Ashes of the Singularity is a game AMD had a big hand in making. There are driver level optimizations the studio paid for. Furthermore the game leverages D3D12's Async Compute which Pascal Architecture can't handle as well as GCN.
That being said in non-AMD sponsored games they two set ups in question (RX480 CFire vs GTX1080) do manage to stay within step of one another [1].
The real issue becomes overclocking and crossfire.
The GTX1080 has a much higher overclock threshold then RX480 (currently this may change as board partners make custome PCB's). So if you start fiddling with power levels this comparison breaks down.
Also crossfire/SLI is to a degree Russian Roulette. Studio's have to opt-in to supporting it, and sometimes don't. Leaving you only with a single card. Also very weird bugs can sometimes crop up. Some people have no issue with this, some have huge issues. YMMV.
The main thing I'm trying to express is when looking into hardware reviews CPU/GPU's more information from more sources is always better. And all sources/chips aren't equal.
Async Compute is a big mess for nVidia right now. Latest trouble is Total War: Warhammer game. They released dx12 support for it and nVidia cards are losing 10fps+ while AMD cards are gaining 30%+ in performance. Also reason why Ark Survival Evolved is holding off dx12 release because of nVidia ...
It's really difficult to draw conclusions from DX12 benchmarks at the moment.
The problem is that DX12 shifts most of the responsibility for architecture-specific optimizations (and avoiding pathological cases) from the driver to the application. So although you could draw the conclusion that Nvidia are struggling with DX12 due to their Total Warhammer performance, it's just as easily explained by Nvidias veteran DX11 driver team being better at tuning for Nvidia hardware than a game studio who partnered with AMD during development.
You mean the game that AMD is giving away for free with it's cards right now, to showcase it's cards, which has been optimized completely for AMD cards is outperforming nVidia cards? I'm shocked.
So the 1070 is the no-brainer choice in this price range. A single card is always better than XFire/SLI, and you have the option to move to 2x1070s if you need the boost. SLI scaling on the 10xx series has been really impressive.
Is that a good card? I saw an article linked somewhere from this post that I can't find now, but it said that things are about to get very good for graphics card purchases. I'm not sure if the article was old and things have gotten better.
I'm basically wondering what I should replace my current card with (I run Linux and play a few games like rocket league).
The GTX 1070 is good value for the performance you get but it's overkill for games like Rocket League - apparently it pushes over 220fps at 1080p and max settings. You may want to wait a few weeks for the GTX 1060.
You can get new 970 for slightly more than that... I regret buying a 380x instead.
AMD drivers have serious bugs (that also affect the 480!) that they promised to fix a year ago, but not only they didn't fixed it, they also removed it from the "known issues" list on the changelogs. This was my first AMD purchase and will be my last, their hardware is mediocre (super power hungry and throttles because of that), software is outright terrible, and support from both them and Sapphire is nonexistent.
And when I asked for help on forums and chats, people accused me of being nvidia paid troll and even banned me. (And someone later pointed out to me that the places that most aggressively attacked me are the ones that have AMD employees as moderators)
The most infamous issue is nicknamed "black screen" google it and you will find tons of people complaining.
And I had problems with every single software, thus why I don't considered mentioning specifics.
Still, some examples:
CAD software stutters heavily because the card refuset to get out of idle state unless I am gaming, regardlesst of all power saving related settings.
Gtav was unplayable on most Crimson drivers, it works in the 16.6.1 but still has the occasional stutter due to power throttling.
Default fan profile is terrible, launching any AAA game without third party fan software makes the card overheat, release pcb varnish smell and make coil noise.
On default power limit the card throttles a lot and cause game stutter (all of them)
The CCCSlim panel on Crimson didn't let me use it because according to it I had multiple monitors with genlock enabled. I was using a single monitor, and genlock is a FirePro feature.
Driver crash a lot. A easy one to reproduce is entering the ingame herb encyclopedia on Shadow of Mordor.
Lastest drivers Mantle is buggy (known issues list), vulkan is also crappily implemented despite being AMD invention.
Browsers flicker on some driver versions for no obvious reason.
It also seems like RX 480s also overdraw the PCIe bus (drawing more than 75W from the slot--cards are supposed to be able to pull 75W from the slot, 75W from the six-pin connector, and 150W from the eight-pin connector), which is potentially dangerous for the rest of your hardware, too.
I went from being a steady AMD/ATI customer (last card was an HD4870) to a 980Ti and don't see myself going back anytime soon.
It's concerning at stock settings but gets far worse when the card is overclocked, with the card overloading the 12V pins by about 50%. AMD say they are working on a driver fix at least.
For that sort of thing you'd be fine with the previous generation of graphics cards, I've got a Radeon R7 series, it's quite happy running most things with the settings turned up and costs a fraction of a high end card.
But Nvidia's crippled SLI with the 10xx launch and declared many scenarios unsupported or not working. It's a little unclear right now because Nvidia isn't communicating it properly, but what used to work and was supported is not as it stands right now.
They've clarified the situation now. 2-way SLI continues to work as normal, but 3-way and 4-way SLI are no longer supported with the exception of certain whitelisted benchmarking apps (for the sake of people trying to set world records).
3/4-way SLI never worked particularly well in practice so it's not a great loss.
Non-SLI setups are unaffected, you can still use as many cards as will fit in your motherboard.
They deprecated 3/4-way SLI due to the difficulty of implicitly splitting the workload in the driver, but with non-SLI setups for CUDA/OpenCL the load-splitting is done explicitly by the application so it's not their problem.
Does that mean SLI/Crossfire is not required when using Vulkan or DX12 (or Metal) and applications can saturate all cards if they want to? Isn't there a delay if you don't use Nvidia's bridge connector?
Considering the idea was to spend less, I'm amazed he didn't mention power usage of having 2 cards vs 1 (best case you spend double in electricity, worst case you actually need a better PSU)
I would never consider a multi-GPU setup. I know from first hand that support on those setups ranges from pretty good to non-existent, but it is an area that receives only little attention once the initial multi-GPU launch is done (if there even is multi-GPU support). So future patches etc may introduce instability for those setups (and they are often unstable at launch with quite a few issues) and only luck or a driver update (1-2 months) will fix it.
Additionally, with ever increasing resolutions and complex rendering pipelines that need an increasing number of syncs between the individual GPUs; SLI and CrossFire bridges are becoming more and more of a bottleneck.
Why do you think they are so poorly supported? I know relatively few users have multi-GPU systems...it may be as simple as that, but I would have expected DX, OpenGL, and the new one from Valve (blanking right now, Vulcan?) would have improved usage and stability with minimal per-game development.
25 comments
[ 5.0 ms ] story [ 23.0 ms ] threadAshes of the Singularity is a game AMD had a big hand in making. There are driver level optimizations the studio paid for. Furthermore the game leverages D3D12's Async Compute which Pascal Architecture can't handle as well as GCN.
That being said in non-AMD sponsored games they two set ups in question (RX480 CFire vs GTX1080) do manage to stay within step of one another [1].
The real issue becomes overclocking and crossfire.
The GTX1080 has a much higher overclock threshold then RX480 (currently this may change as board partners make custome PCB's). So if you start fiddling with power levels this comparison breaks down.
Also crossfire/SLI is to a degree Russian Roulette. Studio's have to opt-in to supporting it, and sometimes don't. Leaving you only with a single card. Also very weird bugs can sometimes crop up. Some people have no issue with this, some have huge issues. YMMV.
The main thing I'm trying to express is when looking into hardware reviews CPU/GPU's more information from more sources is always better. And all sources/chips aren't equal.
[1] https://youtu.be/cVVJPbFRDEc
The problem is that DX12 shifts most of the responsibility for architecture-specific optimizations (and avoiding pathological cases) from the driver to the application. So although you could draw the conclusion that Nvidia are struggling with DX12 due to their Total Warhammer performance, it's just as easily explained by Nvidias veteran DX11 driver team being better at tuning for Nvidia hardware than a game studio who partnered with AMD during development.
In short: even if you exclude games that don't utilize multiple GPUs, a pair of RX 480s is slower than a GTX 1080 on average.
If you do include non-scaling games they're slower than a GTX 1070 on average, which is cheaper than the 480s and uses half the power.
I'm basically wondering what I should replace my current card with (I run Linux and play a few games like rocket league).
http://videocardz.com/61753/nvidia-geforce-gtx-1060-specific...
AMD drivers have serious bugs (that also affect the 480!) that they promised to fix a year ago, but not only they didn't fixed it, they also removed it from the "known issues" list on the changelogs. This was my first AMD purchase and will be my last, their hardware is mediocre (super power hungry and throttles because of that), software is outright terrible, and support from both them and Sapphire is nonexistent.
And when I asked for help on forums and chats, people accused me of being nvidia paid troll and even banned me. (And someone later pointed out to me that the places that most aggressively attacked me are the ones that have AMD employees as moderators)
1. Which game and which issue would that be? 2. Which driver removed it from the known issue list?
And I had problems with every single software, thus why I don't considered mentioning specifics.
Still, some examples:
CAD software stutters heavily because the card refuset to get out of idle state unless I am gaming, regardlesst of all power saving related settings.
Gtav was unplayable on most Crimson drivers, it works in the 16.6.1 but still has the occasional stutter due to power throttling.
Default fan profile is terrible, launching any AAA game without third party fan software makes the card overheat, release pcb varnish smell and make coil noise.
On default power limit the card throttles a lot and cause game stutter (all of them)
The CCCSlim panel on Crimson didn't let me use it because according to it I had multiple monitors with genlock enabled. I was using a single monitor, and genlock is a FirePro feature.
Driver crash a lot. A easy one to reproduce is entering the ingame herb encyclopedia on Shadow of Mordor.
Lastest drivers Mantle is buggy (known issues list), vulkan is also crappily implemented despite being AMD invention.
Browsers flicker on some driver versions for no obvious reason.
It also seems like RX 480s also overdraw the PCIe bus (drawing more than 75W from the slot--cards are supposed to be able to pull 75W from the slot, 75W from the six-pin connector, and 150W from the eight-pin connector), which is potentially dangerous for the rest of your hardware, too.
I went from being a steady AMD/ATI customer (last card was an HD4870) to a 980Ti and don't see myself going back anytime soon.
It's concerning at stock settings but gets far worse when the card is overclocked, with the card overloading the 12V pins by about 50%. AMD say they are working on a driver fix at least.
3/4-way SLI never worked particularly well in practice so it's not a great loss.
They deprecated 3/4-way SLI due to the difficulty of implicitly splitting the workload in the driver, but with non-SLI setups for CUDA/OpenCL the load-splitting is done explicitly by the application so it's not their problem.
http://www.anandtech.com/show/10067/ashes-of-the-singularity...
I don't expect many developers to actually use this functionality though - it's a lot of work to only benefit a tiny minority of users.
Additionally, with ever increasing resolutions and complex rendering pipelines that need an increasing number of syncs between the individual GPUs; SLI and CrossFire bridges are becoming more and more of a bottleneck.
Would not recommend.