Today we have another big current and previous generation GPU benchmark, this time focused on a little game you might have heard of called Battlefield 1… yeah apparently it’s pretty gooood.
Back when the game was in the open Beta stages Steve and myself put quite a few hours into the game, more than we should probably admit. Thankfully we did capture some of the action which was shown in my triple 4K monitor Titan X teaser video.
Walking away from the beta we were shocked by not just how amazing the game ran, but also how incredibly well optimised it seemed to be. Now that the retail version is available we are keen to see if any of the magic has been lost.
For testing, well over a dozen GPUs have been used in our Core i7 test and of course the latest AMD and Nvidia drivers were installed. Testing the multiplayer aspect of the game is near on impossible with this many graphics cards – you simply can’t gather accurate results… So like all the big name tech sites I will be forced to measure the game’s performance using the single player campaign.
Now this title does support both the DX11 and DX12 APIs and you might think I would naturally focus my efforts on testing with DX12. This was the plan but it soon became clear that for this title the low-level API has nothing to offer at this point. The game ran just as well using DX11; in fact for the most part it ran better. Typically the DX11 mode allowed both AMD and Nvidia GPUs to render 3 to 5 fps more and at times the margins were as high as 10 fps in favor of the older API. So I’ve included one graph comparing the DX11 and DX12 performance of half a dozen GPUs but believe me, it is the same situation for all GPUs, Maxwell, Pascal and GCN.
The benchmark pass lasts 60 seconds, we start at the beginning of the “Through Mud & Blood” story where you take command of a British Mark 5 Tank. This test features plenty of AI controlled characters and more importantly an easy to follow path that allows us to reproduce the test over and over again with a high degree of accuracy.
Previous Gen 1080p
As we often do let’s kick things off with the previous generation GPUs. Battlefield 1 is a fast paced first-person shooter so ideally gamers will be looking for an average of at least 60fps for optimal gameplay. With the image quality settings maxed out the R9 380 came close to a 60fps average which is very impressive. The GTX 960 wasn’t bad either though owners will want to employ some overclocking here.
The GTX 970 pushed well above 60fps at all times and the R9 390 was even more impressive. Beyond that last season’s high-end pixel pushers had no trouble rendering well over 60 fps.
Previous Gen 1440p
Jumping up to 1440p quickly gives the mid-range GPUs a quick reality check. The R9 380 and GTX 960 drop down into the 30s at times. So the game was still playable but as a member of the PCMasterRace it was far from ideal. The GTX 970 is really required here while the R9 390 would be prefered, assuming both are compared at the stock reference clock speeds which they are in this case.
Previous Gen 4K
Now at 4K there are no surprises here. The 980 Ti and Fury X were never really able to handle the latest triple A titles without multi-GPU support. The 980 Ti puts forward a gallant effort rendering 50fps and again while playable dipping into the 40’s kind of spoils the 4K experience.
Current Gen 1080p
Fast forward to the present GPU battle and we find the 4th Gen GCN cards taking on Nvidia’s mighty Pascal lineup. The only overlapping product here are the GTX 1060 and RX 480, both of which deliverd similar performance. Not just that but the performance that was delivered allowed for extremely smooth gameplay at 1080p.
Something you might have noticed is the performance discrepancy between the 2GB and 4GB versions of the RX 460. Having spent quite a lot of time testing frame rate and frame time performance I found that for over 90% of the games out there the difference was negligible. The only anomaly at the time was Doom. Recently we discovered when testing Gears of War 4 for example, there was also very little difference between the two.
However here we see that the 2GB model is over 20% slower than the 4GB model and this resulted in a massive drop from 45 fps to just 35 fps. This makes the 4GB version the must have option here and a complete contradiction to my RX460 2GB vs. 4GB conclusion, or is it?
At the time I was chasing 60 fps and here with the 4GB card we are only averaging 45 fps. This means we would lower the quality settings at 1080p. Without spending much time tweaking the quality settings we simply downgraded to the high quality preset. This bootest the 4GB card by 16%, a much nicer 52 fps average. However, we saw a significantly larger 40% performance boost for the 2GB model and it’s now averaging 49 fps. This means the 2GB version is just 6% slower and more importantly capable of delivering smooth playable performance.
Current Gen 1440p
Even at 1440p the RX 480 and GTX 1060 still provided a nice gaming experience. The RX 470 wasn’t bad either with an average of 56 fps. As you might expect, 1440p is just a bit too much for the RX 460, even if it is armed with a 4GB framebuffer.
Current Gen 4K
The GTX 1070 is really the slowest GPU you want to be gaming with using these quality settings at the 4K resolution. Obviously if you can afford it the 1080 is a much better option as it never dropped below 55 fps and managed to average an impressive 65 fps. It has to be said, while stupidly expensive, the Titan X does look quite amazing here.
All cards – 1080p
Here is a busy looking graph that includes the current and previous generation GPUs together in a symphony of blue bars. As you can see we have a number of varied options producing between 80 and 100 fps, so GTX 1060 and RX 480 territory then.
All cards – 1440p
Here is the story at 1440p. The GTX 1060 and 980 delivered similar performance, as did the GTX 1070 and 980 Ti. The RX 470 looks pretty good sitting next to the GTX 970 while the RX 460 was just rubbed out by the GTX 950.
All cards – 4K
Not much more needs to be said for the 4K results. Gamers will want a GTX 1080 here or if they can survive without food for a few months the Titan X is a nice option as well.
DX11 vs. DX12
So at this point you might be thinking, what the hell Matt “where the DX12 tests at, I don’t wanna see this DX11 rubbish”. Well okay you might not be thinking that but I can assure you someone has already made the comment! As I explained at the beginning of the video the Battlefield 1 DX12 performance is very disappointing, let’s take a look shall we.
Here you can see that for the most part DX12 is a step in the wrong direction, as far as frame rate performance is concerned. It isn’t a huge step but here we see most AMD and Nvidia graphics cards drop a few frames when using DX12 over DX11. This is pretty much the same issue we have seen in other titles that have been designed initially for DX11 and then basically ported to DX12.
CPU utilization sat at around 70 to 80% in both the singleplayer and multiplayer portions of the game. Keep in mind those utilization results are based on a Core i7-6700K clocked at 4.5GHz. So needless to say the game requires some serious processing power.
That said the game only ran a few frames slower on the 6600K at the same clock speed. My Core i3-6100 system also played very well and pushed well over 60fps at all times when gaming at 1080p with a GTX 1080.
Well there you have it, Battlefield 1 is an awesome looking game that is very well optimised. The only headache we encountered was the account limitation which locks you out for 24 hours after five hardware changes, that made testing a pain and rather expensive as three copies of the game were required.
With that mostly behind us for now we can look to the positives. Gamers running modest rigs can really enjoy this one. Those rocking a previous generation Radeon R9 380, GeForce GTX 960 or something of equivalent power can enjoy Battlefield 4 in all of its glory.
Meanwhile 1440p gamers won’t need to spend their Friday nights looking for work in St.Kilda, affordable mid-range options such as the RX 480 and GTX 1060 will more than suffice.
It was disappointing to find that DX12 offered nothing over the ageing DX11 API. If anything DX12 was a step backwards. Of course the image quality remains the same so there is no advantage to using one over the other in that respect. The issue being that for the most part DX12 is slower or at best able to match the performance of DX11. I did notice on my lower end Core i3 test rig that the minimum frame rates were slightly higher with the low-level API. So for those running weaker CPUs DX12 might provide a small performance bump.
For everyone else at this point it doesn’t matter too much with API you run, though I personally will be sticking with DX11 for those few extra frames. What do you guys think about Battlefield 1?