Comparing the AMD RX480 and Nvidia GTX1060

[FordGT90Concept]

GCN Vulkan does appear to be CPU restricted. It's not really clear why at this point and I haven't seen anyone really dive into it.

 

[RejZoR]

You need to feed GPU faster if it's doing more work in parallel (ie finishing given tasks faster).

DX11/OGL workload analogy:
I mean, imagine 4 excavators (4 CPU cores) filling up 1 truck (GPU) that goes up and down the highway (CPU-GPU path), depicting single threaded workload.

DX12/VK workload analogy:
Vulkan is essentially same 4 trucks going up and down the highway, but you still have just 4 excavators filling them up. You either need excavators that work faster (higher IPC) or more of them.

This is especially obvious since GPU doesn't have a single workload thread anymore, but several of them, menaing it will most likely also request more work from CPU which can be instructed to use more cores in parallel as well, far more than it does now.

 

[FordGT90Concept]

The CPU shouldn't be doing much/any more work on GTX 1060 compared to RX 480.

 

[RejZoR]

In heavy multithreaded workloads (DX12/Vulkan), yes, it should/would.

 

[Nergal]

If the guy feels better with his 1060, than that´s that. Personally in his situation I would indeed first try to make the 480 "smoother"; but if I don´t know how/can´t bother with it; I would also take the 1060. Apparently he is financially well off enough to have both a 480 + 1060 and considering to be rid of them both to buy something else in the near future.

In that case, dear OP, look into getting a quiet 1070.
Selling both 1060+480 now will get you almost there anyway. (I assure you, all games in FULL HD will run smooth for a long time to come)

 

[arbiter]

You need to feed GPU faster if it's doing more work in parallel (ie finishing given tasks faster).

DX11/OGL workload analogy:
I mean, imagine 4 excavators (4 CPU cores) filling up 1 truck (GPU) that goes up and down the highway (CPU-GPU path), depicting single threaded workload.

DX12/VK workload analogy:
Vulkan is essentially same 4 trucks going up and down the highway, but you still have just 4 excavators filling them up. You either need excavators that work faster (higher IPC) or more of them.

This is especially obvious since GPU doesn't have a single workload thread anymore, but several of them, menaing it will most likely also request more work from CPU which can be instructed to use more cores in parallel as well, far more than it does now.

The CPU shouldn't be doing much/any more work on GTX 1060 compared to RX 480.

In heavy multithreaded workloads (DX12/Vulkan), yes, it should/would.

How would would load in vulkan using either rx 480 and gtx1060 be different? Which gpu is in the machine shouldn't make 1 bit of difference in the work load as FordGT90Concept said. AMD loved the idea mantle would make lower end hardware more effective but old issue of drivers being inefficient is cropping back up. I remember an article some time ago before dx12/vulkan/mantle, that when compared next to each other Nvidia cards had noticeably lower cpu load compared to AMD cards. Kinda sounds like that load issue might be coming back to rear its head on lower end cpu's were a gtx1060/RX480 would most likely find its home in.

This I think should be something looked in to on future vulkan/DX12 games to see if its a recurring trend or just some oddity in just doom

 

[FordGT90Concept]

Drivers are the logical place to look. It would be interesting to compare a GCN card in a game like Talos Principle running Vulkan on Linux and Windows. The drivers have to be different in that case but performance should reasonably be similar. If changing CPUs across operating systems don't get near equal results, it should become pretty apparent what driver needs love from AMD.

 

[RejZoR]

How would would load in vulkan using either rx 480 and gtx1060 be different? Which gpu is in the machine shouldn't make 1 bit of difference in the work load as FordGT90Concept said. AMD loved the idea mantle would make lower end hardware more effective but old issue of drivers being inefficient is cropping back up. I remember an article some time ago before dx12/vulkan/mantle, that when compared next to each other Nvidia cards had noticeably lower cpu load compared to AMD cards. Kinda sounds like that load issue might be coming back to rear its head on lower end cpu's were a gtx1060/RX480 would most likely find its home in.

This I think should be something looked in to on future vulkan/DX12 games to see if its a recurring trend or just some oddity in just doom

I'm talking about async, I don't know what you two are talking about. There is no "oddity" about it, it's how these API's work. You have more direct control of hardware (lower overhead) and you have async compute. I never said there is any difference between RX480 and GTX 1060 assuming GTX 1060 has an actual functional async (which is still debatable even though people say GTX 1000 series have it).

You don't seem to understand that while games will generally perform better with these API's, if you'll want to push them further, it'll all depend on CPU being able to feed all the GPU demands. What this means in a nutshell is that gap between lower and top end GPU's will become smaller in terms of framerate if CPU is the same in both cases.

 

[FordGT90Concept]

Both cards utilize async compute in Vulkan. It in no way explains the massive drop of FPS on RX 480 on slower CPUs. Async increases GPU utilization for very little additional load on CPU (just doing the tasks now rather than later for each frame).

I think the most plausible explanation isn't that the CPUs are older, but the clockspeed is much lower. Something is taking many CPU cycles that the GPU is waiting for that gets done almost twice as fast on the 4.5 GHz 6700K. Whatever that "something" is doesn't exist on GTX 1060.

As far as I know, the none of the CPUs tested hit 100% load. If it did, then yeah, that explains the drop in FPS. Otherwise, there is something quirky that needs to be examined.

 

[Xzibit]

Both cards utilize async compute in Vulkan. It in no way explains the massive drop of FPS on RX 480 on slower CPUs. Async increases GPU utilization for very little additional load on CPU (just doing the tasks now rather than later for each frame).

I think the most plausible explanation isn't that the CPUs are older, but the clockspeed is much lower. Something is taking many CPU cycles that the GPU is waiting for that gets done almost twice as fast on the 4.5 GHz 6700K. Whatever that "something" is doesn't exist on GTX 1060.

As far as I know, the none of the CPUs tested hit 100% load. If it did, then yeah, that explains the drop in FPS. Otherwise, there is something quirky that needs to be examined.

Not in DOOM

Does DOOM support asynchronous compute when running on the Vulkan API?

Asynchronous compute is a feature that provides additional performance gains on top of the baseline id Tech 6 Vulkan feature set.
Currently asynchronous compute is only supported on AMD GPUs and requires DOOM Vulkan supported drivers to run. We are working with NVIDIA to enable asynchronous compute in Vulkan on NVIDIA GPUs. We hope to have an update soon.

 

[FordGT90Concept]

Ah, I was thinking Time Spy which uses async compute on GCN and Pascal.

From that perspective, async compute could be to blame. If it is, that's pretty alarming in terms of CPU load.

 

[Kanan]

Doom is heavily AMD biased anyway in Vulkan mode . It has some kind of vsync turned on on nvidia Gpus that drops fps quite a bit.
See here:
http://www.pcgameshardware.de/Doom-2016-Spiel-56369/Specials/Vulkan-Benchmarks-Frametimes-1202711/

The adjusted graphs show that Vulkan perf on nvidia Gpus is normally higher, if you deactivate that strange vsync like "feature". Also Vulkan perf on nvidia is still generally bad compared to OpenGL.

 

[RejZoR]

"Strange" async he said. XD That's like saying Intel's HT is a "strange" feature...

 

[Kanan]

"Strange" async he said. XD That's like saying Intel's HT is a "strange" feature...

I spoke about vsync not async. Also read the article (Google translator maybe).

 

[RejZoR]

What has Vsync to do with anything? You either have Vsync or you have a framerate limiter. One syncs to monitor refresh, other just limits framerate to certain range and can still be out of sync even if using same value as monitor refresh (60fps+60Hz). Don't see what "else" one would use.

 

[ChristTheGreat]

I did a comparison of PowerColor Red Devil RX 480 (1330 MHz) and the Founder's Edition GTX 1060 at 1920x1080. 4 tests were DX12 where the RX 480 won 3 of the 4 (Rise of the Tomb Raider); 8 tests were DX11 where the GTX 1060 won 7 of the 8 (Alien: Isolation). The cards are well matched:

DX12:
GTX 1060=65.75 fps
Red Devil = 68 fps

DX11:
GTX 1060=80.25 fps
Red Devil = 76.13 fps

Average for both:
GTX 1060 = 75.41 fps
Red Devil = 73.42 fps

Source: http://www.guru3d.com/articles-pages/powercolor-radeon-rx-480-red-devil-review,1.html

[sarcasm]As you can see, a whooping 2 FPS difference (2.7%)! [/sarcasm]


If you plan on holding on to the card for >3 years, I'd go with RX 480. If you don't, I'd go with GTX 1060. Not only is the RX 480 likely to last that long because it is built like a tank, it is also going to perform better in DX12/Vulkan games which will be widely available by then.

If planning to run 1080P games. I don't think higher resolution, in 3 years will be enought. as for DX12, this is truely true, AMD card are doing quite well, (same on Vulkan)

I would like to compare that PowerColor Red Devil RX 480 with an overclocked GTX 1060.

 

[rtwjunkie]

I've decided to keep the 1060 at least for now... once more DX12 games are released I may sell the 1060 and get a custom cooled RX480 depending on performance.

By the time there are enough DX12 games for that to matter the next gen will be here. It really matters not right now what you buy as far as DX12 thinking.

 

[medi01]

I think one needs to dig down to how that "became faster" actually happened, as there might be several ways.

1) AMD bumps up perf of the card by 10%+ over 1-3 years
2) nVidia abandons old card optimizations for NEW games, while AMD continues to support them, new games gradually replace old games, so AMD card climbs in ranks

Besides the "faster CPU" option that was just mentioned.

By the time there are enough DX12 games for that to matter the next gen will be here.

That's a false approach, to be honest.
It really depends on what one plans to play. If Doom is your thing, Vulkan is right there for ya, today, you shouldn't wait. The same with BF1.

 

[RejZoR]

And how many DX11 games are unplayable on RX480? Unless you're stupid enough to insist on using 4K screen, all games will work at high framerate regardless.

 

[Fluffmeister]

And how many DX11 games are unplayable on RX480? Unless you're stupid enough to insist on using 4K screen, all games will work at high framerate regardless.

Ditto for the 1060? And that seems less CPU dependent.

Swing and roundabouts.

 

[rtwjunkie]

That's a false approach, to be honest.
It really depends on what one plans to play. If Doom is your thing, Vulkan is right there for ya, today, you shouldn't wait. The same with BF1.

How is it false? It's a direct reply to the OP, who says nothing about Doom. He stated his intent was to wait until there were more DX12 games before making a permanent choice of 1060 or 480.

My response was by the time you reach that point, the next gens will be here and it will be time to pick a new card, not these cards. So, with his thinking, just pick a card, any card is what I am saying.

 

[yogurt_21]

If it were me, I'd go 480 merely because you'll be more likely to sell the 1060 for more than you paid for it and the performance is close.

 

[medi01]

My response was by the time you reach that point, the next gens will be here and it will be time to pick a new card, not these cards.

My point is, you don't really play "a gen". I know a lot of people who stick with a couple of games (I do!) and when it's more than that, perf differences... are hardly noticeable (e.g. adventures).

Second thing is, that you assume that people buying cards in this tier upgrade them every year or two. Which again, is not the case. (and rightly so, as older cards continue to be more than adequate for newer games for much longer than that)

 

[Recon-UK]

Comparing Apples to Pears.

 

[Dethroy]

Comparing Apples to Pears.

Don't make this thread about operating systems...