chips with a bandwidth of 8 Gb / s, he will receive brand new GDDR6 12 Gb / s. In addition to the increased frame rate in games, an important factor for some users will be the full-fledged integrated NVENC video encoder that the TU116 has, while the TU117 got a similar, less powerful unit from Volta chips.

NVIDIA keeps pricing information for the GeForce GTX 1650 SUPER under wraps, and AMD doesn’t disclose its plans for the Radeon RX 5500 in advance. If we recall how the situation with the price of the Radeon RX 5700 and Radeon RX 5700 XT developed after the announcement of the GeForce GTX 2060 and 2070 SUPER , NVIDIA and AMD can again play an interesting multi-move, but we will not see this until at least November 22.

Now back to the hero of today’s review. Let’s not beat around the bush, analyzing its technical characteristics, because all that has undergone a change in the GeForce GTX 1660 SUPER is the type and bandwidth of RAM. In the past, NVIDIA has done a number of mid-term upgrades of a product with accelerated RAM when the right GDDR chips are in commercial shipments and at the same time there is a compelling market conjuncture. So, in the specifications of the GeForce GTX 1060, an option once appeared with GDDR5 chips with a bandwidth of 9 Gb / s instead of 8 Gb / s, and the GTX 1080 received an upgrade from 10 to 11 Gb / s per pin. However, in neither case did the video card with the updated RAM configuration receive a separate position in the NVIDIA lineup.

An exception was made for the GeForce GTX 1660 SUPER for a reason, because this time the GDDR5 8 Gb / s chip set is changed immediately to GDDR6, and NVIDIA did not stint on fast chips with a bandwidth of 14 Gb / s, which are used in all older accelerators based on Turing chips up to to TITAN RTX. Thus, the GeForce GTX 1660 SUPER has at its disposal a larger total memory bandwidth (336 GB / s) than the GTX 1660 Ti, and compared to the base model, it has grown by 75%. The remaining parameters of the TU116 chip in the SUPER version of the GeForce GTX 1660 remained the same, including the base and boost frequencies. Only the power consumption range had to be expanded from 120W to 125W to match the power consumption of the GDDR6 array.

If NVIDIA decided to give odds to the GeForce GTX 1660 in anticipation of the imminent debut of AMD’s RDNA-based video cards, then the RAM upgrade was the ideal and, in fact, the only possible solution. In addition to the type of RAM, the GeForce GTX 1660 Ti differs from «just a GTX 1660» only in the presence of an additional 128 shader ALUs, which, after correcting for the lower clock speeds of the older model, gives 8% of the theoretical performance. Under such task conditions, it is impossible to squeeze any new model between the GTX 1660 and Ti by manipulating one GPU configuration, but the performance of RAM even for the TU116 chip means a lot. Like it or not, the Ti is ahead of its stripped-down counterpart in FPS gaming tests by 20%but not by 8%. At the same time, won’t everything lead to the fact that the updated GTX 1660 will reach the Ti level in terms of performance, since the theoretical difference between the full and partially blocked configurations of the TU116 is so insignificant, and the memory bandwidth of the SUPER is even higher? Looking ahead, we admit that by and large it is so.

Manufacturer NVIDIA
Model GeForce GTX 1650 GeForce GTX 1650 SUPER GeForce GTX 1660 GeForce GTX 1660 SUPER GeForce GTX 1660 Ti
Name TU117 TU116 TU116 TU116 TU116
microarchitecture Turing Turing Turing Turing Turing
Process technology, nm 12 nm FFN 12 nm FFN 12 nm FFN 12 nm FFN 12 nm FFN
Number of transistors, million 4700 6 600 6 600 6 600 6 600
Clock frequency, MHz: Base Clock / Boost Clock 1485/1665 1530/1725 1530/1785 1530/1785 1500/1770
Number of shader ALUs 896 1280 1408 1408 1536
Number of texture overlays 56 80 88 88 96
Number of ROPs 32 32 48 48 48
Number of tensor cores Not Not Not Not Not
Number of RT cores Not Not Not Not Not
Bus width, bit 128 128 192 192 192
Clock frequency, MHz (bandwidth per contact, Mbps) 2000 (8000) 1 500 (12 000) 2000 (8000) 1,750 (14,000) 1 500 (12 000)
Volume, MB 4096 4096 6 144 6 144 6 144
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
Peak performance FP32, GFLOPS (based on maximum specified frequency) 2984 4570 5027 5027 5437
Performance FP64/FP32 1/32 1/32 1/32 1/32 1/32
Performance FP16/FP32 2/1 2/1 2/1 2/1 2/1
RAM bandwidth, GB/s 128 192 192 336 288
Image Output
Image output interfaces 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 DL DVI-D, DisplayPort 1.4a, HDMI 2.0b
TBP/TDP, W 75 one hundred 120 125 120
Retail price (USA, without tax), $ 149 (recommended at the time of release) ND 219 (recommended) 229 (recommended) 279 (recommended)
Retail price (Russia), rub. ND ND 17,990 (recommended) 17,990 (recommended) 22 990 (recommended)

So, has the GeForce GTX 1660 SUPER come to replace the Ti, like the SUPER versions of the GeForce RTX 2070 and RTX 2080? But no — both previous models are not discontinued and will coexist in the market with the GTX 1660 SUPER, embarrassing technically uneducated buyers. The pretentious SUPER index is certainly more ingrained in the mind than the restrained Ti, but a closer look at the price list should raise the question of why the SUPER is cheaper. NVIDIA has announced a MSRP of $229 for the US market, $50 less than the GTX 1660 Ti. On the other hand, the SUPER only costs $10 more than the base GeForce GTX 1660.

The situation on the Russian market, as always, is more complicated than in the US, where at any time (with the exception of such anomalies as the shortage of AMD video cards during the crypto boom), you can find a video card on Amazon or Newegg with a price tag exactly corresponding to the recommended price without tax for sale. In our latitudes, NVIDIA is going to sell the GeForce GTX 1660 SUPER for an amount of 17,990 rubles, and there is no mistake in the plate with specifications and recommended prices — that’s how much the original GTX 1660 should have cost. However, since the spring debut of the GeForce 16 series, its representatives have pretty much fallen in price : according to, the simplest version of the GeForce GTX 1660 can be bought for 13,462 rubles, and the GTX 1660 Ti — for 16,470. So the GTX 1660 SUPER has already dropped below its recommended price to 15,974 rubles and will probably continue to move downward along as the local market becomes saturated with offers.

Regardless of the level at which the price of the GeForce GTX 1660 SUPER stabilizes, the new product feels at ease in the huge gap that the competitor left between the Radeon RX 590 and the Radeon RX Vega 56. The first is now available for 13,115 rubles, and the second for 19,210, but if the difference between the GTX 1660 and SUPER comes in line with the US market, by then AMD will need to respond to NVIDIA’s challenge.

This is what the GeForce GTX 1660 SUPER looks like in general terms — the latest addition to the swollen lineup of NVIDIA gaming accelerators, in fact, an alternative to the GTX 1660 Ti at a discount. Let’s get acquainted with the video card, which has the honor to present the new model to the readers.

⇡#GIGABYTE GeForce GTX 1660 SUPER OC Construction

It so happened that all NVIDIA accelerators from the GeForce 16 series first appeared on the pages of 3DNews in the guise of GIGABYTE, and only then we took versions of the same models from other manufacturers for comparison. However, we are not complaining: the operating parameters of video cards — power consumption, temperature and clock speeds — are more convenient to study when the design of the cooling system is the same. So the newly arrived GeForce GTX 1660 SUPER is no different from the related GTX 1660 and GTX 1660 Ti, in addition to memory chips of a different type and, as we will see in the next section of the review, a set of voltage regulator components.

The GIGABYTE catalog contains relatively expensive and sophisticated modifications of the 16-series video cards, but once again we got a product with the most simplified design, which, however, fully meets the moderate demands of the TU116 chip for cooling quality and allowed us to endow this version of the GeForce GTX 1660 SUPER with factory overclocking at 45 MHz according to the declared Boost Clock.

The unpretentious exterior of the GIGABYTE GeForce GTX 1660 SUPER OC lacks the luxury features of more established brands, such as LED backlighting (not to mention RGB with adjustable hue), LED strip power connectors and external fans. From all sides, except for the plate with video interfaces, the video card is covered by a black plastic casing. This is done to protect delicate PCB components from accidental damage, as well as to hide the tail of the cooling system protruding beyond the compact printed circuit board.

The GIGABYTE GeForce GTX 1660 SUPER OC cooling system itself is served by two impellers with a diameter of 87 mm, rotating in opposite directions — thus reducing the turbulence of the air flow. Below them lies a radiator of an extremely simplified design, although GIGABYTE put all the checkmarks in front of the key design points: the radiator fins are type-setting, not milled, there is a heat pipe (one), and their own heat sinks are adjacent to the RAM chips and VRM components. There is probably nothing more to add to the results of the external examination of the accelerator. Earlier, we made sure that GIGABYTE’s chosen TU116 chip binding provides the GeForce GTX 1660 Ti with quite comfortable working conditions, which means it will also suit the GTX 1660 SUPER.


The printed circuit board, which underlies the gigabyte GeForce GTX 1660 SUPER, also appeared before us for the third time. But the list of devices in which GIGABYTE uses it is actually even wider and extends from the original version of the GTX 1660 to the RTX 2070, because the TU116, TU106 and TU104 chips are electrically compatible with each other. As a result, the layout of the board is made with the expectation of a fairly powerful voltage regulator: up to six phases under the control of the uPI Semiconductor uP9512R PWM controller. The memory is powered by a two-phase uPI Semiconductor uP1666Q chip.

Due to the fact that the TU116 does not require a complete six-phase voltage regulator, two past modifications of the GeForce GTX 1660 from GIGABYTE — the regular one and the Ti — managed with four phases. But, surprisingly, in the GTX 1660 SUPER, the manufacturer found a way to save even more: there are only three power stages (Alpha and Omega AOZ5332QI) in the GPU power circuit with a rated current of 50 A. If we take a rough correction for the power consumption of memory chips and thermal losses , then the GPU of the video card at a supply voltage of about 1 V can within the standard TDP 125 Wset the VMR load to 100 A. This will certainly not affect the power quality so much as to cause unstable operation (there must certainly be a negative effect on voltage constancy), and serious overclocking of NVIDIA video cards is still closed due to software volt mod restrictions. Nevertheless, this is perhaps the first video card in our practice in recent years with such a meager power reserve of the main voltage regulator (although we know that some of the Taiwanese follow the example of GIGABYTE), and, accordingly, the low efficiency of the latter. Depending on the PWM frequency of the GeForce GTX 1660 SUPER, you can expect 11-12W of heat loss in the GPU power supply under full load (according to AOZ5332QI specifications). It is a pity that the API of green accelerators does not give monitoring programs data on the temperature of VRM keys, so that we can find out how hot they are without a thin thermocouple in the gap between the heatsink and the board.

It’s good that at least with the power supply of RAM chips everything is in order here: a pair of dual MOSFETs Alpha and Omega AON6994 with a nominal value of 26 A with an excess is enough to service six GDDR6 chips. Micron chips, marked with the line 9PA77 D9WCW, operate at their nominal 14 Gb / s bandwidth.

⇡#New NVIDIA Driver Features

Finally, before the performance tests of the GeForce GTX 1660 SUPER, let’s digress from the hardware and devote time to the new features of NVIDIA software, which, admittedly, also arose under the influence of AMD, like the SUPER prefix in the names of new GeForce models. To coincide with the launch of the Radeon RX 5700 and RX 5700 XT, rival NVIDIA timed an update to the Radeon Software package with two functions — at first glance, very simple, but in the end so noticeable that the green team hastened to integrate their full counterparts into the GameReady Driver.

Radeon Software has introduced a Radeon Anti-Lag option that AMD claims can minimize the lag between user interaction and game response on the monitor. NVIDIA now has a similar «technology» NVIDIA Ultra-Low Latency (although this option has been present in the control panel of GeForce video cards from time immemorial, only without a beautiful name), but in reality, both of them just open access to setting the length of the frame queue, which the CPU can prepare in advance for rendering on the GPU in the Direct3D 9 and 11 software graphics pipeline. Direct3D 12 and Vulkan games determine the queue length on their own and, as a result, do not obey the Anti-Lag and NULL instructions. In addition, a short frame queue does not guarantee short delays in all cases. If system performance is limited by the CPU, not the GPU, it will only get worse, but we suggest that owners of NVIDIA accelerators find out in practice whether to enable NULL or entrust API frame queue management. At least the latest driver 441.07 allows you to do this without abandoning G-SYNC, as was the case in previous versions of the GameReady package.

Another innovation — a post-processing filter that increases sharpness — migrated from the FreeStyle interface (part of the GeForce Experience application) to the general NVIDIA control panel. And most importantly, now the filter performs not only sharpening by enhancing local contrast, but also scaling the frame to the screen resolution, when you need to save on viewport resolution in order to maintain performance in demanding games. The function works in games under the last three versions of Direct3D and is compatible not only with Turing chips, but also with graphics processors of previous generations. However, only Turing can ennoble the picture with a 5-tap filter (borrowed from video codecs) on top of the standard bilinear scaling.

In theory, video cards of the GeForce RTX series do not need all this, because there is DLSS technology, which, with due diligence on the part of game developers, hides the difference between honest rendering at full resolution and a stretched picture, but we have seen no less examples of a mediocre implementation of a scaling neural network than successful, and accelerators without tensor cores are denied access to DLSS in any case.

And finally, you can now drop filters created for the ReShade utility into the FreeStyle proprietary interface. The latter, created long before FreeStyle, has a similar functionality — injecting a post-processing step into the game’s graphics pipeline — but thanks to NVIDIA support, filters are activated at the driver level, rather than by gentle hacking of individual applications. FreeStyle can use any filter from the extensive library built by users over the years of ReShade’s existence, but competitive multiplayer games are severely limited to only about three dozen NVIDIA-approved filters.

⇡#Test stand, testing methodology

test stand
CPU Intel Core i9-9900K (4.9GHz, 4.8GHz in AVX, fixed frequency)
RAM G.Skill Trident Z RGB F4-3200C14D-16GTZR, 2 x 8 GB (3200 MHz, CL14)
ROM Intel SSD 760p, 1024 GB
Power Supply Corsair AX1200i 1200W
CPU cooling system Corsair Hydro Series H115i
Frame CoolerMaster Test Bench V1.0
Monitor NEC EA244UHD
Operating system Windows 10 Pro x64
AMD GPU software
All video cards AMD Radeon Software Adrenalin 2019 Edition 19.10.2
NVIDIA GPU software
All video cards NVIDIA GeForce Game Ready Driver 441.08
Synthetic 3D Graphics Benchmarks
Test API Permission Full screen anti-aliasing
3D Mark Fire Strike 1.1 DirectX 11 (feature level 11_0) 1920×1080 Off
3DMark Fire Strike 1.1 Extreme 2560×1440
3DMark Fire Strike 1.1 Ultra 3840×2160
3D Mark Time Spy 1.1 DirectX 12 (feature level 11_0) 2560×1440
3DMark Time Spy Extreme 1.1 3840×2160
Game tests
Game (in order of release date) API Settings, test method Full screen anti-aliasing
1920×1080 / 2560×1440 3840×2160
Grand Theft Auto V DirectX 11 Built-in benchmark. Max. graphics quality MSAA 4x + FXAA + Reflection MSAA 4x FXAA
Ashes of the Singularity: Escalation Vulkan Built-in benchmark. Max. graphics quality MSAA 4x + TAA Ultra Off
Far Cry 5 DirectX 11 Built-in benchmark. Max. graphics quality TAA TAA
Strange Brigade Vulkan Built-in benchmark. Max. graphics quality AA Ultra AA Ultra
Shadow of the Tomb Raider DirectX 12 Built-in benchmark. Max. graphics quality SMAA 4x SMAAT 2x
Forza Horizon 4 DirectX 12 Built-in benchmark. Max. graphics quality MSAA 4x + FXAA FXAA
Assassin’s Creed Odyssey DirectX 11 Built-in benchmark. Max. graphics quality AA High (TAA) AA High (TAA)
Battlefield V DirectX 12 OCAT, Liberte mission. Max. graphics quality. DXR off, DLSS off TAA High TAA High
Metro Exodus DirectX 12 Built-in benchmark. Max. graphics quality. DXR off, DLSS off, Shading Rate 100% TAA TAA
DiRT Rally 2.0 DirectX 11 Built-in benchmark. Max. graphics quality MSAA 4x + TAA TAA
Tom Clancy’s The Division 2 DirectX 12 Built-in benchmark. Max. graphics quality AA Ultra (TAA) AA Ultra (TAA)
Total War: THREE KINGDOMS DirectX 12 Built-in benchmark (Battle Benchmark). Max. graphics quality TAA TAA

In most test games, the average and minimum frame rates are derived from the render time array of individual frames, which is recorded by the built-in benchmark (or the OCAT utility, if not available).

The average frame rate in the charts is the reciprocal of the average frame time. To estimate the minimum frame rate, the number of frames generated in each second of the test is calculated. From this array of numbers, the value corresponding to the 1st percentile of the distribution is selected.

The exceptions to this technique are three games with a built-in benchmark — DiRT Rally 2.0, Far Cry 5 and Forza Horizon 4. The Far Cry 5 benchmark itself records the number of frames per second of the test to a file — the average FPS value and the 1st percentile are calculated from of these numbers, not by the rendering time of individual frames. The DiRT Rally 2.0 benchmark records only the average frame rate and the minimum frame rate calculated from the maximum frame time. Forza Horizon 4 does not output any performance parameters to a file and also reports average and minimum frame rate totals.

General purpose computing, video encoding/decoding
Program Settings
Adobe Premiere Pro CC 2019 (Puget Systems Adobe Premiere Pro CC Benchmark) 4K Heavy GPU Effects, live playback (ProRes 422 59.94 FPS) OpenCL CUDA
4K Heavy GPU Effects, export to H.264 40 Mbps UHD and ProRes 422HQ UHD (59.94 FPS)
Blender 2.8 Beta (Cycles) Classroom Demo
CompuBench 2.0 Ocean Surface Simulation OpenCL CUDA
N-Body Simulation 1024K
DXVA Checker 4.1.2, Decode Benchmark H.264 1920 × 1080 (High Profile, L4.1), 3840 × 2160 (High Profile, L5.1). Microsoft H264 Video Decoder
H.265 1920 × 1080 (Main Profile, L4.0), 3840 × 2160 (Main Profile, L5.0), 7680 × 4320 (Main Profile, L6.0). Microsoft HEVC Video Extensions
VP9 1920×1080, 3840×2160, 7680×4320. Microsoft VP9 Video Extensions
FFmpeg 4.0.2, H.264 encoding 1920×1080 -c:v h264_amf -quality speed -coder cabac -level 4.1 -refs 1 -b:v 3M -c:v h264_nvenc -preset fast -coder cabac -level 4.1 -refs 1 -b:v 3M
3840×2160 -c:v h264_amf -quality speed -coder cabac -level 5.1 -refs 1 -b:v 7.5M -c:v h264_nvenc -preset fast -coder cabac -level 5.1 -refs 1 -b:v 7.5M
FFmpeg 4.0.2, H.265 encoding 1920×1080 -c:v hevc_amf -quality speed -level 4 -b:v 3M -c:v hevc_nvenc -preset fast -level 4 -b:v 3M
3840×2160 -c:v hevc_amf -quality speed -level 5 -b:v 7.5M -c:v hevc_nvenc -preset fast -level 5 -b:v 7.5M
7680×4320 -c:v hevc_nvenc -preset fast -level 6 -refs 1 -b:v 20M
Lux Mark 3.1 Hotel Lobby (Complex Benchmark)
SiSoftware Sandra Titanium (2018) SP3b GPGPU Processing (FP16/FP32/FP64) OpenCL CUDA
GPGPU Scientific Analysis (FP16/FP32/FP64)

The power of video cards is recorded separately from the CPU and other PC components using the JUNTEK VAT-1050 ammeter. In order to simultaneously measure the current passing through the additional power connectors and the motherboard slot, the video card is connected through a PCI Express x16 hard riser, in which the power lines are broken and brought out to a separate cable.

The test load for power and noise tests is Crysis 3 at 3840 x 2160 resolution without FSAA and maximum graphics quality settings, as well as the FurMark stress test at the most aggressive settings (3840 x 2160 resolution, MSAA 8x). Measurements of all parameters are performed after the video card warms up, when the GPU temperature and clock speeds stabilize.

⇡#Test participants

The following video cards took part in testing:

  • GIGABYTE GeForce GTX 1660 Ti OC (1500MHz, 12000Mbps, 6GB);
  • GIGABYTE GeForce GTX 1660 SUPER OC (1530MHz, 14000Mbps, 6GB);
  • GIGABYTE GeForce GTX 1660 OC (1530 MHz, 8000 Mbps, 6 GB);
  • Inno3D GeForce GTX 1660 Twin X2 (1530 MHz, 8000 Mbps, 6 GB);
  • AMD Radeon RX Vega 56 (1590 MHz, 1600 Mbps, 8 GB);
  • SAPPHIRE NITRO+ Radeon RX 590 SE (1545 MHz, 8000 Mbps, 8 GB).

⇡#Clock speeds, power consumption, temperature, noise and overclocking

Unlike the older Turing-based graphics cards — the GeForce RTX 2060 and above — there are no reference samples in the GeForce 16 series that would allow comparison of performance, power consumption and clock speeds, all other things being equal (although some partners use NVIDIA-branded reference boards). However, we were lucky enough to collect the necessary data within the same GIGABYTE product line. Yes, all these cards have factory overclocking, but, as it should be for budget implementations of this or that model, moderate enough so that we can extend our observations of the main operating parameters to similar products.

The spread in clock frequencies between the three modifications of the GeForce GTX 1660 ranges from 1870 to 1935 MHz, so the performance in benchmarks will depend entirely on other, more significant discrepancies — the configuration of the shader ALUs and the bandwidth of the RAM. The power consumption of both versions of the GeForce GTX 1660 and GTX 1660 Ti also remains virtually unchanged at stock clock speeds — from 107 to 111 MHz in resource-intensive gaming tests. The scale on the corresponding graph is set by rival AMD video cards — the Radeon RX 590 and Vega 56 — whose appetites are exactly twice as large as that of any member of the GeForce GTX 1660 family.

Conclusions about the properties of the cooling system, in turn, relate specifically to GIGABYTE boards and can say little about the requirements of various modifications of the TU116 chip to the video card piping. In budget accelerators, this firm puts emphasis on GPU temperature (between 63-66 ° C under load) rather than noise level, and as a result, the three devices turned out to be quite vociferous for their modest power (36-38 dBA). However, in absolute terms, these are quite acceptable results — especially against the background of the reference Vega 56, which among competing accelerators is closest to the «vanilla» GeForce GTX 1660, Ti and SUPER. The Radeon RX 590 in the NITRO + modification, on the contrary, is extremely quiet despite the high power consumption — it’s all about a high-quality cooler, so the partner versions of the Vega 56 may well be quieter under load than the simple GTX 1660 SUPER versions.

Load Performance (Crysis 3)
video card Settings GPU clock frequency, MHz GPU supply voltage, V Fan speed, rpm (% of max.) Fan speed 2, rpm (% of max.)
Avg. Max. Limit Avg. Max. Avg. Avg.
GIGABYTE GeForce GTX 1660 SUPER OC (1530MHz, 14000Mbps, 6GB) 1887 1890 1965 0.998 1.012 1988 (71%) ND
GIGABYTE GeForce GTX 1660 SUPER OC (1630MHz, 15500Mbps, 6GB) +20% TDP 1995 1995 2055 1.031 1.031 2147 (77%) ND
GIGABYTE GeForce GTX 1660 OC (1530MHz, 8000Mbps, 6GB) 1935 1935 1995 1.031 1.031 1915 (69%) ND
GIGABYTE GeForce GTX 1660 Ti OC (1500MHz, 12000Mbps, 6GB) 1870 1905 1980 0.939 0.975 1950 (64%) ND
SAPPHIRE NITRO+ Radeon RX 590 SE(1545MHz, 8000Mbps, 8GB) WattMan: Balanced, Silent UEFI 1543 1545 1545 1.140 1.181 1252 (28%) ND
AMD Radeon RX Vega 56 (1590MHz, 1600Mbps, 8GB) WattMan: Balanced 1312 1319 1590 0.940 1.075 1868 (38%) ND

Note: all parameters are measured after the GPU has warmed up and the clock speeds have stabilized.

To prepare the gigabyte GeForce GTX 1660 SUPER for overclocking, you need to increase the TDP of the video card to the allowed limit of 120%. The GPU voltage control on the GTX 1660 SUPER is still closed in MSI Afterburner and similar utilities, and, as we have repeatedly seen, it does not really work on Turing chips. Without a software volt mod, most modern NVIDIA video cards allow you to increase the base core frequency by 100–150 MHz, and with it, shift the entire curve connecting the positions of frequencies and voltages upwards. The GeForce GTX 1660 SUPER was no exception: the «extra» 100 MHz Base Clock resulted in an increase in the stable GPU frequency in the gaming test from 1887 to 1995 MHz. In addition to this, GDDR6 memory chips with a nominal bandwidth of 14 Gb / s agreed to operate at 15.5 Gb / s — again, a typical overclocking result for all NVIDIA cards with a similar RAM configuration.

The power consumption of the overclocked GeForce GTX 1660 SUPER increased from 107 to 124 watts under load in Crysis 3. It would seem quite a bit, but the simplified GIGABYTE cooling system with a single heat pipe is forced to respond with increased fan speed and, as a result, the noise level is on the verge of 40 dBA. Whatever one may say, budget video cards do not favor overclocking, and now we will find out whether it is really profitable to exchange silence for increased frequencies in such cases.

⇡#Game tests (1920×1080)

Before the advent of the GTX 1660 SUPER, when the GeForce 16 family had only two video cards based on the TU116 chip, differing at the same time in the configuration of the GPU execution units and the type of RAM, it was not easy to find out how much the bandwidth of the GDDR5 chips was holding back the performance of the younger model.

Now that the GTX 1660 SUPER has better RAM chips than the GTX 1660 Ti, the memory bandwidth factor is clearly visible in the gaming benchmarks. It turns out that replacing GDDR5 with GDDR6 can increase the performance of the TU116 by as much as 15%. When both the older models of the GeForce 16 series — the GTX 1660 SUPER and the Ti — have plenty of bandwidth to spare, the difference in the composition of the shader ALUs allowed the Ti to add only 4% to the average SUPER frame rate in a dozen test games. It is not surprising that the overclocking of the GeForce GTX 1660 SUPER — especially so insignificant without software voltmod — is expressed in the same 4% increase in performance.

Among video cards based on AMD chips for the GeForce GTX 1660 SUPER, you can’t find an equal opponent yet. The Radeon RX 590 competes with the base GTX 1660 but is no match for the GTX 1660 SUPER (24% faster on average). Radeon RX Vega 56, in turn, outperforms the new product by 9% in terms of average frame rate.

⇡#Game tests (2560×1440)

In the results of gaming benchmarks at a resolution of 1440p, we did not find significant differences from what the test participants have already demonstrated in 1080p mode — for such video cards, the notorious processor dependence can be safely deleted from the formula when the comparison is carried out on a powerful eight-core platform with CPU frequencies tending to 5 GHz . However, let’s summarize the findings.

The transition to a higher screen resolution has most of all affected the ratio between the base modification of the GeForce GTX 1660 with GDDR5 memory and the new GTX 1660 SUPER equipped with GDDR6 chips. The gap increased by only 2% — from 15% to 17% — but the PSP factor in this case should have played a role. The GTX 1660 SUPER and Ti are 3% apart in average frame rates in 12 games, but the gap between the GTX 1660 SUPER is still easy to overcome with overclocking.

AMD GPUs based on GCN architecture can be expected to perform better than similar NVIDIA products when the screen resolution increases, but the difference in load between 1080p and 1440p is still not large enough for the Radeon RX 590 to regain its lost ground forever: GeForce GTX The 1660 SUPER outperforms the older Polaris by 23% FPS. On the other hand, the more productive Vega 56 retains a 9% advantage over the green novelty.

The GeForce 16 series graphics cards are positioned as low-cost gaming solutions for gaming at 1080p, and the 1440p benchmark results suggest that they really should not go into more demanding modes than Full-HD — at least with maximum quality settings charts. Only sparing AAA-class projects (Battlefield V, Forza Horizon 4, Far Cry 5 and Strange Brigade) allowed the GeForce GTX 1660 SUPER to rise above 60 FPS. In most others, it dropped into the range from 30 to 60, and without a sure margin of minimum frame rates, and the most demanding benchmarks (Ashes of the Singularity, Metro Exodus and Total War: THREE KINGDOMS) did not even allow the minimum program to be executed (30 FPS) . To be fair, the Vega 56 — a more powerful accelerator that was once designed for resolutions up to 4K — no longer feels comfortable in 1440p tests that reflect modern realities. Suitable performance should be sought in products of a higher price category: GeForce RTX 2060 or Radeon RX 5700.

⇡#General purpose computing

Graphics cards like the GeForce GTX 1660 SUPER are generally not considered as general-purpose computing accelerators for demanding workloads: in such use cases, both a powerful GPU and a suitable array of RAM are required. However, modern gaming accelerators, even in the mass market category, are comparable in performance and RAM to flagship products of past generations, and not all work tasks have become more resource-intensive since then. In addition, all Turing chips share a number of optimizations, thanks to which NVIDIA has managed to push the traditionally strong solutions of the GCN architecture on the performance graphs of work applications.

So, in the Blender and LuxMark rendering tests, all three modifications of the GeForce GTX 1660 are ahead of the Radeon RX 590. And, characteristically, the best result belongs to the GTX 1660 SUPER — obviously, due to the fact that this model has the highest memory bandwidth compared to the other two. But the Radeon RX Vega 56 is still out of competition here.

Premiere Pro is initially better optimized for NVIDIA accelerators, and even Vega 56 loses to the GTX 1660 triad in all aspects of performance, in addition to exporting the project to H.264 format. However, none of the GeForce GTX 1660 models can reproduce 4K sources in full resolution even at 24 FPS, especially the original GTX 1660. The SUPER and Ti versions in Premiere Pro are significantly faster and almost equivalent in performance.

In physics simulation tests, the GTX 1660 SUPER once again showed a significant advantage over the original product and came out on par with the GTX 1660 Ti. The performance of one of the competing video cards — the Radeon RX 590 — is highly dependent on the nature of the task, but the leading position of the Radeon RX Vega 56 is again beyond doubt.

The isolated matrix multiplication rate benchmark in SiSoftware Sandra is pure «synthetics» that perfectly matches, with rare exceptions, the theoretical GPU bandwidth in gigaflops. As a result, both the Radeon RX 590, and even more so the Vega 56, are ahead of any modifications of the GeForce GTX 1660 in standard accuracy calculation points (FP32) (the difference between the latter is rather insignificant on a general scale). In addition, by the standards of consumer devices, AMD GCN-based video cards perform well in niche tasks related to operations on double-precision real numbers (FP64). But for some reason, this SiSoftware Sandra benchmark cannot effectively use the dedicated array of FP16 blocks in the TU116 chip, so the results of the GTX 1660/SUPER/Ti turned out to be four times less than their theoretical capabilities.

Sandra’s compute shader test, which is more complex than isolated GEMM loading, placed the Radeon RX 590 between the GTX 1660 and 1660 Ti, but Vega 56 was once again at the top of the chart. But here the TU116 chip showed its high half-precision performance (FP16), while the Radeon RX 590 is only able to process FP16 at the same speed as FP32.

The overall result of two Sandra tests is that SUPER turned out to be the weakest among the three modifications of the GTX 1660. The lag behind the GTX 1660 Ti is due to the reduced configuration of shader ALUs, and from the base model, apparently, the relatively high latency of the new type of RAM.

⇡#Video encoding/decoding

There are practically no differences between the GeForce GTX 1660, SUPER and Ti in terms of video encoding and decoding speed — after all, they have the same set of specialized NVDEC and NVENC blocks. The performance of the TU116 in these tasks is more than enough for playing movies at resolutions up to 8K, transcoding and streaming in high (by the standards of integrated ASICs) quality.

AMD, in turn, is upgrading a similar component of its GPUs with a long delay. The Radeon RX 590 and Vega 56 can take over the decoding of 4K H.264 and HEVC recordings at high frame rates, but both cards are weak in converting and streaming due to the low bandwidth of the built-in encoder. In addition, both are denied access to 8K resolution, and they do not have the ability to decode the VP9 format without the help of shader ALUs.

⇡#Gaming Performance Test Results and Pricing

3DMark (Graphics Score)
Permission NVIDIA GeForce GTX 1660 SUPER (1530MHz, 14000Mbps, 6GB) NVIDIA GeForce GTX 1660 SUPER (1630MHz, 15500Mbps, 6GB) NVIDIA GeForce GTX 1660 (1530MHz, 8000Mbps, 6GB) NVIDIA GeForce GTX 1660 Ti (1500MHz, 12000Mbps, 6GB) AMD Radeon RX 590 (1545MHz, 8000Mbps, 8GB) AMD Radeon RX Vega 56 (1590MHz, 1600Mbps, 8GB)
fire strike 1920×1080 16295 17160 14011 16740 16235 20229
Fire Strike Extreme 2560×1440 7385 7779 6314 7511 7048 9383
Fire Strike Ultra 3840×2160 3212 3395 2769 3320 3582 4770
time spy 2560×1440 6082 6523 5328 6297 4719 6270
Time Spy Extreme 3840×2160 2789 2984 2359 2839 2085 3040
Max. +7% −12% +4% +12% +49%
Avg. +6% −14% +3% −8% +22%
Min. +5% −15% +2% −25% +3%
Full screen anti-aliasing NVIDIA GeForce GTX 1660 SUPER (1530MHz, 14000Mbps, 6GB) NVIDIA GeForce GTX 1660 SUPER (1630MHz, 15500Mbps, 6GB) NVIDIA GeForce GTX 1660 (1530MHz, 8000Mbps, 6GB) NVIDIA GeForce GTX 1660 Ti (1500MHz, 12000Mbps, 6GB) AMD Radeon RX 590 (1545MHz, 8000Mbps, 8GB) AMD Radeon RX Vega 56 (1590MHz, 1600Mbps, 8GB)
Ashes of the Singularity: Escalation MSAA 4x + TAA Ultra 20 / 31 22/34 18/26 22/33 19/28 23/37
Assassin’s Creed Odyssey TAA High 43/51 45 / 52 43 / 49 47 / 55 35 / 39 44 / 54
Battlefield V TAA High 97 / 107 86 / 108 76 / 93 93 / 112 76 / 91 99 / 114
DiRT Rally 2.0 MSAA 4x + TAA 45/65 47/68 40/58 47/68 42/51 59 / 74
Far Cry 5 TAA 83 / 95 86 / 96 74 / 82 89 / 99 74 / 82 95 / 106
Forza Horizon 4 MSAA 4x + FXAA 73 / 87 75 / 88 65 / 78 77 / 91 57 / 69 81 / 96
Grand Theft Auto V MSAA 4x + FXAA + Reflection MSAA 4x 44/71 46 / 73 37/60 45/73 33/47 45 / 64
Metro Exodus TAA 19/32 19/34 17/28 19/33 16/27 22/39
Shadow of the Tomb Raider SMAA 4x 44/55 43/57 38 / 48 45/57 32/43 38 / 53
Strange Brigade AA Ultra 89 / 116 95 / 123 77 / 97 93 / 118 71 / 96 90 / 128
Tom Clancy’s The Division 2 TAA Ultra 47 / 58 50 / 61 43/52 50 / 60 43/52 56 / 68
Total War: THREE KINGDOMS TAA 27/35 28/36 23/29 28/35 19/26 28/37
Max. +10% −4% +8% −10% +22%
Avg. +4% −13% +4% −19% +9%
Min. +1% −17% 0% −34% −10%
Full screen anti-aliasing NVIDIA GeForce GTX 1660 SUPER (1530MHz, 14000Mbps, 6GB) NVIDIA GeForce GTX 1660 SUPER (1630MHz, 15500Mbps, 6GB) NVIDIA GeForce GTX 1660 (1530MHz, 8000Mbps, 6GB) NVIDIA GeForce GTX 1660 Ti (1500MHz, 12000Mbps, 6GB) AMD Radeon RX 590 (1545MHz, 8000Mbps, 8GB) AMD Radeon RX Vega 56 (1590MHz, 1600Mbps, 8GB)
Ashes of the Singularity: Escalation MSAA 4x + TAA Ultra 19/26 19/28 16/22 18/27 16/25 21/32
Assassin’s Creed Odyssey TAA High 37 / 42 38 / 44 34 / 37 37 / 42 29/33 36/41
Battlefield V TAA High 64/81 66 / 82 53/68 67 / 83 55 / 65 74 / 90
DiRT Rally 2.0 MSAA 4x + TAA 36/48 37 / 48 32/42 38 / 50 31 / 38 43/52
Far Cry 5 TAA 59/65 60 / 68 51 / 56 61 / 68 51 / 56 65 / 74
Forza Horizon 4 MSAA 4x + FXAA 56 / 66 56 / 67 49/58 58 / 69 44/53 60 / 72
Grand Theft Auto V MSAA 4x + FXAA + Reflection MSAA 4x 33/50 34/52 28/41 34/51 24/33 33/46
Metro Exodus TAA 16/26 17/28 14/23 17/27 13/22 19/32
Shadow of the Tomb Raider SMAA 4x 29/35 30 / 36 25 / 30 30 / 36 22/28 27/35
Strange Brigade AA Ultra 69 / 84 73 / 89 56 / 69 70 / 85 54/68 69 / 91
Tom Clancy’s The Division 2 TAA Ultra 34 / 40 36 / 42 30 / 35 35/41 31 / 35 41 / 47
Total War: THREE KINGDOMS TAA 19/24 20/25 16 / 20 20/24 15/18 20 / 26
Max. +8% −12% +5% −4% +23%
Avg. +4% −14% +3% −19% +9%
Min. 0% −18% 0% −34% −8%


The performance and price charts above characterize the GeForce GTX 1660 SUPER better than a lengthy discussion of the current state of the mainstream gaming graphics card market. The green novelty, by and large, has become an alternative version of the GeForce GTX 1660 Ti. As a consequence, everything we’ve already said about the Ti’s performance applies equally to the SUPER: it’s a product for gaming at 1080p with enough speed to keep the coveted 60 FPS in most modern AAA projects, and enviable power efficiency at against the background of the competitor’s solutions closest in price and capabilities. However, even if the Radeon RX 590 is not a rival to the new product — the performance gap is too large — the Radeon Vega 56 looks like a good alternative (especially if you take into account the additional 2 GB of RAM), because the next step in the NVIDIA lineup is the GeForce RTX 2060, which costs much more.

From a practical point of view, for a budget gamer, there is no difference in favor of which of the two older GTX 1660 models to choose, and individual differences between NVIDIA partner products — in terms of cooling quality and noise level — are more important than the additional 4% performance that the GTX 1660 Ti has. . Moreover, the minimum prices of SUPER and Ti differ by the cost of a couple of cups of coffee. However, SUPER still has a chance to get cheaper, because there are not many new video cards in Russian retail yet, and the prices recommended by NVIDIA for the US still provide for a 50-dollar gap between them. Then both the older and younger GTX 1660 models will simply be superfluous on the market: the first does not have a sufficient performance advantage, and the second, on the contrary, is too slow compared to SUPER.

If we ignore attempts to restore order in the modern GeForce catalog — NVIDIA systematically complicates this task — then, as in the case of RTX video cards, the SUPER prefix says one thing: performance per ruble is growing. And the reason for the changes this time is probably the same. AMD has announced a budget Radeon RX 5500 accelerator that will take on the GeForce GTX 1650 rather than the GTX 1660, but it’s likely that NVIDIA knows something about the competitor’s plans that we don’t know yet. Whatever one may say, AMD needs to close the gap between the Radeon RX 590 and the Radeon RX Vega 56, which houses the GTX 1660 SUPER. Judging by the fact that NVIDIA hastened to make a proactive move, it will become clear in the near future how this story will end.

As for the specific implementation of the new product — GIGABYTE GeForce GTX 1660 SUPER OC — thanks to which we got acquainted in practice with the latest initiative of the greens, then, like the previous models of the Taiwanese company in the GeForce 16 series that passed through the 3DNews laboratory, this one is one of the simplest implementations GTX 1660 SUPER. It cools the TU116 chip well and is pretty quiet for the recorded GPU temperature, but it’s time to stop saving on GIGABYTE voltage regulator components despite the economical orientation of the GTX 1660 SUPER.


От Yraa

Добавить комментарий