Associate
- Joined
- 27 Dec 2014
- Posts
- 1,686
- Location
- Southampton
wow, this thread has heated up quite a bit.
-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
GTX 1060 Vs RX 480 - head to head showdown
- Thread starter Gregster
- Start date
Nvidia keeps saying it support it, it will be on next driver we promise etc etc. How many months of it not been there before you give up?
I don't need to understand how dx12 works or how async works. However it works, Nvidia ain't using it and dont look like they ever will. Or if it does, then it makes no difference in the fps so who cares????
GCN get double the boost pascal does though..................
SO its the game developers fault and not a case, as in doom, that Nvidia are still "working" on a driver to support it?
This. Does anyone really thinks that ID Software deliberately choose to not include the required support for Nvidia cards with the Vulkan API?
From what I can see...
GTX 1060 seems like the safer SHORT TERM solution. If you plan to move on from the card within a year this is a safe bet card to buy. It performs better with most DX11 titles (that I've seen in reviews).
GTX 480 is the better LONG TERM solution (1.5 years plus) It performs better (marginally) in DX12 situations and has other things going for it like hardware encoding (I could be wrong with the encoding).
Obviously things could change over time for either. Not sure what to make of overclocking at the moment on either card.
GTX 1060 seems like the safer SHORT TERM solution. If you plan to move on from the card within a year this is a safe bet card to buy. It performs better with most DX11 titles (that I've seen in reviews).
GTX 480 is the better LONG TERM solution (1.5 years plus) It performs better (marginally) in DX12 situations and has other things going for it like hardware encoding (I could be wrong with the encoding).
Obviously things could change over time for either. Not sure what to make of overclocking at the moment on either card.
Especially since they worked with NVIDIA to show off DOOM on Vulkan at GDC for Pascal.
Do you have a link to where NVidia said it would be in the next driver please. I will give them some what for, as there has been 3? drivers and Pascal still isn't showing gains in Vulkan
It was already stated by futuremark itself, that they only wrote a single render path which only pushes things as far as Pascal could go. They call this the unbiased solution.
Its like a car race where they force a speed limit which happens to be the slowest cars max speed.
At GDC NV and AMD together had a DX12 presentation where people from both company talked, and stated, that if you don't use hardware specific render paths you shouldn't touch DX12, just stay on DX11.
Associate
- Joined
- 19 Aug 2008
- Posts
- 25
So I'm looking for a new graphics card in the 200-250 quid range and thought I'd check the forums while I watched the stock on the main site... some of these threads are a ridiculous.
I currently write software that uses D3D12 as a part of my living (AMA on root sigs, descriptor tables etc. if you like), not cryengine mods or unity/unreal with D3D12 support, but actually work with the API directly.
Futuremark didn't mess up.
In D3D12 you build command lists on the GPU and submit them to command queues for execution. There is no vendor specific way to submit command lists. There is no AMD async path and/or Nvidia Async path. You execute your graphics lists on a direct queue, and typically submit any compute command lists on a compute queue. Then if the hardware supports some form of async you'll get a lower frametime than submitting everything only to the direct queue.
It is true that you can and sometimes should tailor your render path to different hardware to improve performance in D3D12. That's also been true in D3D11. There are things you can do that unfairly cripple one vendor's performance while the other handles it fine. Some hardware prefers shared data in a constant buffer, while other GPUs work faster if you put that data in a 1D buffer. Some lose half their export rate if you render to certain buffer formats, and others **** the bed if you pass more than 64 bytes between shader stages, or pass a non-multiple of 16 bytes.
I already stated there's no alternative method to submit compute commands that favours one vendor in D3D12. But you could conceivably change the commands themselves depending on the detected hardware. Maybe one vendor prefers a larger or smaller threadgroup size, or whatever. That's a separate render path isn't it? So what are you now measuring in your benchmark, the performance of different hardware or the performance of different threadgroup sizes? You can't fault Futuremark for making the hardware/drivers the only variables in a hardware benchmark.
I hope you guys get past this whole "that's not real async" thing. Go look around beyond3d and see what known developers think of it - https://forum.beyond3d.com/threads/no-dx12-software-is-suitable-for-benchmarking-spawn.58013/page-17
All I wanted was to read some discussion about the 480 vs 1060 to help me decide, I didn't expect to see so many pages on AMDs very specific meaning of asynchronous or how mistaken futuremark are.
I currently write software that uses D3D12 as a part of my living (AMA on root sigs, descriptor tables etc. if you like), not cryengine mods or unity/unreal with D3D12 support, but actually work with the API directly.
Futuremark didn't mess up.
In D3D12 you build command lists on the GPU and submit them to command queues for execution. There is no vendor specific way to submit command lists. There is no AMD async path and/or Nvidia Async path. You execute your graphics lists on a direct queue, and typically submit any compute command lists on a compute queue. Then if the hardware supports some form of async you'll get a lower frametime than submitting everything only to the direct queue.
It is true that you can and sometimes should tailor your render path to different hardware to improve performance in D3D12. That's also been true in D3D11. There are things you can do that unfairly cripple one vendor's performance while the other handles it fine. Some hardware prefers shared data in a constant buffer, while other GPUs work faster if you put that data in a 1D buffer. Some lose half their export rate if you render to certain buffer formats, and others **** the bed if you pass more than 64 bytes between shader stages, or pass a non-multiple of 16 bytes.
I already stated there's no alternative method to submit compute commands that favours one vendor in D3D12. But you could conceivably change the commands themselves depending on the detected hardware. Maybe one vendor prefers a larger or smaller threadgroup size, or whatever. That's a separate render path isn't it? So what are you now measuring in your benchmark, the performance of different hardware or the performance of different threadgroup sizes? You can't fault Futuremark for making the hardware/drivers the only variables in a hardware benchmark.
I hope you guys get past this whole "that's not real async" thing. Go look around beyond3d and see what known developers think of it - https://forum.beyond3d.com/threads/no-dx12-software-is-suitable-for-benchmarking-spawn.58013/page-17
All I wanted was to read some discussion about the 480 vs 1060 to help me decide, I didn't expect to see so many pages on AMDs very specific meaning of asynchronous or how mistaken futuremark are.
Soldato
- Joined
- 28 Sep 2014
- Posts
- 3,577
- Location
- Scotland
Bethesda released DOOM patch supported Vulkan last week.
https://community.bethesda.net/thread/54585?tstart=0
I don't know if it will be in next driver but I think it probably will take Bethesda few months to add Nvidia Async Compute support in future patch.
I don't see how, or why. NVIDIA have been working with them on Vulkan for a while now.
So much so the first time we even saw Vulkan DOOM running was at the NVIDIA presentation at GDC to show off Pascal. NVIDIA were also the first of the two companies to get proper Vulkan drivers, and support out.
I think the hold up might be more on NVIDIA's side, since they need a driver/software based solution for concurrent pre-emption; while AMD's hardware ACE's just do the work themselves. Although I'm sure they'll get the driver/game patch out in due time.
Last edited:
Caporegime
- Joined
- 9 Nov 2009
- Posts
- 25,458
- Location
- Planet Earth
This and when it comes to DX12 and Vulkan I am getting fed up of all the excuses made by some that Nvidia is fine and that Nvidia can fo async fine,it just needs to be tailored to their way of doing things.
Yet,apparently the company which spends more on software,games optimisations and engineers according to some here,still seems to have problems in DX12 and Vulkan overall.
Doom is a hilarious one - I tried running Vulkan on my GTX960 4GB and it ran slower than OpenGL.
My mates ancient HD7870XT which is a slower card than my GTX960 got a decent performance boost when switching to Vulkan - that is a GCN1.0 based card FFS,which came out long before all this DX12,Vulkan and even Mantle malarkey.
Plus Doom is an iD game and we saw how previous iD games had issues on AMD cards - now suddenly its a 3DMark benchmark which is more important than actual games.
The studio is owned by Bethesda and if you look at games like FO4,they worked closely with Nvidia.
So when games from devs who tend to work more closely with Nvidia start showing improvements in Vulkan that favour AMD cards,that tells you something.
Last edited:
if you playing bf1 at 1080 res any will be fine.any res above i wouldn't buy any for bf1.they will struggle.
The RX480 vs GTX 1060 tested in DX12 and Vulkan Games:
http://www.golem.de/news/geforce-gtx-1060-vs-radeon-rx-480-das-bringen-direct3d-12-und-vulkan-1607-122214.html
We can see the stock 480 is beating te 1060 in most of the games so a good indicator of things to come. The custom 480's will no doubt increase the lead.
http://www.golem.de/news/geforce-gtx-1060-vs-radeon-rx-480-das-bringen-direct3d-12-und-vulkan-1607-122214.html
We can see the stock 480 is beating te 1060 in most of the games so a good indicator of things to come. The custom 480's will no doubt increase the lead.
Yea AMD is doing way better in Vulkan and dx12 games. Theres no point in looking at dx11 benchmarks because those are a thing of the past now. All current cards can run dx11 good enough and we aren't going to see much more on that front.
Last edited:
Do people stop playing DX11 games now then as well?
For people on a 3 year upgrade cycle (and there are many) Future performance is more important then what is the norm today.
Getting a 1060 doesn't make any sense in that regard