When approaching testing methodology, I prefer to emulate an environment where most consumers will utilize their hardware (i.e. in a case on their desk), in addition to stressing the components to provide a worst case scenario. This is done to provide context so that consumers may understand the worst performance they may expect in a relatively common environment. In other words: how hard can I push this card under relatively common environmental factors?
Here are the components used for this review.
- CPU: Intel i7 8700K (OC 5 GHz)
- GPU: This is the testing variable and will be compared against the GTX 1080 Ti FE for comparisons.
- RAM: 16GB G.Skill TridentZ DDR4 @ 3200 MHz
- CPU Cooler: Corsair H100i v2
- Motherboard: Gigabyte Z370 Aorus Gaming 7
- Chassis: Corsair Carbide 400C
- PSU: Corsair RM 750i
- SSD: Samsung 850 EVO 1TB, Samsung 750 EVO 500 GB
A few things to note here. The 1080 Ti tested for comparisons is an Nvidia Founders Edition card with stock clocks on the core and memory, in addition to stock power and temperature limits. Although the RTX 2070 may seemingly appear to compete more directly with the GTX 1080 and RX Vega 64, the 1080 Ti is the card on hand at the moment. Additionally, the GTX 1080 Ti is known for pushing 4K visuals at higher framerates, thus it will be useful to understand just how close the RTX 2070 gets to achieving this on a new architecture.
When performing GPU benchmarking, it’s important to reduce any potential bottlenecks. To do so, an i7 8700k is being utilized with a 5.0 GHz overclock applied.
The games and software being tested are listed below, with considerations listed accordingly:
The selection of software represents a mix of DirectX 11 and DirectX 12 APIs, in addition to a mix of genres including first person shooter, open world RPGs, and action titles. Furthermore, these titles were chosen due to their relevance. That is, considerations were made regarding which games most players are most likely to be playing at the time of review. Finally, if in-game benchmarks were available, and if the author deemed these benchmarks representative of in-game stresses, these benchmarks were used.
In the case of Battlefield 1, the opening prologue level was played for several minutes in a repeatable run to capture data. Note that there is some variance here due to some randomness in enemy spawn.
Regarding testing, the numbers presented throughout this review represent an average of three runs at 1080p, 1440p, and 2160p (commonly referred to as 4K). Framerate is calculated utilizing frametime data as a function of time captured from FRAPS benchmarking. Using this frametime data, we are able to accurately calculate true framerates to provide an average framerate, 1% low framerate, and 0.1% low framerate. We are not capturing minimum framerate as these do not provide an accurate measurement of true moment-to-moment frame pacing.
Instead, by calculating 1% lows and 0.1% lows, we are better able to understand the overall fluidity of the game in question. This helps to account for those instances where frametimes spike, thus producing stutter that may not be reflected in a framerate number, but will be felt and adversely impact the overall fluidity of the gaming experience. For further reading, I highly recommend Scott Watson’s excellent article detailing why utilizing frametime data for performance is a better method.
When measuring temperatures, MSI Afterburner 4.6.0 Beta 9 was used to capture and recorded maximum temps over these titles in lieu of placing thermocouples on the MOSFETs and VRMs.
Regarding overclocking, MSI Afterburner 4.6.0 Beta 9 was used to modify core clocks, memory clocks, power limit, and max out temperature limits. Overclocking was tested utilizing 3D Mark Time Spy Extreme. The Overclocking section of this review will include a table documenting steps and results.
Synthetic testing consisted of utilizing 3D Mark Time Spy Extreme and Fire Strike Ultra. Curiously, the ROG Strix 2070 scored just barely higher Graphics in Time Spy Extreme than the 1080 Ti FE. In Fire Strike Ultra, we see that the 1080 Ti handily outscores the ROG Strix 2070 in both the Graphics and Combined scores. However, the ROG Strix 2070 did manage to score higher in Physics compared to the 1080 Ti FE.
The 1080 Ti FE had higher average framerates for all games tested at 1080p, except for GTA V which very well could be margin of error. However, at this resolution, the limiting factor here becomes the CPU and not necessarily the GPU. As such, we see very similar average framerates, with the largest discrepancy being Shadow of the Tomb Raider where the 1080 Ti FE commands a 17% performance lead compared to the ROG Strix 2070.
However, keep in mind that framerate does not scale linearly, with percentage differences being more pronounced at lower framerates. Meaning, a ~17% increase in framerate going from 51 fps to 60 fps is more easily felt than that same 17% increase from 100 fps to 117 fps. With this in mind, 98 fps for the ROG Strix 2070 vs 115 fps in Tomb Raider for the 1080 Ti FE should be difficult to discern for most players.
At 1440p, the biggest difference was once again in Shadow of the Tomb Raider which saw a 14% increase in the 1080 Ti FE over the ROG Strix 2070. However, in games like Assassin’s Creed Odyssey and Ghost Recon Wildlands -- two very GPU intensive titles -- we see the gap narrow between the 1080 Ti FE and the ROG Strix 2070. Overall, the ROG Strix 2070 seems to be a very capable 1440p performer.
At 4K, the gap narrows further with the 1080 Ti FE managing to hold on to a performance lead. This time, however, the largest percentage difference is in Ghost Recon Wildlands. However, looking at the true framerates further illustrates the non-linear scaling of framerates. A 15% difference between the two cards here is only 5 fps in reality. This is demonstrated further by looking at a greater 11 fps difference in GTA V, but a lower 13% difference due to the higher overall framerates.
Note, however, the 0.1% Lows for Tomb Raider at 19 fps (the frametimes clustered around 52 ms). These spikes in frametimes -- upwards of nearly 100 ms -- indicate greater frametime variance which lead to stutter. So overall, while the ROG Strix 2070 can push 4K visuals, the performance trade-off and overall compromise in fluidity may not be worth it for most, opting to stick to 1440p for most titles.
Temperature, Fan Speeds, Noise
I was pleasantly surprised by the temperature and fan performance of the ROG Strix 2070. Running with the stock fan curves, the hottest temperature was a very acceptable 74 C, with fans topping out at 2208 RPM. For reference, this is 67% of total fan speed. The card remained relatively quiet overall. Most importantly, the pitch of the noise created by the fans was unobtrusive and sounded more like a gentle whoosh than the high pitched jet engine of cards from yesteryear. Overall, I was impressed.
When overclocking, 3D Mark Time Spy Extreme was used in DirectX 12 mode with all graphics sliders set to their default state and async ON. A vanilla run was performed with no tweaks to clocks, fans, and power. From there, multiple runs were performed, slowly increasing the following in order, until a peak threshold was hit before moving on to the next item:
- Power Limit and Temperature
- Core Clock
- Memory Clock
- 100% fan speed
- Nvidia Scanner
This methodology is inspired by similar methodology leveraged by Gamers Nexus, a highly respected hardware reviewer outlet, and produces detailed results given its thorough nature. The goal by leveraging such methodology is to understand where the performance limiter is for this card. By modifying these values in turn, we are better able to understand any perceived performance gains, in addition to the impact of these modified values on temperature.
Note here, memory clock and offset are double data rate, so any values there should be doubled. Meaning, a +500 offset to the memory results in a net +1000 MHz offset, yielding memory clocks of 15000 MHz overall.
I managed to achieve a stable overclock with +130 on the core and +500 on the memory, yielding average clocks of ~1977 MHz on the core and 8000 MHz on the GDDR6 memory. Temperatures at these clocks using the stock temperature curve remained relatively cool, sitting well below 70 C. Pushing clocks further saw artefacting at +130 on core and +1000 on memory, leading to a crash in Time Spy Extreme.
Running Nvidia Scanner, Nvidia’s one-click overclocking API built into EVGA Precision X1 and MSI Afterburner 4.6.0 Beta 9, yielded +105 on the core at the top end of the voltage curve netting an average core clock of 1995 MHz -- an impressive result. It’s important to note, Scanner does not touch memory, temperature, or power limits. After running Scanner, be sure to max out the power and temp sliders. Thus, while Scanner may be useful for a simple one-click no-hassle overclock, manual overclocking allows for much finer control.
Throughout all tests, power remained the limiting factor affecting any further performance. It appears that the VRMs should be able to take more voltage, but without fully unlocking the voltage, it will be difficult to determine just how far we can push the GPU.
Should You Buy It?
No -- not yet.
The xx70 tier of cards historically were always meant to be that “affordable performance” bracket for gamers who wanted a card with most of the performance of a flagship for considerably less money. While it no doubt appears to be a capable 1440p card, the Asus ROG Strix GeForce RTX 2070 OC Edition flies in the face of that ethos for the asking price of $630. Keep in mind also, of the trio of cards with ray tracing and Tensor functionality, the RTX 2070 is the least capable based purely on specs alone.
It is increasingly difficult to justify this performance for $630 -- $130 over MSRP -- when a more powerful card in the 1080 Ti with 3 GB more memory is only $70 more.
Perhaps most egregious, the namesake of this card (ray tracing) simply doesn’t exist yet for games, and neither does DLSS. Nvidia are essentially asking you to buy this product based on a promise, and an unfulfilled one at that. As a critic, I cannot in good conscience recommend that you go out there and spend your hard-earned $630 on this graphics card. Either wait until games actually get ray tracing and DLSS, or buy a (still very good) Pascal card.
- Striking design
- Good thermals
- Great acoustics
- Solid 1440p performance
- Decent overclocking
- Costs far too much for a xx70 “affordable performance” tier
- Last gen Pascal cards are not much more, offering better performance
- Zero DLSS supported titles
- Zero ray tracing supported titles
The product discussed in this article was provided by the manufacturer for the purposes of review.