apparently even 980Ti can't handle async-compute without a crippling context switch

[soccerballtux]

This is an interesting development. We'll be seeing more benefits of this soon for AMD with all the DX12 titles coming out later this year, and in the future with the console ports.

Maxwell (970/980/Ti/etc) SUPPORTS async-compute, but not without a performance crippling context-switch (that's baked into the Maxwell hardware design) for every compute task--which kills the whole point of having it available for games in the first place

Considering pipelining the render stage is the new hotness and because it enabled NaughtyDog to triple their framerate from 12fps to ~30 in the 3 weeks before release, meeting the release deadline, this is a pretty significant oversight for Nvidia, and another boon for AMD consumers.

As to why Async Compute/Shaders are so important in DX12 & future cross-platform games:

• Compute is used for global illumination, dynamic lighting, shadows, physics, post-processing (including even AA). If it can be offloaded from the main rendering pipeline and done asynchronously in parallel, it can lead to major performance gains. As such, GPUs that support it will see major performance uplift and in theory, GPUs that do not support it, will have no benefit, it reverts back to the normal serial rendering of graphics & compute.
  
• Async Shaders are vital for a good VR experience, as it helps lower latency of head movement to visual/photon output. I posted on this topic awhile ago...

Anyways, I don't really have anything more to say about this, I'm just excited to keep my 7950 for another 4 years. Discuss.

https://www.reddit.com/r/pcgaming/comments/3j1916/get_your_popcorn_ready_nv_gpus_do_not_support/

 

[ShintaiDK]

https://forum.beyond3d.com/threads/dx12-performance-thread.57188/page-5#post-1868411

Not to get too into the weeds here, but what exactly is the benchmark doing? You have to be a bit careful when "testing async compute" as the notion isn't even a well-defined concept. Depending on the specifics of the load on the various units on the machine, certain architectures may or may not get performance benefits from async queues, but it's certainly not as simple as "supported or not". Similar notion with async copy as well: you're providing the driver/hardware with some additional parallelism that it may be able to make use of to improve performance, but it depends highly on the specific characteristics of both the net workloads running on the machine at the time, and the architecture.

And let's remember, an ideal architecture would not require additional parallelism to reach full throughput, so while the API is nice to have, seeing "no speedup" from async compute is not a bad thing if it's because the architecture had no issues keeping the relevant units busy without the additional help It is quite analogous to CPU architectures that require higher degrees of multi-threading to run at full throughput vs. ones with higher IPCs.

 

[cmdrdredd]

I'll probably have a new card by the time this really matters to us in terms of games available. I want DX12 to take off but I don't think it'll be as quickly supported as I hoped.

Anyway we will see what the actual games do when they are out to retail using DX12.

 

[Carfax83]

AMD also uses context switches to perform asynchronous compute. And you do realize there's a massive thread devoted to this very topic right? Why on Earth would you start a new one?

 

[Gikaseixas]

IMO

- By the time DX12 games arrive, 28nm will be irrelevant
- Fury, Fury X might surpass TI and Titan x but again, at unplayable frame rates
- Pascal will most likely have Async Compute
- Arctic Islands might have an edge since AMD will have more experience tweaking it's drivers

 

[Magee_MC]

https://forum.beyond3d.com/threads/dx12-performance-thread.57188/page-5#post-1868411

As I understand it, that bolded statement is irrelevant in regards to Maxwell and async compute. As it stands there is no way to know if Maxwell would benefit from async compute because the needed context changes prevent that experiment from being run.

The key phrase is "if it's because the architecture had no issues keeping the relevant units busy without the additional help". As it stands there isn't a way for us to know whether or not this is the case.

 

[cmdrdredd]

Saw this post on reddit. Summarizes my feelings. I don't regret buying my 970s. The performance for what I paid is worth it. I don't expect a first gen DX12 part to be flawless in DX12. That wasn't the case for DX11 or DX10 either, it took newer cards to really push the performance up.

Every game that I currently own, and every game due out in the near future that I have an interest in, is ≤ DX11.

I care more about the fact that AMD are still single-threaded in DX11, while NVIDIA are multi-threaded in DX11, than something which may be an issue a year or two down the line.

What I've seen recently is that more and more games seem to have CPU-limited performance rather than being GPU-limited. Now obviously DX12 is the solution for that, but it doesn't help existing games which are running single-threaded on AMD.



But I'm also of the opinion that none of the current-generation GPUs are "future-proof" at the moment, and people are kidding themselves if they think that. It's the reason why I ended up buying a mid-range card (GTX 960) instead of something high-end.

Nothing currently does 4K gaming well - at least not if you are trying to play any of the "big-budget games" released in the last year or two.

Consumer versions of VR headsets are yet to ship, keep being pushed back, and VR is going to have significant performance demands - possibly more than trying to run games at 4K and 60+ FPS.

What we typically see is that GPUs built around a new generation of DirectX have significantly better performance than the older GPUs which are able to support it. That has been true for DirectX 9, 10, 11 etc. and I don't see any reason to think that it will be different with DX12.

By the time that DX12 and VR gaming actually matters, we'll be on to the next generation of GPUs that are shipping with a new architecture, on a smaller process, with 8-32GB HBM2, and with DisplayPort 1.3 connections to properly support high framerates at high resolutions. (VR, 4K and beyond)

Make your purchases based on what actually matters today, not what may happen at some point in the future.

I could have built a PC "for VR" in 2012 when the Oculus DK1 was shipping, and the requirements for CV1 or the Vive in 2016 are going to be very different.

 

[ShintaiDK]

Its a pre alpha benchmark of a game.
Its AMD sponsored.
Drivers and OS are still in early state.
Async compute is only needed on some uarchs.

Pick between hysteria and viral marketing.

What happens if ARK, Fable or some other DX12 game shows a very different picture. Sample size matters.

 

[TheELF]

and because it enabled NaughtyDog to triple their framerate from 12fps to ~30 in the 3 weeks before release, meeting the release deadline

Wow,this PS4 cpu is so wonky that it would only give you 12 FPS on it's own?

 

[poofyhairguy]

IMO

- By the time DX12 games arrive, 28nm will be irrelevant
- Fury, Fury X might surpass TI and Titan x but again, at unplayable frame rates
- Pascal will most likely have Async Compute
- Arctic Islands might have an edge since AMD will have more experience tweaking it's drivers

-Nvidia will make sure most of the big games have optimizations that benefit their hardware and not AMDs anyway

It really doesn't matter to me personally. I only buy games a year after release for half price or less, which means even if I am wrong and EVERY developer moves to this technology it will be three plus years until it effects me. At that point I might just get a cheap end-of-gen PS4 and play all the exclusives I missed until the market settles.

 

[3DVagabond]

I'll probably have a new card by the time this really matters to us in terms of games available. I want DX12 to take off but I don't think it'll be as quickly supported as I hoped.

Anyway we will see what the actual games do when they are out to retail using DX12.

Remember all those devs who signed up for Mantle to get a leg up with DX12? Maybe it'll take off faster than some think?

 

[el etro]

IMO

- By the time DX12 games arrive, 28nm will be irrelevant
- Fury, Fury X might surpass TI and Titan x but again, at unplayable frame rates
- Pascal will most likely have Async Compute
- Arctic Islands might have an edge since AMD will have more experience tweaking it's drivers

It looks like will be exactly this that will happen.

 

[tential]

no longer relevant comment

 

[cmdrdredd]

Remember all those devs who signed up for Mantle to get a leg up with DX12? Maybe it'll take off faster than some think?

Dunno, have to see.

 

[soccerballtux]

AMD also uses context switches to perform asynchronous compute. And you do realize there's a massive thread devoted to this very topic right? Why on Earth would you start a new one?

no, I didn't

the context switches aren't baked into the hardware.

 

[DooKey]

No biggie to me. I'll buy whatever it takes to get top performance.

 

[Keysplayr]

AMD also uses context switches to perform asynchronous compute. And you do realize there's a massive thread devoted to this very topic right? Why on Earth would you start a new one?

Its like Mantle all over again.

 

[poofyhairguy]

Can we get any more condescending to a development studio?

Sure we can. I can pull some Daikatana reviews out of the Way Back Machine if you don't believe me.

Edit: And I deleted that part from my post before you posted for a reason. The viability of development houses embracing Directx 12 doesn't matter. I applaud AMD for pushing the standard in gaming and I hope that future games are more advanced because of it.

 

[cen1]

Too much rumors and people talking about things they have no idea how they actually work. I'll rather wait for a reputable source to research into this.. like Anandtech, although they are quite slow these days.

I'll admit that watching various forum flame wars is kinda entertaining tho.

 

[boozzer]

Sure we can. I can pull some Daikatana reviews out of the Way Back Machine if you don't believe me.

Edit: And I deleted that part from my post before you posted for a reason. The viability of development houses embracing Directx 12 doesn't matter. I applaud AMD for pushing the standard in gaming and I hope that future games are more advanced because of it.

without mantle lighting a fire under microsoft's ass, we will never have dx12. not this early or how it is made anyway. :biggrin:

 

[poofyhairguy]

without mantle lighting a fire under microsoft's ass, we will never have dx12. not this early or how it is made anyway. :biggrin:

Maybe. It seems like the weak CPU in the Xbox One was just as big of a factor.

 

[NTMBK]

*grabs popcorn* Should be fun to watch what happens with these new engines. Wonder how much the new Deus Ex will use async shaders?

 

[Snafuh]

*grabs popcorn* Should be fun to watch what happens with these new engines. Wonder how much the new Deus Ex will use async shaders?

In some Siggraph paper about Deus Ex there was async compute mentioned. Same for the new Tomb Raider game.
Gears of War and Fable will be the first proper DirectX 12 game. Let's see how AMD and NVidia GPUs will perform there.

 

[x3sphere]

This will backfire horribly for AMD if only a select few DX12 games actually end up being faster. People are already setting up unrealistic expectations off of one game, which as an RTS seems to rely heavily on compute and draw calls. Not all necesssarily will.

If they do get a huge perf advantage, great. Still an OC'd 980 Ti is around 20% faster than Fury at 1440p - so async compute would really have to give a ~30% or so advantage to topple that. And they still may be considered even in some eyes as it's not like everyone stopped playing DX11 games.

 

[Arachnotronic]

Seems like a great selling point for NVIDIA's next GPU relative to prior generation ones