Overclocked Performance and Conclusion
I was expecting the RTX 2060 to hit 2 GHz based on what I was hearing leading up to the launch, and was eager to push it as far as I could when I returned from CES. The result? I was able to achieve a stable overclock of +170 MHz on the core, and +500 MHz (effective) memory, and while I could have tried to push memory further I was unable to get anywhere with core speeds above +170. Power and temp limits were pushed all the way up, with overclocking numbers reached using the latest version of MSI's AfterBurner utility and the latest graphics driver from NVIDIA.
What these numbers translated to was a sustained boost of between 2055 and 2070 MHz in the benchmarks to follow, and the 500 MHz increase in memory clocks (250 MHz actual) provided a lift in memory bandwidth from 336 GB/s to 360 GB/s.
So how close can we get to a stock RTX 2070? Let's start with the 3DMark Time Spy benchmark.
This is not the most realistic benchmark with regard to game performance as we will see, but the added performance from this OC is significant, with the overclocked RTX 2060 actually coming out slightly ahead of the RTX 2070 – in this test, anyhow. (Spoiler: this is the only test with this outcome.)
Unigine Superposition is next.
Here the RTX 2070 regains its lead, but the performance gain is sizable for the overclocked RTX 2060.
On to the game benchmarks, and while I won't cover every game I did test a few from the previous results to see if the overclock held up beyond the synthetic benchmarks. We'll start with Far Cry 5.
The RTX 2060 with this overclock comes pretty close to the RTX 2070, and is providing a nearly 10 FPS increase over the stock clocks. Frame time variance is nearly identical, with about a 2.6 ms gap between the average and 99th percentile frame times, and the performance was smooth to the naked eye once again.
Up next is Middle Earth: Shadow of War:
Once again the RTX 2060 leapfrogs the GTX 1080, but comes up short of the RTX 2070 in FPS numbers. The frame time variance actually improves slightly with the overclocked card, dropping about 0.4 ms from the result with the stock RTX 2060.
Moving on to the DX12 benchmarks, we start with Ashes of the Singularity: Escalation.
This was not the most impressive benchmark for the overclocked RTX 2060, with performance below the GTX 1080 in this game when looking at FPS numbers, but there was a big improvement to frame time variance compared to the stock results, with the gap between average and 99th percentile closing to 12.9 ms from 19.3 ms – not good, but much better than it was.
Finally we have Shadow of the Tomb Raider:
Once again the overclocked RTX 2060 surpasses the GTX 1080 in performance, sitting between the 1080 and the RTX 2070 on the chart. Frame time variance was about the same as before, and results were still a little uneven with some additional spikes compared to the earlier GTX 1080 and 1070 results.
As to power draw, replicating the original test I saw power go up from 243 to 270 W total system power (measured using a watts up? Pro meter from the wall). Temperature results were not really useful in comparison to earlier results as I was using a custom fan profile during overclocking, but I topped out at 59 C with an ambient of 19.5 C. I don't set cards to 100% fans when overclocking, relying instead on a linear rise beginning at 30 C, and with my profile the fans on this Founders Edition card maxed out at ~70% during the tests.
Conclusion
It has been a couple of weeks since part one of this review was published, and in that time I've had numerous conversations about this card's performance and where it sits relative to the market. I get the sense that NVIDIA has created a bit of confusion based on the nomenclature, with the RTX 2060 apparently assumed to be a direct successor to the GTX 1060, and with that assumption the $349 price tag seems high. But this is not a midrange card as we have seen, but a product that sits between what we might have expected from a card called "2060" and the existing RTX 2070. The closest parallel in performance is the GTX 1070 Ti, which launched at $449, and depending on the game you are getting performance at or even above that level for $100 less a year later.
So are we actually getting a 2060 Ti without the "Ti" branding here? Not really – though it could be argued we are getting the "RTX" version of a card before the "GTX" version. The comparison I keep making in my head is to the GTX 560 Ti 448 (if anyone remembers that card), which was based on the same core as the GTX 570 from back at the end of the Fermi era. Why this comparison? Besides my obsession with obsolete hardware it actually makes some sense to bring up this odd limited-edition card as this RTX 2060 is actually based on the same TU106 core as the RTX 2070, but with 30 SMs enabled vs. 36 with the 2070. While the reduction in memory bandwidth is significant (the RTX 2060 moves from 256-bit to 192-bit GDDR6, and from 448 to 336 GB/s), as we saw with our overclocked results it is possible to get close to the RTX 2070 – and with a higher memory overclock the gap could get smaller.
Ultimately this card is a 1440p gaming card with a name that might confuse those expecting the successor to the popular GTX 1060 (easily the #1 card in Steam hardware surveys), but the RTX 2060 blows away the GTX 1060 6GB card by nearly 2x across the board, besting the GTX 1070 and even with (and occasionally above) the GTX 1070 Ti in the benchmarks we just saw. In fact, when you look at this as a direct successor to the GTX 1070 Ti, but with ray tracing capabilities, suddenly the $349 price looks much better. I contend that it's priced quite fairly based on the performance we have seen. But that's not the end of the story, as ray tracing has not been explored here. I will be covering ray tracing (and hopefully DLSS) performance between RTX cards in a followup article, but from results available so far the RTX 2060 is clearly intended as not so much a ray tracing option as a DLSS (deep learning super sampling) option. We also need more games to test before we can make a determination about ray tracing performance in general.
As things stand the GeForce RTX 2060 is a mighty little card with performance that challenges the "midrange" label, and carves out a position as a formidable 1440p gaming option with the overclocking potential to flirt with RTX 2070-level performance. Not bad for $349, after all.
You might want to re-check
You might want to re-check that World of Tanks frame time/frame rate math.
9.8 ms average frame time does NOT work out to 50.7 fps average frame rate.The rest of the graphs seem fine.
You’re right – those are the
You're right – those are the 2160 numbers. Wrong chart, will be fixed shortly. Thanks!
What monitors did you test
What monitors did you test with? If they support adaptive sync/freesync, did you notice any issues?
Standard 60Hz UHD monitor
Standard 60Hz UHD monitor from Samsung (this one I believe) – nothing special, no VRR. AS/FS testing will have to be a separate project.
no matter how you try to
no matter how you try to justify it, prices at all tiers of RTX have been raised SIGNIFICANTLY over previous generations.
Also, have you guys given up on the in depth graphs that Ryan and Ken used to do?
also, f@#k this @#$##ing catcha and all of those f#^%ing ridiculous “click the pictures with ‘x’ in them”
Obviously prices for the RTX
Obviously prices for the RTX cards are higher compared directly to the GTX 10 series. I made observations about relative performance and price for the RTX 2060 alone, which in my opinion is priced fairly based on its performance. I don't expect everyone to agree with that, and I welcome your opinions.
I have not given up on the graphs. For this review I have started from scratch with all new benchmarks, all of which I have personally conducted, and I have amassed a lot of OCAT data which can be graphed in a similar way to the previous results. The main difference is that FCAT is no longer used, and without it the path to those graphs involves creating a different workflow. OCAT does have a built-in visualizer with the same type of frame time graph we were used to, but only for one result at a time. I was going to make charts from frame time data in Excel but with the number of cards I used in this review it would be a visual mess. Once I have a viable solution I'll work it in to future reviews. A smaller test group is going to be important (at least on a per-chart basis) for frame time analysis.
Good review and summary. The
Good review and summary. The RTX cards are all great performers, no question. It’s the VALUE associated with that performance that has people swallowing hard. $1200 and $800 US for the 2080Ti and 2080 made them extremely unaffordable for most people. The RTX 2060 is the first card in the RTX series that actually has an affordable price tag and great value for it. This easily going to be their best selling card.
The 2060 lacks the
The 2060 lacks the performance to back up the value RTX value option. 2080 and 2080ti struggle to run ray tracing even at lower resolutions, why would you want to get a lower tier card to try and do the same thing? It’s pretty much a useless feature on a lower tier card. that is until more content comes out and bugs get worked out and optimized, but by then next gen *60 cards will come out that do rtx much better, so rtx 2060 is not a great option. It might be the best selling RTX card, but not nvidia’s best selling turing card.
What will be their best selling card(s) are the gtx 1660ti or the gtx1660(and lower tier cards like 1650 or 1650ti if naming stays intact). Just like AMD and previous nvidia generations, the best selling cards are almost always $250 or under.
Hey Sebastian,
what do you
Hey Sebastian,
what do you think about graphs like this:
https://www.gamestar.de/artikel/msi-rtx-2070-armor,3335875,seite2.html
I really like having a hover over to see the power of different cards in relation to the one I am interested in. Would be great to get a quick idea about the price performance.
Do you think those graphs would make sense? Would your workflow be compatible with doing such graphs?
I think interactive charts
I think interactive charts are great, and when we are running a new site design we will be able to do much better things with charts than our current static images.
A very thorough review. Good
A very thorough review. Good work, Sebastian!
Thanks for the review very
Thanks for the review very well done thank you.
If you look at this card as just a Graphics card and do not intend run games that support ray tracing turned on this card is a mighty little performer for sure. I really think Nvidia should have just removed RT cores left the tenser cores for DLSS only and priced it $25-$50 lower and released it that way.
Also this DLSS option is it true that it only works while in 4K res? If so can this card even handle 4K very well. I guess I should go and check out the other review you did that has the 4K numbers for this card. My own opinion is that the 6GB frame buffer would limit the cards ability to run the most current games at high settings and get a lot worse when the new games roll out through out 2019.
I guess the question is still this does or will DLSS support 1440p res which would really enhance this cards performance even more at that res if DLSS is only supported in 4K then I think DLSS should be unchecked form the list of features since the cards limited 6GB memory would become the limiting factor.
DLSS works at 1440p too; you
DLSS works at 1440p too; you just wouldn’t want to use it because for that output resolution it renders the game at 1708×960.
It should work decently well to enable “4K” on the 2060, though, because at 4K DLSS renders at 1440p. In games that support DLSS you’ll get 1440p performance and visual quality that’s… a bit better than up-scaling 1440p to 4K. It’s not a miracle by any means, but this card is actually likely to need it more than something like a 2080Ti.
Appreciate the benchmarks but
Appreciate the benchmarks but for the millions of Sim Racers such as myself could you toss in another SIM racer? Assetto Corsa has it’s own benchmark and without checking, I believe Project Cars 2 may as well.
Anyway, it would be great and thanks.
I thought about adding more
I thought about adding more racing titles, but with F1 already there I held off. Going forward we can look at it, and if I expand the list of games in later reviews I'll make sure racing games are represented.
Since this card was my first attempt at a GPU review since taking over as EiC I was going with games that are kind of consensus picks in wide use on review sites, not trying to reinvent the wheel (pun intended).
I might have missed why but I
I might have missed why but I would have want to see how the rtx2060 goes against the Vega 64 vs the RX590.
If you put just those three
If you put just those three cards together it makes no sense. But I'm pretty sure I benchmarked more than just those cards.
I am an idiot and did not see
I am an idiot and did not see VEGA 64 on the charts somehow. Sorry.
Very nice follow-up review
Very nice follow-up review here. Having both the average FPS as well as the frame time graphs is great for comparison. Thanks.