-
Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.
Oxide: Nvidia GPU's do not support DX12 Asynchronous Compute/Shaders.
- Thread starter Arkanius
- Start date
I bought the XB27HU, its' the best IPS monitor right now. I'm using the 144Hz right now and the IPS (Came from an old TN monitor)
The G-sync upgrade later on would just be a bonus. I couldn't splurge money for a high end GPU and a monitor at the same time. I decided to do it on a phased roll-out[/QUOTE]
Thanks for this, makes a lot more sense. So basically Nvidia have banked on another architecture that is counter productive to the way DX12 works.
shiftplusone
Member
Is this "correctable" in a driver update or does it require completely new hardware?
Portugeezer
Member
Tomorrow Children uses async computing and that gave them a ~20% performance boost.
Dictator93
Member
I still think the OP and title are inaccurate. Heck, the dev does not even know about Maxwell 2 and there is competing information.
BriareosGAF
Member
There is no such thing as an async shader, there is an async pipe/ring (generally exposed as an 'async device') upon which compute shaders run.
These threads, so painful.
*shrug* this isn't a particularly rare. Intel drivers have been guilty of this. I wouldn't be surprised that this has happened with AMD drivers too.
This may contribute, but more importantly AMD GPUs had shifting and integer operations that Nvidia cards did not have until Maxwell. You should note that even pre-GCN AMD cards were faster than Nvidia cards. These operations made them much more suitable for hashing algorithms; AMD cards were often used in brute force password cracking for a quite a while.
Nvidia also reported the 970 as a 4gB GPU
Lemonte
Member
I'm starting to and my 970 isn't even 2 weeks old.
derExperte
Member
Has happened a few times over the years but AMD could never profit in a meaningful way so I'm not holding my breath but everything that helps them is good for us imo.
Hmm now I think about it, could be drivers, as Nvidia cards have their own queue engine.
(someone more knowledgeable correct me)
edit: Mahigan from Hardforum theorized that Maxwell 2 AWS does not function independently out of order with error checking.
Don't mix async timewarp [an engine rendering technique for sampling last user posistion as late as possible] with async compute [taking advantage from multiple hardware pipelines heading to the Compute Units and filling the "empty spots" in the rendering pipeline with more tasks]. AMD intentionally increased the number of ACE pipelines in GCN 1.2, so devs can extract full performance from their cards [esentially, all DX9/11 cards we used for years have wasted a lot of their performance].
First implementation of Async Compute in Tommorow Children on PS4 enabled bunching up of GPU tasks into the tighter schedule that was previously full of holes:
18% increase of performance just from going from traditional rendering pipeline to taking advantage of radeon ACEs [and that's just in a first implementation of the tech].
The version of GCN that Xbone uses have 4x less ACEs, and I think that Gforce cards have similar sparse setup.
edit - video explanation of the asynchronous compute
https://www.youtube.com/watch?v=v3dUhep0rBs
Portugeezer
Member
Doubtful. Nvidia has the marketing monies and AMD does not. AMD is going to have to have multiple games that trounce Nvidia's Pascal performance for a market shift like that. They may get a slight up-tick if some people lose their shit and decide to bank on what could be rather than what is.
dr. apocalipsis
Banned
Best DX12 improvements still being draw calls and memory management enhancements, so every GPU including Fermi or GNC 1.0 will enjoy the DX12 boost in performance.
That aside, Nvidia GPUs being bad at switching between compute and graphical tasks is old news. We all have suffered FPS tanking when activating Physx even on high end GPUs.
That aside, Nvidia GPUs being bad at switching between compute and graphical tasks is old news. We all have suffered FPS tanking when activating Physx even on high end GPUs.
justsomeguy
Member
Tough call then... Dx11 still the key performance requirement today. How long until async compute is too important to be ignored & dx11 support has faded?
Foutrologue
Banned
ASYNC Compute is part of DX12 so every GPU "supports" it even if it's not implemented in hardware which is the case with NVIDIA's current GPUs. To quote REDLYNX's Sebbi:
https://forum.beyond3d.com/posts/1868942/
So AMD GPU: HW Support for Async Compute
NVIDIA GPU: Async Compute code will run single thread
https://forum.beyond3d.com/posts/1868942/
So AMD GPU: HW Support for Async Compute
NVIDIA GPU: Async Compute code will run single thread
"Our belief is that by the middle of the PlayStation 4 console lifetime, asynchronous compute is a very large and important part of games technology." - Mark Cerny
More reading material from few months back:
http://www.anandtech.com/show/9124/amd-dives-deep-on-asynchronous-shading
http://wccftech.com/amd-improves-dx12-performance-45-gpu-asynchronous-compute-engines/
We are just about to enter the "middle of the PlayStation 4 console lifetime", and with the big help from AMD, low-level APIs will arrive at the same time on PC.
More reading material from few months back:
http://www.anandtech.com/show/9124/amd-dives-deep-on-asynchronous-shading
http://wccftech.com/amd-improves-dx12-performance-45-gpu-asynchronous-compute-engines/
We are just about to enter the "middle of the PlayStation 4 console lifetime", and with the big help from AMD, low-level APIs will arrive at the same time on PC.
dragonelite
Member
So you can just program async, with a penalty on nvidia hardware?
Maybe the performance gains can nullify the performance penalty.
Foutrologue
Banned
More up to date is UBI Soft/ REDLYNX's Siggraph presentation.
http://advances.realtimerendering.c...iggraph2015_combined_final_footer_220dpi.pptx
Most obvious gain of DX12 is decoupling DX11 CPU processing from old singlethreaded-focus into full multithreaded approach. After that, devs get full controll over GPU and memory they are accessing, including features such as async compute.
DX11 is here to stay because it is easy, but DX12 will be there for studios [or engines] who are focused on performance [pls Ubi, switch AC games [even old ones] to DX12 ASAP].
icecold1983
Member
the future doesnt look good for nvidia. even without this amd has been catching up quickly in performance as more and more games are designed around gcn because of consoles.
please no.
please no.
People wondering why Nvidia is doing a bit better in DX11 than DX12. Thats because Nvidia optimized their DX11 path in their drivers for Ashes of the Singularity. With DX12 there are no tangible driver optimizations because the Game Engine speaks almost directly to the Graphics Hardware. So none were made. Nvidia is at the mercy of the programmers talents as well as their own Maxwell architectures thread parallelism performance under DX12. The Devellopers programmed for thread parallelism in Ashes of the Singularity in order to be able to better draw all those objects on the screen. Therefore what were seeing with the Nvidia numbers is the Nvidia draw call bottleneck showing up under DX12. Nvidia works around this with its own optimizations in DX11 by prioritizing workloads and replacing shaders. Yes, the nVIDIA driver contains a compiler which re-compiles and replaces shaders which are not fine tuned to their architecture on a per game basis. NVidias driver is also Multi-Threaded, making use of the idling CPU cores in order to recompile/replace shaders. The work nVIDIA does in software, under DX11, is the work AMD do in Hardware, under DX12, with their Asynchronous Compute Engines.
I like this quote, where Nvidia's drivers gave them big boosts before, they're losing in DX12 because drivers aren't nearly as important.
Pretty much what AMD has been on about for the past 3 years. While Nvidia focused on the market as-is, AMD was getting in early on DX12 and it's cost them dearly in their already low market share. Now that DX12 is the current focus, I'm sure Pascal won't have these problems and it honestly can't if they want to compete. When Pascal comes around, we'll probably start seeing the first DX12 games hit the market and if their top-tier card is competing or even winning, that's all anyone is going to be talking about, not Maxwell performance.
I like this quote, where Nvidia's drivers gave them big boosts before, they're losing in DX12 because drivers aren't nearly as important.
Pretty much what AMD has been on about for the past 3 years. While Nvidia focused on the market as-is, AMD was getting in early on DX12 and it's cost them dearly in their already low market share. Now that DX12 is the current focus, I'm sure Pascal won't have these problems and it honestly can't if they want to compete. When Pascal comes around, we'll probably start seeing the first DX12 games hit the market and if their top-tier card is competing or even winning, that's all anyone is going to be talking about, not Maxwell performance.
Not sure, even UE4 doesn't seem to support it (except for X1)
icecold1983
Member
pascal doesnt seem to be a new architecture from the info available. its a maxwell 2 derivative with hbm and fp 16 support according to all current information.
The problem with Pascal is this:
A GPU Architecture takes around 3-4 from planning to fruition.
I honestly want to believe that Nvidia had the vision 4 years prior of seeing they need to go Parallel.
DX12 started being talked last year, this year it was released. Vulkan comes next year, so does Pascal.
I'm not that convinced that Nvidia is ready
dr. apocalipsis
Banned
People painting an obscure future for Nvidia is funny, because even on that biased Ashes of the Singularity benchmark, geared towards AMD hardware even when the engine it's based on doesn't shows that strange behaviour, still sees better performance on Nvidia side.
Usual AMD forums fud.
Usual AMD forums fud.
Coulomb_Barrier
Member
IIRC it was 290X versus 980Ti which is two grades above and twice the price
Then again, Fury didn't shine so more data is indeed needed.
NumberThree
Member
dr. apocalipsis
Banned
Any reason for Nvidia not to switch to parallel architecture for Pascal onwards? By the time a meaningful amount of DX12 games are out (2~3 years maybe?) I guess most people will probably be in for a GPU upgrade anyway. And maybe I'm missing something but I can't see any massive differences in the benchmark graphs; I only skimmed through the computerbase article tho.
Nostremitus
Member
Nvidia has the marketshare and the money to focus on what works best "now" then release something different when it's needed. AMD doesn't have that luxury. They have to aim at future proofing because they don't have the R&D budget to constantly adapt like Nvidia. So, when they release a new microarchitecture, it's aimed a few year ahead and may not play well right off the bat with what is currently popular. This also means that when AMD has a misstep, they are stuck with that mistake for years (as with their CPUs.)
yes
Dictator93
Member
I am pretty sure NVs research budget is not about "the now" and AMDs "is about the future". The world is more diverse than the picture you paint. Because if you look at any other data point than async compute (which still has not even been proven that Maxwell 2 does not support it) you would see NV also has hardware features which are "future oriented".
dragonelite
Member
Coulomb_Barrier
Member
This pic should be on OP, imo.
What does this prove, exactly?
thequickandthedead
Member
Pardon my ignorance but what main feature are you talking about? I thought G-Sync monitors just eliminated screen-tearing by syncing your GPU to the monitor's refresh rate?
Poetic.Injustice
Member
TEH BIAS.
icecold1983
Member
they dont seem to be nearly as beneficial as async compute. also, consoles not supporting them pretty much makes them useless. you will see the occasional 2 titles a year use them to do almost nothing to enhance the visuals. they will probably become the new tessellation.
