RTX 3070 Ti Performance
Performance Summary Charts & Graphs
Gaming Performance Summary Charts
Here are the summary charts of 25 games and 3 synthetic tests. The highest settings were always chosen and the settings are listed on the chart. The benches were run at 1920×1080, 2560×1440 and at 3840×2160. Five cards were compared and they are listed in order starting from left to right with the RTX 3070 FE, the reference RX 6800, the RTX 3070 Ti, the RX 6800 XT, the RTX 3080 FE, the RTX 3080 Ti FE, the RTX 3090 FE, and the Red Devil RX 6900 XT (which was benchmarked in April).
Most results, except for synthetic scores, show average framerates, and higher is better. Minimum framerates are next to the averages in italics and in a slightly smaller font. Games benched with OCAT show average framerates, but the minimums are expressed by frametimes (99th-percentile) in ms where lower are better. Performance wins between the RTX 3070 Ti and the RX 6800 are given in yellow text.
Please click on each chart to open a pop-up window for its best viewing experience.
Although there is some game-dependent variability, the RTX 3070 Ti FE is only around 3-10% faster than the RTX 3070 FE but it is enough to now easily trade blows with the reference RX 6800 in rasterized games, winning more than it loses, and is much faster in most ray traced games and a lot faster when DLSS is used.
Next we look at overclocked performance.
These benchmarks are run with the RTX 3070 Ti overclocked +150MHz on the core and +800MHz on the memory versus at stock clocks. The RTX 3070 Ti overclocked results are presented first and the stock results are shown in the second column.
There is a small performance increase from overclocking the RTX 3070 Ti Founders Edition. Unfortunately, although we did not have enough time to optimize our overclock, it’s clear that NVIDIA has locked down Ampere cards’ overclocking in an attempt to maximize performance for all Founders Edition gamers. We would also suggest that the RTX 3070 Ti FE is rather voltage constrained and if you want a higher overclock, pick a factory-overclocked partner version instead of a Founders Edition.
Let’s next look at VR gaming with the Vive Pro 2. The following is not our review of the Vive Pro 2 – the full review will follow next week. Instead we are going to focus on performance.
VR Gaming with the Vive Pro 2
The Vive Pro 2 is a much more demanding headset than the Vive Pro or the Valve Index by virtue of its higher resolution. Image resolution has been increased per eye from the Pro’s (or Valve Index’) 1440 x 1600 to 2448 x 2448. This higher resolution gives it exceptional clarity with no screen door effect, but it is also demanding on video cards. By default at the Ultra or Extreme preset, the Vive console uses 150% SteamVR Render Resolution for the Vive Pro 2 which appeared to be set to 2748×2748 per eye for high end NVIDIA cards at the time we benchmarked our games.
Here is the OpenVR benchmark result which requires 100% SteamVR Render Resolution for its default run. We used the Vive Console Ultra setting at native resolution and 90Hz. We did not test the Extreme setting which allows up to 120Hz.
Although SteamVR sets the same resolution for the RTX 3090 and the RTX 3070 Ti, it uses a lower resolution for AMD cards at either 100% (2244×2244) or at 150%. In fact, yesterday’s Vive software update lowered the default SteamVR resolution slightly for NVIDIA cards which suggests that it is still a work in progress and is being fine-tuned. The 100% SteamVR render resolution was lowered from 2556×2556 to 2532×2532 yesterday. Our results reflect the higher render setting.Some VR gamers prefer to lower the SteamVR Render Resolution which is set at 150% and is mostly used to compensate for the lens’ distortion instead of lowering a game’s preset or by dropping individual settings. We decided to initially test at 100% which is what we test the Reverb G2, the Vive Pro, and the Valve Index. Our follow up review will also benchmark at the default 150% resolution.
Yesterday, in response to our questions, Vive suggested that the SteamVR default Render Resolution should be left at 150%. Vive told BTR:
“Motion Compensation is the same as Motion Smoothing. The new lens and display requires our own motion compensation, and VIVE Console is the software that is driving the displays, so motion compensation is built into that.
For VIVE Pro 2, we set Steam’s supersampling setting as 150% by default, which makes up for the lens distortion. We found this to be the best value for SteamVR’s automatic performance scaling to scale and still reach 90 or 120 Hz on the majority of PCs we expect to be used to run VIVE Pro 2. However, users can still go into SteamVR to manually adjust their supersampling settings.
If we had set it to 100%, a lot of PCs would struggle under automatic settings. Render resolution is set by SteamVR and automatically scales to what it thinks is best for your system, VIVE Console handles display resolution.”
Motion Smoothing is disabled in SteamVR, but we actually didn’t see any FPS performance difference disabling or enabling Motion Compensation in the Vive console using fpsVR although the frametimes suffered. We see relatively minor visual differences between 100% and 150% SteamVR Render Resolution but even at the higher setting, lens distortion is still slightly visible to us particularly at the edges of the display.
At 50% SteamVR Render Resolution, there is a clear degradation of visuals which indicates that the SteamVR Render Resolution is working properly. However, at 150% Super Resolution, the frametime rates go up (which is bad) for several games that we tested although the FPS remain at 45 FPS which suggested to us that Vive’s Motion Compensation may still be on although Vive assures us it can be switched off in their console. We noticed that Motion Compensation artifacting became prominent and even disturbing if settings are pushed too high as we found with Elite Dangerous.
Please note that FCAT VR doesn’t distinguish dropped frames from synthesized frames using the Pro 2 (or the Reverb G2) like it properly does for the Valve Index and the Vive Pro. We suggest that the vast majority of the frames reported as dropped are actually synthetically generated (reprojected) frames. It is likely that FCAT VR is not yet optimized for the Pro 2.
It is important to remember that BTR’s charts use frametimes in ms where lower is better, but we also compare “unconstrained framerates” which shows what a video card could deliver (headroom) if it wasn’t locked to either 90 FPS or to 45 FPS by the HMD. In the case of unconstrained FPS which measures just one important performance metric, faster is better.
Let’s individually look at our five VR games’ performance using FCAT VR. All of our games were benchmarked at 100% SteamVR resolution.
First up, Assetto Corsa Competizione.
Assetto Corsa Competizione (ACC)
BTR’s sim/racing editor, Sean Kaldahl created the replay benchmark run that we use for both the pancake game and the VR game. It is run at night with 20 cars, lots of geometry, and the lighting effects of the headlights, tail lights, and everything around the track looks spectacular.
The RTX 3070 Ti delivered 102.85 unconstrained FPS with 15 dropped or synthesized frames and no Warp misses.
The RTX 3070 Ti has a little performance headroom and it is possible to play it using enhanced individual settings with minimal reprojected or synthesized frames but it is best suited for playing ACC on VR Low. VR High is unplayable.
Next, we check out Elite Dangerous.
Elite Dangerous (ED)
A player will probably spend a lot of time piloting his space cruiser while completing a multitude of tasks as well as visiting space stations and orbiting a multitude of different planets (~400 billion). Elite Dangerous is also co-op and multiplayer with a very dedicated following of players.
We picked the Ultra Preset with the maximum FoV originally but the shimmering and artifacting from reprojection/Motion Compensation was awful, so we set everything to Medium leaving the FoV at maximum. Here is the frametime plot.
The RTX 3070 Ti delivered 128.79 unconstrained FPS with no Warp Misses nor any dropped or synthetic frames.
The experience playing Elite Dangerous at Ultra settings is awful but Medium seems perfect with some performance headroom to increase individual settings.
Next, we will check out a really demanding VR game, No Man’s Sky.
No Man’s Sky
No Man’s Sky is an action-adventure survival single and multiplayer game that emphasizes survival, exploration, fighting, and trading. It is set in a procedurally generated deterministic open universe, which includes over 18 quintillion unique planets using its own custom game engine.
The player takes the role of a Traveller in an uncharted universe by starting on a random planet with a damaged spacecraft equipped with only a jetpack-equipped exosuit and a versatile multi-tool that can also be used for defense. The player is encouraged to find resources to repair his spacecraft allowing for intra- and inter-planetary travel, and to interact with other players.
We set the settings to Enhanced which is above Low and below High, but we also set the anisotropic filtering to 16X and upgraded to FXAA+TAA. The game has recently implemented DLSS 2.1 and we used the highest visual quality preset, Quality which gives a much smaller performance boost than the others DLSS settings.
Here is the No Man’s Sky Frametime plot.
The RTX 3070 Ti produced 85.37 unconstrained FPS with no dropped frames or Warp misses, but it required 3200 (50%) synthetic frames.
The Low Preset may be better suited for play with the RTX 3070 Ti, or else individual setting may be lowered to maintain a balance of performance to visuals. However, it may be best to use DLSS Performance instead and accept a slight artifacting. We were very impressed with the Enhanced preset using DLSS Quality, and the high resolution screen of the Vive Pro 2 makes playing this game an even more extraordinary experience where the game comes more alive.
Let’s continue with another demanding VR game, Project CARS 2, that we still like better than its successor.
Project CARS 2 (PC2)
There is a real sense of immersion that comes from playing Project CARS 2 in VR using a wheel and pedals. It uses its in-house Madness engine, and the physics implementation is outstanding. We are disappointed with Project CARS 3, and will continue to use the older game instead for VR benching.
Project CARS 2 offers many performance options and settings and we prefer playing with SMAA rather than to use MSAA.
We originally tried maximum settings including for Motion Blur but that wasn’t possible so we set everything to Medium.
The RTX 3070 Ti delivered 77.49 unconstrained FPS with 4802 (50%) synthesized or dropped frames and with no Warp misses.
The experience playing Project CARS 2 on the Medium preset requires that we would recommend lowering individual settings or even lower the resolution a as needed to stay out of reprojection. However, even on Medium, the game looks great using the Vive Pro 2.
Let’s benchmark Skyrim VR.
Skyrim VR is an older game that is no longer supported by Bethesda, but fortunately the modding community has adopted it. It is not as demanding as many of the newer VR ports so its performance is still very good on maxed-out settings using its Creation engine.
We benchmarked Skyrim VR using its highest settings, but we did not increase its in game supersampling.
The RTX 3070 Ti managed 130.68 unconstrained FPS with no dropped frames, no synthetic frames, and no Warp misses.
The RTX 3070 Ti can play Skyrim at its maxed out in-game settings although we did not benchmark in-game Supersampling since we saw reprojecting or synthesized frames. Since there is some performance headroom, it suggests to us that mods may be used with the Vive Pro 2 and a RTX 3070 Ti class of video card.
These benchmarks results bring up more questions than answers that we hope to cover in a follow up review dedicated to the Vive Pro 2 next week. However, we love the Pro 2 and have ordered our own headset and will keep it for future VR benchmarking.
To see if the RTX 3070 Ti is a good upgrade from the other video cards we test workstation, creative, and GPGPU benchmarks starting with Blender.
Blender 2.92 Benchmark
Blender is a very popular open source 3D content creation suite. It supports every aspect of 3D development with a complete range of tools for professional 3D creation.
We benchmarked three Blender 2.92 benchmarks which measure GPU performance by timing how long it takes to render production files. We tested seven of our comparison cards with both CUDA and Optix running on the GPU instead of using the CPU. We benchmarked the RX 6800 XT and the RTX 3080 using OpenCL because Radeons do not support CUDA.
Here are the RTX 3070 Ti’s CUDA and OPTIX scores.
Blender’s benchmark performance is slower using the RTX 3070 Ti compared with the RTX 3080 and slightly faster than te RTX 3070.
Next we look at the OctaneBench.
OctaneBench allows you to benchmark your GPU using OctaneRender. The hardware and software requirements to run OctaneBench are the same as for OctaneRender Standalone.
We run OctaneBench 2020.1.5 for Windows and here are the RTX 3070 Ti’s complete results with an overall score of 454.87.
Here is the summary chart comparing our five GeForce cards. Radeons cannot run the Octane benchmark.
The RTX 3070 Ti is a decent card when used for rendering but closer to the RTX 3070 in performance than the RTX 3080.
Next, we move on to AIDA64 GPGPU benchmarks.
AIDA64’s benchmark code methods are written in Assembly language, and they are well-optimized for every popular AMD, Intel, NVIDIA and VIA processor by utilizing the appropriate instruction set extensions. We use the Engineer’s full version of AIDA64 courtesy of FinalWire. AIDA64 is free to to try and use for 30 days.
Here is the chart summary of the AIDA64 GPGPU benchmarks with seven of our competing cards side-by-side.
The RTX 3070 Ti is a fast GPGPU card that is slightly faster than the RTX 3070. So let’s look at Sandra 2020 next.
To see where the CPU, GPU, and motherboard performance results differ, there is no better tool than SiSoft’s Sandra 2020. SiSoftware SANDRA (the System ANalyser, Diagnostic and Reporting Assistant) is a excellent information & diagnostic utility in a complete package. It is able to provide all the information about your hardware, software, and other devices for diagnosis and for benchmarking. Sandra is derived from a Greek name that implies “defender” or “helper”.
There are several versions of Sandra, including a free version of Sandra Lite that anyone can download and use. Sandra 2021 R2 is the latest version, and we are using the full engineer suite courtesy of SiSoft. Sandra 2020/21 features continuous multiple monthly incremental improvements over earlier versions of Sandra. It will benchmark and analyze all of the important PC subsystems and even rank your PC while giving recommendations for improvement.
We ran Sandra’s intensive GPGPU benchmarks and charted the results summarizing them. There was a bug in one Processing benchmark that affected the Red Devil RX 6800 XT with OpenCL that was addressed by SiSoft by the time we tested the RX 6800.
In Sandra GPGPU benchmarks, the RTX 3070 Ti is similar in performance to the RTX 3070. Interestingly, the RTX 3070 Ti (and RTX 3080 Ti’s) Hashing bandwidth is much lower than the RTX 3080/RTX 3070 and even the RX 6800 XT as NVIDIA has limited its cryptocurrency mining ability. However, since the architectures are different, each card exhibits different characteristics with different strengths and weaknesses.
All the SPECworkstation3 benchmarks are based on professional applications, most of which are in the CAD/CAM or media and entertainment fields. All of these benchmarks are free except for vendors of computer-related products and/or services.
The most comprehensive workstation benchmark is SPECworkstation3. It’s a free-standing benchmark which does not require ancillary software. It measures GPU, CPU, storage and all other major aspects of workstation performance based on actual applications and representative workloads. We only tested the GPU-related workstation performance as checked in the image above.
Here are the SPECworkstation3 results summarized in a chart along with six competing cards. Higher is better.
Using SPEC benchmarks, the RTX 3070 Ti is closer in performance to the RTX 3070 than it is to the RTX 3080. However, since the architectures are different, the cards each exhibit different characteristics with different strengths and weaknesses.
SPECviewperf 2020 GPU Benches
The SPEC Graphics Performance Characterization Group (SPECgpc) has released a 2020 version of its SPECviewperf benchmark that features updated viewsets, new models, support for both 2K and 4K display resolutions, and improved set-up and results management.
We benchmarked at 4K and here is the summary for the RTX 3070 Ti.
Here are SPECviewperf 2020 GPU benchmarks summarized in a chart together with six other cards.
Again the RTX 3070 Ti is slightly faster than the RTX 3070 but not close to RTX 3080 performance.
After seeing these benches, some creative users may wish to upgrade their existing systems with a new RTX 30X0 series card based on the performance increases and the associated increases in productivity that they require. The question to buy an RTX 3070 Ti should be based on the workflow and requirements of each user as well as their budget. Time is money depending on how these apps are used. However, the target demographic for the RTX 3070 Ti is primarily gaming for gamers, especially at 1440P and at 1080P.
Let’s head to our conclusion.