Turing RTX 2080 and RTX 2080 Ti Benchmarked with 36 Games

Benchmarking Turing RTX 2080 and RTX 2080 Ti with 36 Games

BTR received a RTX 2080 Ti and a RTX 2080 from NVIDIA last week, and we have been testing them using 36 games versus the TITAN Xp, the GTX 1080 Ti, the GTX 1080, and also versus AMD’s flagship, the liquid-cooled Vega 64. In addition to bringing revolutionary new features including ray tracing to significantly improve visuals, NVIDIA promises that Turing will bring large performance increases over Pascal.

Since we got the drivers for the new cards on Friday night, this review has to focus on performance as well as consider whether the new RTX 2080 Ti Founders Edition at $1199 ($999 for entry-level AIBs) or the RTX 2080 Founders Edition at $799 ($699 from AIBs) represents good value as an upgrade from Pascal.

Turing uses architecture derived from Volta featuring Tensor cores for inference and added RT cores which can offer ray tracing in real time, and it represents a huge leap forward visually. AI denoising will make possible a new kind of deep learning supersampling (DLSS) AA that improves on TAA’s blurry images while bringing higher performance. Turing’s huge dies coupled with other advanced features including fast and overclockable GDDR6 together with improved memory compression, and a completely redesigned dual-fan Founders Edition come with a high materials cost.

The Pascal GTX 1080 ($699) and the GTX 1070 ($379) both launched in May 2016 as the mid-sized GP104. Later Nvidia launched the $1199 TITAN Xp as GP102, the big chip Pascal GPU with 12 GB of vRAM followed by a slightly cut-down 11 GB vRAM-equipped flagship GeForce gaming card, the GTX 1080 Ti for $699. The Turing RTX 2080 Ti is priced as an upgrade for the TITAN Xp, but it is also the only upgrade for GTX 1080 Ti owners.

We are testing using Windows 10 64-bit Home edition at 1920×1080, 2560×1440 and at 3840×2160 using Core i7-8700K at 4.7 GHz and 16 GB of HyperX 3333 MHz DDR4. All games are patched to their latest versions and we are using the latest drivers. But before we get to the unboxing, overclocking, and performance testing, we want to briefly cover Turing architecture, as well as detail the specifications and features of the new RTX 2080 Ti and RTX 2080.

Key Features of the Turing RTX 2080 and RTX 2080 Ti

Turing uses new hardware-based accelerators and hybrid rendering to combine rasterization and real-time ray tracing using tensor cores, plus AI using RT cores to accelerate NVIDIA’s next generation of PC graphics. Turing offers large increases in performance, memory bandwidth, and power efficiency over the current Pascal architecture and it has implications for advancing VR.

So far, the 18.6 billion transistor Turing TU102 GPU is only used for the new Quadro RTX 6000 and for the RTX 2080 Ti while the 13.6 billion transistor TU104 is used for the RTX 1080. The RTX 2080 and the 2080 Ti both use fast 14 GBits/s GDDR6 for high memory bandwidth, and new architectural refinements make it faster than Pascal.

NVIDIA has broken with their pattern by releasing their fastest gaming card first. NVIDIA says the RTX 2080 Ti is as fast or faster than the developer-oriented TITAN V at $2999. Created for ultra 4K, the RTX 2080 Ti is the highest performing Turing GeForce gaming card just as the GTX 1080 Ti is to the Pascal family although the hybrid TITAN Xp is faster.

To achieve 4K gaming, the Turing Streaming Multiprocessor (SM) delivers a 1.5X boost to shading efficiency performance per CUDA Core compared to Pascal. The Turing SM adds a new independent integer datapath, allowing integer and floating point operations to run concurrently. The SM memory path has been redesigned to unify shared memory, texture caching, and memory load caching into one unit for nearly double the bandwidth, but it offers more than double the increase of available capacity of L1 cache over Pascal.

Turing is built for 4K, ray tracing, AI, and especially for NVIDIA’s RTX. NVIDIA claims RTX is their greatest innovation since CUDA in 2006, and it will significantly improve PC game visuals. No longer is NVIDIA racing to deliver the same rasterized visuals at faster framerates, but they are dedicating a lot of valuable die space for hardware acceleration to deep learning/AI and to ray tracing.

Deep Learning Super-Sampling (DLSS)

Turing’s 114 TFLOP Tensor Cores are specialized execution units that performing tensor matrix operations that are the core compute function used in deep learning. Deep Learning SuperSampling (DLSS) leverages a deep neural network to extract features of a rendered scene and it accurately combines details from multiple frames to construct a high-quality image. Turing GPUs use half of the samples for rendering and then use AI to fill in data to create a final image. The result is a less blurry image with similar quality to TAA but with higher performance.

Here is a comparison taken from the soon-to-be-released Final Fantasy XV demo which runs a timedemo at maximum settings using either DLSS or TAA. The screenshots are similar but not identical. The screenshot using DLSS is on the left and TAA is on the right.

In some places the DLSS image looks a bit sharper although it is hard to tell from this example. The images are quite similar but the performance using DLSS is much higher than using TAA as the following chart shows.

By using NVIDIA’s supercomputer to train their neural network which scans a game’s images, DLSS becomes an easy integration for developers. So far, developers have announced that 25 games will support DLSS so that gamers will soon be able to have great DLSS anti-aliasing at a much lower performance cost than by using TAA.

Ray Tracing & RTX

Ray tracing has long been considered the holy grail of PC computer graphics and it is how many CGI feature movies simulate their lighting by using render farms to produce one frame at a time. By leveraging Turing’s RT cores with RTX software, NVIDIA can achieve ray tracing in real time for lighting and the use of AI for graphics and other applications. All together, these features are called Hybrid Rendering which is a combination of traditional rasterized and ray traced rendering, plus AI to produce game images that simply “look better”.

The next generation of ray traced games will not only look better but run faster on Turing RTX cards. NVIDIA has also developed a number of advancements for virtual reality by improving visual quality, acoustic simulation, and bringing a whole range of new tools to VR devs.

Turing GPUs also include hardware support for USB Type-C and VirtualLink, a new
open industry standard developed to meet the power, display, and bandwidth demands of
next-generation VR headsets through a single USB-C connector.

The RTX 2080 Ti

The GeForce RTX 2080 Ti is built on the TU102 Turing GPU and it is clearly designed for gamers who want the best. It delivers 4K HDR/60 FPS for most games and it replaces the TITAN Xp in all but name. It would also be the logical upgrade for a GTX 1080 Ti owner.

The GeForce RTX 2080 Ti has 4352 CUDA Cores, 68 RT Cores, 544 Tensor Cores, and runs at a GPU 4.0 Boost clock speed of 1545 MHz. The Founders Edition graphics card is factory overclocked by 90 MHz to GPU Boost clock speeds of 1635 MHz and will be sold for $1199 on the NVIDIA website beginning September 27, while the board partners will also offer their own stock and factory overclocked variants of the GeForce RTX 2080 Ti GPU starting at $999.

The RTX 2080 Ti uses 11 GB of 14 Gbps GDDR6 consisting of eleven 32-bit memory controllers (352-bit) providing peak memory bandwidth of 616 GB/sec, which is a 27% improvement over the Pascal GTX 1080 Ti.

The RTX 2080

The GeForce RTX 2080 is based on the Turing TU104 GPU which contains 13.6 Billion transistors, making it more complex than the 7.2 billion transistors used in the GTX 1080. The GeForce RTX 2080 is designed for gamers who want high-fidelity 4K gaming. It delivers 4K HDR at 60Hz similar in performance to a GTX 1080 Ti.

The GeForce RTX 2080 features 2944 CUDA Cores, 46 RT Cores, and 368 Tensor Cores, and ships with a GPU Boost clock speed of 1710 MHz. A Founders Edition overclocked by 90 MHz to 1800 MHz is offered directly from NVIDIA’s website beginning on September 20.

The memory subsystem of GeForce RTX 2080 uses eight 32-bit memory controllers (256-bit), complimented by 8GB of 14 Gbps GDDR6 memory which provides up to 448 GB/sec of peak memory bandwidth. There are 64 ROP Units with a peak Pixel Fill Rate of 109.4 Gigapixels/second. The GeForce RTX 2080 also has 184 Texture Units providing a Texture Rate of 314.6 Gigatexels/second.

The specifications, features, and the cards themselves look awesome, so let’s unbox them before overclocking and benching them.

Unboxing

Just like with the Founders Editions before them, the RTX 2080 and RTX 2080 Ti come in the same style of boxes which may be used to display the cards.

The contents and warranty information are on the bottom of each box. They have similar power requirements. The thick packing of the box and anti-static film protect the cards.

A forged and machine-finished aluminum shroud creates the new look for the RTX 2080 and RTX 2080 Ti Founders Edition to provides a premium and solid heavy feel to the industrial design.

A full-length vapor chamber covers the entire PCB of the graphics card providing a large surface area for cooling so the heat is readily transferred to a dual-slot aluminium fin stack which also covers the entire PCB. The dual-axial fans exhaust the heat out of the back and the sides of the card which may not work well for mini-ITX cases.

Instead of using a regular backplate, it is fully enveloped by the shroud, providing a cleaner look. The backplate rests on top of low-profile board components including the power circuitry and memory modules, and it provides extra cooling for them.

The RTX 2080 Ti uses two 8-pin connectors, and the RTX 2080 uses a 6-pin plus an 8-pin connector. A 13-phase iMON DrMOS power supply is used for the RTX 2080 Ti and an 8-phase unit is used for the RTX 2080 to provide clean power for overclocking.

These RTX Founders Editions are both beautiful cards. Before we install them and check out overclocking and noise, let’s look at our test setup.

Test Configuration – Hardware

• Intel Core i7-8700K (HyperThreading and Turbo boost is on to 4.7 GHz for all cores; Coffee Lake DX11 CPU graphics).
• EVGA Z370 FTW motherboard (Intel Z370 chipset, latest BIOS, PCIe 3.0/3.1 specification, CrossFire/SLI 8x+8x), supplied by EVGA
• HyperX 16GB DDR4 (2×8 GB, dual channel at 3333 MHz), supplied by HyperX
• RTX 2080 8 GB Founders Edition, stock FE clocks, supplied by NVIDIA
• RTX 2080 Ti 11 GB Founders Edition, stock FE clocks, supplied by NVIDIA
• TITAN Xp 12 GB Star Wars Edition, stock clocks, supplied by NVIDIA
• GTX 1080 Ti 11 GB Founders Edition, stock clocks, supplied by NVIDIA
• GTX 1080 8 GB Founders Edition, stock clocks, supplied by NVIDIA
• GTX 1070 8 GB Founders Edition, stock clocks, supplied by NVIDIA
• GTX 980 4 GB, Founders Edition stock clocks, supplied by NVIDIA
• Gigabyte Vega 64 Liquid-cooled edition 8GB, at factory overclocked AIO clocks
• PowerColor Fury X 4 GB, at stock clocks
• 2 x 480 GB Team Group SSDs – one for AMD, and one for NVIDIA
• 1.92 TB San Disk enterprise class SSD
• 2 TB Micron 1100 enterprise class SSD
• Seasonic 850W Gold Focus power supply unit
• EVGA CLC 280mm CPU water cooler, supplied by EVGA
• Onboard Realtek Audio
• Genius SP-D150 speakers, supplied by Genius
• EVGA DG-77, mid-tower case supplied by EVGA
• LG 43″ HDR 4K TV
• Monoprice Crystal Pro 4K

Test Configuration – Software

• Nvidia’s GeForce 411.51 drivers used for the RTX 2080/Ti. 399.24 is used for the GTX 1080 Ti and for the TITAN Xp, and 399.07 for the GTX 1080. 398.36 WHQL drivers are used for the GTX 980 Ti and for the GTX 1070. See Control Panel image below.
• AMD Adrenalin Software 18.9.1 drivers were used for the Vega 64 and 18.7.1 was used for the Fury X. See Control Panel image below.
• VSync is forced off.
• AA enabled as noted in games; all in-game settings are specified with 16xAF always applied
• Gaming results show average frame rates in bold including minimum frame rates shown on the chart next to the averages in a smaller italics font.
• Highest quality sound (stereo) used in all games.
• Windows 10 64-bit Home edition. All DX11 titles were run under DX11 render paths. DX12 titles are generally run under the DX12 render path unless performance is worse than with DX11. Three games use the Vulkan API.
• Latest DirectX
• All 36 games are patched to their latest versions at time of publication.
• WattMan used to set overclocking, cooling and power options. See Control Panel image below.
• MSI’s Afterburner, used to adjust GTXes and to set temp and power limits to maximum.
• PrecisionX1 pre-release version for RTX settings and for overclocking
• OCAT, latest version
• Fraps, latest version

36 PC Game benchmark suite & 4 synthetic tests

Synthetic

• Firestrike – Basic & Extreme
• Time Spy DX12

DX11 Games

• Grand Theft Auto V
• The Witcher 3
• Fallout 4
• Watch Dogs 2
• Just Cause 3
• Rainbow Six Siege
• Battlefield 1
• Resident Evil 7
• For Honor
• Ghost Recon Wildlands
• Mass Effect: Andromeda
• Prey
• DiRT 4
• Project CARS 2
• Middle Earth: Shadow of War
• Assassin’s Creed Origins
• Destiny 2
• Call of Duty WW II
• Star Wars: Battlefront II
• Final Fantasy XV
• Far Cry 5
• The Crew 2

DX12 Games

• Tom Clancy’s The Division
• Ashes of the Singularity: Escalation
• Hitman
• Rise of the Tomb Raider
• Deus Ex Mankind Divided
• Gears of War 4
• Civilization VI
• Sniper Elite 4
• Total War: Warhammer II
• Forza 7
• Shadow of the Tomb Raider

Vulkan Games

• DOOM
• Wolfenstein: The New Colossus
• Strange Brigade

Additional Demo

• FFXV standalone Demo – DLSS vs. TAA

AMD Adrenalin Control Center Settings:

All AMD settings are set so as to be apples-to-apples when compared to NVIDIA’s control panel settings – all optimizations are off, Vsync is forced off, Texture filtering is set to High, and Tessellation uses application settings.

We used Wattman to set the liquid-cooled Gigabyte Vega 64’s power, temperature and fan settings to their maximums.

NVIDIA Control Panel settings:

Here are the NVIDIA Control Panel settings that match AMD’s settings.

We used MSI’s Afterburner for all GTXes to set their power and temperature limits to maximum but we used the as yet unreleased EVGA PrecisionX1 scanner for adjusting RTX card settings. See the overclocking section.

By setting the Power Limits and Temperature limits to maximum for each card, they do not throttle, but they can each reach and maintain their individual maximum clocks. This is particularly beneficial for high power cards including the Vega 64 liquid-cooled edition.

Let’s check out overclocking, temperatures and noise next.

Overclocking using PrecisionX1, temperatures & noise

The RTX 2080 and RTX 2080 Ti are both very quiet cards even when overclocked. We could not hear either card over the fans of our system even when they ramp up. We will spend more time with overclocking in our follow up review, but we did manage to overclock both of our RTXes using NVIDIA’s new automatic scanner in the form of PrecisionX1 which is not yet available to the public.

The above image shows the RTX 2080 Ti at idle on the desktop. The voltage seems to be a bit high with or without temperature and power targets set to maximum. We benchmark our video cards with these targets at maximum to unleash the cards full performance potential at stock clocks.

We used NVIDIA’s new scanner and it found it to be quite accurate compared with the scanner that released with Pascal. To find the best automatic overclock, click on the Scan button and wait for about 20 minutes. You can also test the stability of your overclock by using the Test button.PrecisionX1 decided that 1950 to 1980 MHz was the maximum safe overclock, set it automatically, and gave us a score of 193.

We tried out own manual overclock and were able to add 200 MHz to the core to peg the clocks mostly at 2000 MHz after we pushed the voltage slider to maximum with peak Boost around 2050 MHz. We also were able to add +800 MHz to the RTX 2080 Ti’s memory clocks(!)

Our preliminary RTX 2080 Ti overclock is +180 MHz offset to the core and +800 MHz to the memory. Our final overclock will be explored in Turing, Part 2, later this week. In the meantime, we tested 10 games at 4K resolution to see if our overclock has made any performance difference.

We see several games where an overclock may help to improve a game’s fluidity at 4K. It appears from our very limited time with RTX 2080 Ti that overclocking is good and similar to Pascal overclocking, but it appears that a steady 2000 MHz is close to the limit for Turing with peak Boost around 2050 MHz. We also like NVIDIA’s new scanner, and EVGA’s new PrecisionX1 is much more stable than when it released for Pascal.

We also used PrecisionX1 to overclock the RTX 2080 with similar good results. It gave us a score of 115 and a top boost of 2025-2040 MHz.

Our preliminary overclock for the RTX 2080 is +115 MHz offset to the core and +700 MHz to the memory. Temperatures for the RTX 2080 remained cool in the 70s C range although the RTX 2080 Ti would reach into the low-80s C under maximum load and during the highest overclock in the PrecisionX1 scan. The noise from the RTX cooling fans never became noticeable over the other PC case fans.

Let’s check out performance after we look at our test configuration on the next page.