Hey guys it’s time once again for another huge benchmark session; this time the focus is on Nvidia’s new GeForce GTX 1060. Two weeks ago I was able to reveal the 1060 and most of its specs in the video here. Rather than go over everything again, here’s a table I put together comparing the 1060’s specs to the 1070 and some of last generation’s Maxwell GPUs.
As you can see the 1060 is a tiny 200mm2 die and yet, thanks to the 16nm FinFET process Nvidia was able to cram inside an incredible 4.4 billion transistors. This means despite being 12% smaller than the GTX 960, the 1060 boasts 50% more transistors which has allowed Nvidia to increase the core count by 25%. The 1060 also features 80 Texture units and 48 ROPs.
In total the 1060 ships with 1280 CUDA Cores and 10 SM units. The memory subsystem consists of six 32-bit memory controllers for a total bus width of 192-bit, using 6GB of GDDR5 memory operating at an 8Gbps data rate.
The base clock speed of the 1060 is 1506MHz and the typical Boost Clock speed is 1708MHz. Nvidia says that the Boost Clock speed is based on the average 1060 card running a wide variety of games and applications. Of course as we’ve seen with the 1070 and 1080 the actual Boost Clock will vary depending on conditions, namely thermals and power target.
The GTX 1060 reference board, or Founders Edition if you will, measures 249mm in length. Display outputs include one dual-link DVI, one HDMI and three DisplayPort connectors. Like the RX 480 we find just a single 6-pin PCIe power connector, but with the TDP set at just 120 watts we don’t expect Nvidia will run into the same issues AMD faced.
So with the basic information now out of the way let’s get to the stuff you guys came for – the benchmarks. As always you guys can find all the graphs at the Hardware Unboxed website, which will be linked in the description. After looking at the results from a select few games I’m going to heavily analyse the data from all 25 games across the multiple resolutions to get a clear idea of how the 1060 stacks up.
Before I jump into the benchmarks here are a few quick notes. All tests were conducted using my GPU test rig which is built inside the Corsair Carbide 600C with an Intel Core i7-6700K clocked and locked at 4.5GHz. For a full list of the system specs please check the video description, you’ll also find a detailed video index there as well. Finally, all benchmarks were conducted after a 10 minute warm up period as this is sufficient time to get these cards up to their maximum working temperature. As always I’m using reference AMD and Nvidia graphics cards unless specified. I’ll be focusing on the 1440p resolution for the discussion but I will quickly take a look at the 1080p performance as well…
Let’s start with Doom shall we since it is a game I recently re-benchmarked due to its Vulkan implementation. For you guys we have tested all 12 graphics cards using both the OpenGL and Vulkan APIs so you can see how the AMD and Nvidia cards stack up. I’ll just skim over the 1080p results as I’ve decided to focus on 1440p with the GTX 1060 given its performance. Here we see the 1060 is 13% faster than the RX 480 using OpenGL but shockingly 14% slower with Vulkan. Let’s see if this changes at 1440p.
This time the 1060 was 15% faster than the RX 480 using OpenGL but it was also 14% faster in Vulkan. In fact the 1060 performance in Doom was very sub par as it lost out to not just the GTX 980 but also the 970… hopefully this isn’t a sign of what is to come. I suspect what we’re looking at here is a driver issue, let’s move on to find out if the 1060 does any better in Overwatch.
Okay so things couldn’t look more different in Overwatch. Here the GTX 1060 was 21% faster than the RX 480, 20% faster than the 970 and impressively 8% faster than the 980. The 1060 was 24% slower than the 1070 though that was to be expected, after all it was still 27% slower when testing with Doom.
Increasing the resolution to 2560×1440 helps the 1060 extend its lead over the RX 480 as it’s now 27% faster. The 1060 also increased the margin from the 980 to 11%. That said it was still 23% slower than the 1070.
Here we have The Division and at 1080p we find a very close battle between the GTX 1060 and RX 480. Nvidia does win out here, but only by a very slim 3% margin.
Now at 1440p the 1060 is able to pull further ahead of the 480, extending the margin to 12%. This saw the 1060 beat the 980 by a 4% margin while it matched the 390X.
Hitman was tested using both DirectX 11 and DX12. Even under DX11, Nvidia’s GPUs perform pretty poorly in Hitman. Here we see that the 1060 is 4% slower than the 480 and moving to DX12 only amplifies Nvidia’s issues in this game was the 1060 is now 14% slower.
Even at 1440p the RX 480 along with the R9 390 have no issue beating the new 1060 in Hitman. Although the 1060 was faster than the 980 it was still 4% slower than the 480 when using DX11.
Rise of the Tomb Raider (DX11)
I’m yet to update my results for Tomb Raider using DirectX 12 so for now we’ll have to settle for the DX11 numbers. From what I’ve seen the GTX 1060 does stay well ahead of the RX 480 even when using DX12. Here we see the 1060 beating the 480 by a 16% margin with an average frame rate of 96fps.
The 1060 is able to crack the 60fps barrier at 1440p and is again 16% faster than the RX 480. It was however 3% slower than the GTX 980 which was surprising.
Total War: Warhammer
Total War Warhammer is another game that I’m yet to start testing with DX12. DX12 is marked as Beta at the moment, which is why I’ve been reluctant to make the move. Using DX11 we see that Nvidia does have an advantage, here the 1060 is 12% faster than the RX 480. Surprisingly the 1060 was 10% slower than the GTX 980 despite matching the R9 Nano.
Increasing the resolution to 1440p we see that the 1060 is still faster than the RX 480 – this time 10% faster. Still the 1060 was 15% slower than the GTX 980 which is interesting and makes me wonder if we’ll see better performance out of the 1060 in this title through a driver update.
Mirror’s Edge Catalyst
Mirror’s Edge Catalyst is another new game we’ve started testing with, though this title only supports DirectX 11. As you can see the 1060 is faster than the RX 480, 12% faster in fact at 1080p.
Now at 1440p we see a few interesting things. Most interesting of which is the fact that the 1060 is now quite a bit faster than the GTX 980 whereas they delivered the same performance at 1080p. The 1060 is now 20% faster and this makes me think the 4GB memory buffer of the 980 is to blame here. Mirror’s Edge Catalyst was one of the few games that showed the 8GB RX 480 outperforming the cheaper 4GB model. Speaking of the RX 480 here we see that the 1060 is just 7% faster with an average of 49fps opposed to 46fps.
ARMA 3, ohh ARMA 3 why do I test with you? Ahh that’s right because you guys beg for it every time so here we are. Unfortunately, ARMA 3 doesn’t play nice with the RX 480 and as a result the GTX 1060 is 31% faster in this title. Here the 480 would dip as low as 48fps, while the 1060 never dropped below 68fps.
Again at 1440p the 1060 was 31% faster than the RX 480, though this time it was 4% slower than the R9 390X and GTX 980. It seems strange that the RX 480 is so much slower than the R9 390 in ARMA 3, given it typically matches or beats the R9 390.
Witcher 3: Wild Hunt
I know someone is going to say it in the comments so let me address this real quick. Testing with HairWorks enabled doesn’t put the AMD cards at a disadvantage in The Witcher 3 anymore – we’ve moved on from 2015. The 1060 is just 7% faster than the RX 480 with HairWorks enabled and you can expect to see a similar margin within 1-2% with HairWorks disabled. Okay? So let’s move on to 1440p.
Here the 1060 is just 5% faster than the RX 480 as it isn’t able to beat the GTX 980 this time, trailing by a 6% margin. Still The Witcher 3: Wild Hunt is a demanding game with all the eye candy turned on so an average of 44fps at 1440p is a commendable result.
This is where I expected the 1060 to really impress and well… it has. Total system power draw with the 1060 was 17% lower than with the RX 480 at just 195 watts. Moreover using the 1060 our test system consumed just 15% more power than the GTX 960 which is remarkable given we saw GTX 980-like performance for the most part. Still the 1060 doesn’t exactly destroy the RX 480 here though it has to be said saving 30 – 40 watts is going to make Nvidia’s mid-range offering much easier to cool. Speaking of cool things let’s check out the Founders Edition operating temperature…
As we expected the GTX 1060 does run very cool, the Founders Edition never saw the GPU temp exceed 64 degrees which is a similar result to the GTX 960. This is a huge improvement over both the GTX 980 and Radeon RX 480. For those wondering the 1060 Founders Edition runs really quiet, it definitely couldn’t be heard over my Corsair Carbide 600C case fans which are very quiet to begin with.
Okay so in this video I took an individual look at just 9 of the 25 games tested. For the other 16 games be sure to check out the written version of this review at the Hardware Unboxed website. For this video I didn’t want to take up 30 minutes or so of your time walking you through all 25 games… just the thought is exhausting. Instead I’m going to make a few quick GPU comparisons based on the data from all 25 games tested.
Price vs Power Consumption
But before we jump to that, here’s a look at the performance vs. power consumption data, using a scatter plot graph. For those unaware the further a data plot is to the right on the x axis the faster the frame rate, while the higher the figure on the y axis the greater the power consumption. So in short if you’re a GPU you want ideally want to be as low and as far right on this graph as possible. Doing just that is the GTX 1070 which consumed just 2.91 watts per frame.
The GTX 1060 came in second consuming slightly more power per frame thanks to a much lower average frame rate. Let’s declutter things a little by removing all GPUs except for the 1070, 1060 and RX 480. Here we see when compared to the RX 480 the 1060 did very well and from our 25 game sample, proved it was 25% more efficient.
However let’s remove all the DirectX 11 titles and focus on the games that feature low-level API support. These games include Hitman, Ashes of the Singularity and Doom. As you can see the picture changes quite dramatically, as Nvidia’s GPUs become slightly less efficient and AMD’s RX 480 becomes considerably more efficient, thanks to a bump in performance. Here we see when using DX12 and Vulkan the GTX 1060 and RX 480 are pretty much on par with one another, so that’s interesting to see.
GTX 1060 vs RX 480
Now how do the GTX 1060 and RX 480 stack up at 1080p? On average the 1060 was 12% faster across the 25 games tested. The only real outliers here include Hitman, Anno 2205, ARMA 3 and surprisingly Battlefield 4. Games where the margins were very close include Black Ops III, Batman, Ashes of the Singularity and The Division.
Moving to 1440p the big picture doesn’t really change much, the 1060 is still 12% faster than the 480 and we see the same outliers. Compared to the R9 390 the GTX 1060 was on average 10% faster at 1440p and we see the same outliers from the RX 480 comparison.
GTX 1060 vs GTX 980
So how does the 1060 compare to the 980? After all Nvidia promised the power of the GTX 980 for every gamer. Well they look to have delivered on that promise as overall the 1060 was 1% faster. That said for more than half the games tested the 1060 was slower, though in most of those games the margin was 5% or less.
GTX 1060 vs GTX 1070
What about the next step in the Pascal food chain, the GTX 1070? Well the 1060 was on average 26% slower which isn’t bad given the Founders Edition is at least 26% cheaper while partner cards should be around 40% cheaper.
GTX 1060 vs GTX 960
Those rocking a GTX 960 will certainly notice the upgrade to the 1060 as we found it was on average 81% faster. In fact worse case scenario still saw the 1060 deliver over 50% more performance.
Cost Per Frame
The final and probably the most important data comparison I’m going to make is the cost per frame. The figure next to the GPU model is based on the current “typical” asking price in US dollars from popular retailers such as newgeg.com and Amazon. Most of these GPUs are a bit redundant now such as the Fury X, GTX 980 and 970 for example, but I have included them anyway out of interest.
The 1060 Founders Edition comes in at a cost of $5 per frame based on our average performance from the 25 games tested. This is kind of poor value frankly; the GTX 1070 for example costs only slightly more at $5.30 while the RX 480 is considerably cheaper at $4.44, and that’s for the 8GB model.
For me though the RX 480 4GB model is where the value is at, even if it really is a 4GB card and not a model that can be unlocked. At $200 the 4GB model costs just $3.70 per frame, as it delivers the same performance as the 8GB model in our tests. This means Nvidia really needs those board partner cards to come in at the $250 MSRP if they want to be competitive in the sub $300 price range. At that price the 1060 came at a cost of $4.16 per frame making it 6% cheaper than the 8GB RX 480, but 12% more than the 4GB model.
Quite a few of you didn’t really like the scatter plot graph in my RX 480 review. When it came to showing the cost per frame many of you asked for bar graphs instead. Here we have the GPUs sorted by their average frame rate performance and I’ve highlighted the $250 board partner 1060 as well as the 4GB RX 480.
Here you can see the same graph resorted according to the cost per frame, starting with the cheapest card, ranging down to the most expensive. As you can see the RX 480 4GB does just edge out the board partner 1060’s, assuming they do actually come in at $250 US.
Well there you have it… all of it 🙂 The GTX 1060 is as expected faster than the RX 480 while also going a little easier on your power supply. Our 25 game sample indicates you will see a little over 10% more performance on average but you will also pay a little over 10% more per frame when compared to the 4GB RX 480.
If we do indeed see Nvidia’s partners delivering custom cards to market for $250, then I feel AMD’s RX 480 8GB card could be in trouble. This is down to the fact that it costs more per frame as a result of it being slower. Of course it consumes more power as well and probably won’t offer the same degree of overclocking headroom. Speaking of which you can check out my GTX 1060 Founders Edition overclocking video here, while my first GTX 1060 board parter card review, which also includes overclocking, can be found here.
Overall, I feel it’s too soon to call a real winner here. That said it is fair to say we have a real fight on our hands once again in the sub $300 GPU market and that will likely end up being good news for you guys. A lot of things are going to shape this battle over the next 6 months, from board partner cards to upcoming DX12 titles, so I can’t wait to see how it plays out.
BONUS CONTENT: Additional Benchmarks