AMD Explains Asynchronous Shaders on DirectX 12

Gee who cares about DX12 now anyway... The majority of games won't be using it to its full extend before 2-3 years anyway. So much will have changed by then.
 
Who cares? Gamers for one. Won't take 2-3 years, games are already starting to move on up.

Right... Look at how long it took the game devs to finally scrap DX9.

Edit: Guess I shouldn't have said "who cares". It's actually kinda rude :)
 
Last edited:
Pretty sure the takeup for DX11 took along time (heck we're still getting alot of AAA DX10 games) so all this excitement won't mean anything until we start seeing the vast majority of games using DX12 or even Vulkan.

Lets not turn this into an AMD circle jerk and be realistic. nVidia have a much better funded R&D team and also have the potential to see games are more tailored towards them (be it through financial trickery or marketing). I'd love to see AMD come out, make some cash on top and force nVidia to try hard. AMD still however have to get a couple of basic things right in order to even hope of gaining ground, even with this DX12 lead.

nVidia are still top dog and won't let go of their place without a fight. Even if it means they have to really innovate.
 
Pretty sure the takeup for DX11 took along time (heck we're still getting alot of AAA DX10 games) so all this excitement won't mean anything until we start seeing the vast majority of games using DX12 or even Vulkan.

Lets not turn this into an AMD circle jerk and be realistic. nVidia have a much better funded R&D team and also have the potential to see games are more tailored towards them (be it through financial trickery or marketing). I'd love to see AMD come out, make some cash on top and force nVidia to try hard. AMD still however have to get a couple of basic things right in order to even hope of gaining ground, even with this DX12 lead.

nVidia are still top dog and won't let go of their place without a fight. Even if it means they have to really innovate.

It's been realistic so far, AMD have the lead currently.. and on a hardware level too. Driver's won't be as efficient and are more limited than the hardware way. That's realistic. AMD currently have the less buggy drivers it seems. Sure they don't release as often but they have released 2 within a month, that's a huge progress in the driver department for them. So as of right now, they have a driver lead and hardware lead(both for DX12). No circle jerk'n here. It's just how it is. In the future it's really up to how Nvidia respond and if they can get the Async Shaders supported on a hardware level for Pascal.
 
Last edited:
It's been realistic so far, AMD have the lead currently.. and on a hardware level too.

How so? All that has really happened so far is that they've been making alot of noise about DX12. Its just marketing BS.

The whole benchmark thing with the 290X being as fast as the Titan X ended up with Futuremark saying 'do not use it to compare graphics cards' because its an API test. Until I see actual gains in DX12 GAMES (not benchmarks), be it with my 7870 or my 780Ti I'm going to stay skeptical and presume that all this noise is just AMD trying to hype people into buying their gpus. Don't get me wrong, it'd be cool if AMD wipes the floor but as I said earlier, this IS nVidia and they do have talent and the finances. Even if they do prefer to pay people off occasionally..
 
How so? All that has really happened so far is that they've been making alot of noise about DX12. Its just marketing BS.

The whole benchmark thing with the 290X being as fast as the Titan X ended up with Futuremark saying 'do not use it to compare graphics cards' because its an API test. Until I see actual gains in DX12 GAMES (not benchmarks), be it with my 7870 or my 780Ti I'm going to stay skeptical and presume that all this noise is just AMD trying to hype people into buying their gpus. Don't get me wrong, it'd be cool if AMD wipes the floor but as I said earlier, this IS nVidia and they do have talent and the finances. Even if they do prefer to pay people off occasionally..

Good thing that the only benchmark out also just so happens to Be built in a game... It's not hype. The results are there for you. The game and benchmark use asynchronous compute. It's comparable. The fact the game releases this year also makes it even more valid as it's not a tech demo. AMD have hardly really made any noise, Nvidia keeps whining about the game/benchmark which gives AMD free marketing.
 
Good thing that the only benchmark out also just so happens to Be built in a game... It's not hype. The results are there for you. The game and benchmark use asynchronous compute. It's comparable. The fact the game releases this year also makes it even more valid as it's not a tech demo. AMD have hardly really made any noise, Nvidia keeps whining about the game/benchmark which gives AMD free marketing.

If it was Nvidia that was ahead in performance in AOTS no one would be moaning at all guaranteed but the moment AMD gets a leg up "HELL NO WTF NO WAY !!!!".
 
If it was Nvidia that was ahead in performance in AOTS no one would be moaning at all guaranteed but the moment AMD gets a leg up "HELL NO WTF NO WAY !!!!".

You are slowly taking your fanboyism way too far, even if Nvidia was in amds place right now more than enough people would be as critical as they are now, there is just something in us that doesnt want to believe these benchmarks. (Atleast in me that is)
 
You are slowly taking your fanboyism way too far, even if Nvidia was in amds place right now more than enough people would be as critical as they are now, there is just something in us that doesnt want to believe these benchmarks. (Atleast in me that is)

No I am not an AMD fanboy that's quite a silly thing to say dude without knowing all the facts.

I have 4 Nvidia GPU's in my household, 780 *Non Ti*, Titan Black, 2 x 980's *Non Ti* and 1 x AMD card and just because I have an AMD sig doesn't make me a fanboy if anything I'm a little annoyed at both companies due to the lackluster improvements especially AMD due to the Fury + X having zero amounts of voltage control so overclocking is extremely limited.

I've been watching quite a few different forums that are discussing the AOTS benchmark/game and nigh on all of the Nvidia only users are complaining that AMD has a bit of a lead in DX12 in that 1 and only game with some users on those forums going a little too far.
 
Last edited:
You are slowly taking your fanboyism way too far, even if Nvidia was in amds place right now more than enough people would be as critical as they are now, there is just something in us that doesnt want to believe these benchmarks. (Atleast in me that is)

Don't want to be believe it? Why wouldn't you? It's literally just facts and results that are being consistent throughout everyone's testing. This isn't fanboyism here dude. We are going off what's being reported, I actually OWN the game unlike most people. I have access to the Alpha threads on their website and can see other people's results and even talk to Devs if I wanted. Even everyone there is reporting similar results. Nvidia DO NOT support Asynchronous Compute on a hardware level and are trying to replicate it through Drivers. It's no where near as efficient(at the moment but even certain people believe it simply is limited through software). Sure they can make it great, but making it would also thereby increase the driver overhead slightly, so again it's a balance for them to achieve. Nvidia has a good Driver team, hopefully they can find other ways to do better but as of now if a DX12 software uses Asynchronous Compute, AMD immediately has an edge, And the way Async works, will probably be most titles since it is very efficient at multitask scheduling which also helps the CPU in return.
 
IF AMD keeps up these sorts of numbers in newer games, and nVidia fall behind I suspect it will only be a matter of time before nVidia pull ahead again.

For nVidia it will depend on how long it takes them to pull ahead again as to whether AMD regain some of the market share that they have lost, and personally I hope they do regain quite abit of it but personally I cannot see it happening to any sort of meaningful level.

When I am due to upgrade my graphics cards which I would have normally by now already done, I will be looking at AMD and nVidia to see who has not only the best performance in the games that I play but also who's cards are cooler and use less power.

I have not used an AMD card in my main pc since the 8800 series cards came out from nVidia, same with AMD cpu's I have not used one in my main pc since the Q6600 came out.

I would love to use AMD again but I just want the performance that Intel and nVidia can give, but if I find at upgrade time that AMD are performing better then I will switch.

As for the current state of benchmarks and games using DX12 I will wait till more benchmarks and games, before I make any choice because currently I don't feel there is enough to make a clear choice that will benefit me in the long run.
 
Most AMD cards currently outperform Nvidia cards dude.. It's just most people still think Nvidia have better drivers(as of late they seem to have dropped the ball slightly) and are always faster. That's not the case. While I can understand the Power consumption arugment, in a desktop format, I can't see it being any sort of meaningful argument over who's better tbh. Most people run a 650watt+ PSU. More than enough for any single card you can get. Coolers on both sides work great tbh. Nearly every AMD card from Hawaii chips to Fiji all perform amazing with their coolers despite their heat output. Nvidia also run very cool as well. So taking that out of the equation, AMD generally have better performing cards at every price segment. Obviously excluding the TX however.

Though like you, I really do hope AMD can keep this up and maintain an advantage. They need more market share. People just need to actually buy their products instead of constantly saying we need AMD in the market, then go out and buy a Nvidia card right after.
 
Driver's aren't the problem here for AMD. AMD on a hardware level supports Async Shaders. Nvidia do not. So the way Nvidia gets it's supoprt is through drivers, however it's not nearly as efficient and has it's limits. AMD doesn't have this issue. Your limited by the hardware and the devs have full access to it. So for AMD it's down to the devs more than themselves. Nvidia are just screwed. They better have it for Pascal otherwise they may just start to fall behind more. By now if Pascal on the design level for it's die doesn't have this capability, then they will probably end up losing a lot of marketing power since they won't compete well with AMD and AMDs next gen products. As of now Nvidia only hold the TX as a clear winner. That could easily change with limited drivers.

I've never said Drivers are the problem, all I've said is that Nvidia's financial resources are substancially bigger than AMD's meaning that AMD's jump on HBM might not potentially last long so they need to market this well and get people to jump ship to new tech.

IMO its Naive to say there's a winner or a loser...I see it as 2 companies both currently offering something thats the same but work different slightly.
 
nVIDIA do support Asynchronous Compute on a hardware level but a large part of their scheduler is implemented in software. Therefore what feeds their Asynchronous Warp Schedulers is a series of Software solutions.

nVIDIAs solution is also limited to 32 Compute Queues (or 1 Graphic + 31 Compute). Therefore under any sort of high Asynchronous Compute workloads, they run the risk of hitting a bottleneck. This is further compounded by the Latency introduced under Preemption due to their slow context switching.

Face it... nVIDIA was not as prepared as AMD for DX12. That's to be expected given that AMD has had a lot of experience with Mantle as well as the console API's.

We may need to wait until Pascal in order to see the level of engineering talent over at nVIDIA as it pertains to DX12.
 
I've never said Drivers are the problem, all I've said is that Nvidia's financial resources are substancially bigger than AMD's meaning that AMD's jump on HBM might not potentially last long so they need to market this well and get people to jump ship to new tech.

IMO its Naive to say there's a winner or a loser...I see it as 2 companies both currently offering something thats the same but work different slightly.

Naive? Not at all. They don't offer the same thing. One is hardware, one is software. The hardware solution is much better currently. Drivers for Nvidia are the problem. They need to get them to work very efficiently to avoid bottlenecks. That's all I was saying.
HBM has nothing to do with it tbh. They aren't bandwidth limited. They are CPU limited and then GPU limited.

Link to my AOTS benching thread if anyone wants to see how the cards actually perform on the bench.:)

http://forums.overclockers.co.uk/showthread.php?t=18690671

Be careful posting anything outside of the NDA.. not lifted yet;)

nVIDIA do support Asynchronous Compute on a hardware level but a large part of their scheduler is implemented in software. Therefore what feeds their Asynchronous Warp Schedulers is a series of Software solutions.

nVIDIAs solution is also limited to 32 Compute Queues (or 1 Graphic + 31 Compute). Therefore under any sort of high Asynchronous Compute workloads, they run the risk of hitting a bottleneck. This is further compounded by the Latency introduced under Preemption due to their slow context switching.

Face it... nVIDIA was not as prepared as AMD for DX12. That's to be expected given that AMD has had a lot of experience with Mantle as well as the console API's.

We may need to wait until Pascal in order to see the level of engineering talent over at nVIDIA as it pertains to DX12.

Nvidia do not support it on a 100% hardware level. Sure their current scheduler's can process the Asyn Compute given to them, however need software's help to actually carry out that process. AMD just needs to activate the ACE's in the driver and doesn't go to the extent of aiding the ACE's. Which is why they get hit much harder at higher draw calls as of now, or in this game at the very least. I don't disagree on the performance differences, I'm just disagreeing with the support on a hardware level issue. As it's not technically correct to an extent.
They support 32 queues, however AMD supports 128 natively. Big difference, which again is why we see differences in heavy draw call scenario's.

Although i don't think Console's had much to do with it tbh. Mantle is a maybe. AMD has had all this support since 2011 and that means have planned for it years before that since it takes about 18-22months for a chip to enter mass production.

Pascal is more than likely in prototype stage. Going back and changing or adding anything similar to the ACE's in GCN would mean they would have to delay the launch. Now if they already had that design in there already, then it just took them little longer to realize the advantages of strong scheduler's capable of high queues.
 
Last edited:
Good thing that the only benchmark out also just so happens to Be built in a game... It's not hype. The results are there for you. The game and benchmark use asynchronous compute. It's comparable. The fact the game releases this year also makes it even more valid as it's not a tech demo. AMD have hardly really made any noise, Nvidia keeps whining about the game/benchmark which gives AMD free marketing.

I will apply the same level of skepticism regardless of who it is, even if I feel more 'loyalty' to one company over the other. I mean FFS I'm currently waiting on the furyX2 to come out so I should be happy about this turn of events. So excuse me while I hold judgement until I see a load more games with async, not just this one example.

While nVidia are now supposedly being shady as f*ck when it comes to Oxide now they've got the talent to make async computing work for them. Heck it actually does work, albeit in a buggy manner from what I've read thus far. nVidia are not amd, they'll either make it work or pay people off so that it works in the way that favors nVidia cards.

Early days innit. My next card is almost certainly going to be AMD for a number of reasons but this is not a factor for me yet.
 
I will apply the same level of skepticism regardless of who it is, even if I feel more 'loyalty' to one company over the other. I mean FFS I'm currently waiting on the furyX2 to come out so I should be happy about this turn of events. So excuse me while I hold judgement until I see a load more games with async, not just this one example.

While nVidia are now supposedly being shady as f*ck when it comes to Oxide now they've got the talent to make async computing work for them. Heck it actually does work, albeit in a buggy manner from what I've read thus far. nVidia are not amd, they'll either make it work or pay people off so that it works in the way that favors nVidia cards.

Early days innit. My next card is almost certainly going to be AMD for a number of reasons but this is not a factor for me yet.

The Elephant in the room is despite all NVidia's whining it is their cards that are the fastest on the bench in DX12.
 
Back
Top