[Updated] Oxide responds to AotS Conspiracies, Maxwell Has No Native Support For DX12 Asynchronous Compute

[Mahigan]

I hope I'm not too late with this:https://m.reddit.com/r/nvidia/comments/3j5e9b/analysis_async_compute_is_it_true_nvidia_cant_do/

While it is true that Maxwell 2 can, on paper, handle Asynchronous Compute... nVIDIAs current driver cannot. That post is based on misinformation. The folks who made it were misinterpreting the Data from Beyond3D. Beyond3D wasn't testing for performance, they were testing for latency. They wanted to see if the Graphic and Compute latency added together was less than the Graphic + Compute latency in order to determine if Asynchronous Compute was functioning. It wasn't. It worked for GCN but not for Maxwell 2.

 

Yesterday, the Oxide developer showed up on our forums at overclock.net and informed us that nVIDIA is aware of the issue, that it's a driver issue, and they're working to resolve it in their driver.

 

Now we wait and see for the result of these driver fixes.

[Mahigan]

@Mahigan, and yet the new Maxwell Teslas are selling like hotcakes. The only problem with Maxwell in compute workloads is its lack of native 64-bit support. Otherwise it's just as good and sometimes better than Kepler.

 

And they're bringing it back with Pascal. There are other reasons why Tesla cards sell well... they're selling to an industry which is familiar with coding for CUDA and optimizing for CUDA. There have been problems, however, such as the low DP performance as well as the lack for a hardware scheduler. Everything, from Tesla to the Gaming GPUs are moving to Pascal in Q2 2016. Pascal will also introduce half precision (FP16) in order to handle tasks which don't require a high level of precision.

 

I'm waiting for Pascal and Greenland myself. Not bothering with Fiji or Maxwell 2.

 

 

 

patrickjp93, on 06 Sept 2015 - 05:00 AM, said:

I reference when there are actual things to reference. The equation for flops is known across the damn industry. I shouldn't have to hold the kiddies' hands on everything!

 

You're right, that's how you calculate the theoreticals, but the problem is that a GTX 780 Ti doesn't hit its theoreticals in practice. Under DX11, neither did GCN. With DX12, GCN hits it's theoreticals with a high degree of efficiency because it can even execute tasks in between executions... that's Asynchronous Compute. Ensuring the CUs are never left idling. It can do so with a high degree of efficiency without relying on slow context switching. 

 

Maxwell 2 relies on slow context switching when performing pre-emption... therefore it remains to be seen what sort of performance bump Maxwell 2 will get from Asynchronous Compute.

[Agent181]

so im going to post this here for everyone.....

 

There is no reason to keep fighting over this, ok on the tests amd has now gained performance and nvidia does not gain that much... this is just one engine that will be used for games... remember that there will always be different results depending on the game and the engine it runs on, so amd gets a little boost, good for them since they are struggling to survive, as for nvidia they have a lot more resources and funding to resolve this and improve on the problems....

 

so everyone chill and relax... this is just the beginning and there are more to come so there is no reason to fight over something that isn't even done yet.

 

now to EVERYONE ON THE FORUM calm down and don't fall to people that just join to defend a side....

 

im looking at @Mahigan that only joined 2 days ago to respond to this thread and defend a side, if your a regular on the forum watch out for this and try not to fight over something that has no point.  

 

-Agent

[Kinda Bottlenecked]

so im going to post this here for everyone.....

 

There is no reason to keep fighting over this, ok on the tests amd has now gained performance and nvidia does not gain that much... this is just one engine that will be used for games... remember that there will always be different results depending on the game and the engine it runs on, so amd gets a little boost, good for them since they are struggling to survive, as for nvidia they have a lot more resources and funding to resolve this and improve on the problems....

 

so everyone chill and relax... this is just the beginning and there are more to come so there is no reason to fight over something that isn't even done yet.

 

now to EVERYONE ON THE FORUM calm down and don't fall to people that just join to defend a side....

 

im looking at @Mahigan that only joined 2 days ago to respond to this thread and defend a side, if your a regular on the forum watch out for this and try not to fight over something that has no point.  

 

-Agent

 

I'm sorry but Mahigan isn't actually taking a side. @Mahigan is the Mahigan from here http://www.overclock.net/t/1569897/various-ashes-of-the-singularity-dx12-benchmarks/400#post_24321843

[Agent181]

I'm sorry but Mahigan isn't actually taking a side. @Mahigan is the Mahigan from here http://www.overclock.net/t/1569897/various-ashes-of-the-singularity-dx12-benchmarks/400#post_24321843

Ok that's good, just want everyone to make sure not to fight over something that would have no point, in the past there have been some workers come to forums to defend their products...

 

seen it a lot back when I use to be on hard forum and the start of this one

[Mahigan]

so im going to post this here for everyone.....

 

There is no reason to keep fighting over this, ok on the tests amd has now gained performance and nvidia does not gain that much... this is just one engine that will be used for games... remember that there will always be different results depending on the game and the engine it runs on, so amd gets a little boost, good for them since they are struggling to survive, as for nvidia they have a lot more resources and funding to resolve this and improve on the problems....

 

so everyone chill and relax... this is just the beginning and there are more to come so there is no reason to fight over something that isn't even done yet.

 

now to EVERYONE ON THE FORUM calm down and don't fall to people that just join to defend a side....

 

im looking at @Mahigan that only joined 2 days ago to respond to this thread and defend a side, if your a regular on the forum watch out for this and try not to fight over something that has no point.  

 

-Agent

 

I joined, at first, because of the Tech report thing, I was also angry that nobody, tech journalists I'm looking at you, was working to correct the misinformation being spread across the web about the Async issue. 

People are spreading fear and misinformation across the web about the Asynchronous Issue. Some people claim that Maxwell 2 doesn't support it, because the Oxide developer stated this in one of his responses. Since information appears to be slow and people are running around with "old news", I've been joining forums to keep people in the loop.

 

Everyone needs to calm down, I agree with this statement 100%. Don't sell your graphic cards... don't overreact. nVIDIA is working to implement a fix in their driver. At the same time I'm doing the work for the tech journalists. I'm explaining the issue.

 

As for my post on GCN, AMD has been waiting for DX12 for this reason. It's why they pushed for Mantle. AMD users, however, are going overboard. nVIDIA users are reacting with fear. The whole thing has become idiotic if you ask me.

As for what I can do. I can explain GPU architectures and the ups and downs of them all, basically what a tech journalist is supposed to do rather than run benchmarks and conclude a winner. If people had been told what Asynchronous Compute was all about, from the beginning, none of this would have happened.

 

If there's a "side" I'm defending... it's the truth. I don't fly a green or red banner.

[Humbug]

Maxwell 2 relies on slow context switching when performing pre-emption... therefore it remains to be seen what sort of performance bump Maxwell 2 will get from Asynchronous Compute.

Does that mean that it should perform well as long as you have a strong CPU?

While lagging behind AMD if paired with a less strong CPU?

[Agent181]

snip

I'm glad you did this and for trying to clear up some of the confusion, I was just trying to stop some of the fighting that has happened over this but I think you settled it well here since you know a lot more than many do.  I'm sorry if I came out as a A-hole but I just hate seeing the fanboy wars.  

[ybriK]

It's like people hate it when AMD is winning on something. Looks like karma does bite your back. Also this is only 1 Dx12 title, why is everyone on the green side going mental. It's not like all Dx12 titles will all require async.

[Tomsen]

It's like people hate it when AMD is winning on something. Looks like karma does bite your back. Also this is only 1 Dx12 title, why is everyone on the green side going mental. It's not like all Dx12 titles will all require async.

AMD can't win if i keep spinning my argument, change perspective, fiddle with minor details to fit my world view.

[Zah]

Everyone needs to calm down, I agree with this statement 100%. Don't sell your graphic cards... don't overreact. nVIDIA is working to implement a fix in their driver. At the same time I'm doing the work for the tech journalists. I'm explaining the issue.

Or if you do sell your GPU, make sure you upgrade to the next hardware when it's out, not a sidegrade to another 28nm process.

 

(That's what I'm doing with my 980ti, I run through these cards like gas)

[Humbug]

It's like people hate it when AMD is winning on something. Looks like karma does bite your back. Also this is only 1 Dx12 title, why is everyone on the green side going mental. It's not like all Dx12 titles will all require async.

Even this game does not 'require async'. It's something that can be implemented to better utilize the hardware potential of GCN, future Pascal and other GPUs that support it. It doesn't hurt Maxwell. It just helps other architectures especially upcoming ones. So I see no reason why devs should not use it barring political reasons or other conspiracy theories. Worst case scenario they can turn it off for Maxwell.

[Prysin]

 I swear if social Darwinism was legal more than two thirds of this forum wouldn't be considered eligible for breeding.

That includes you too.... for reasons you wouldnt admit.

 

 

Where not trolling, we are just asking you to reference the information you provide correctly. Oh and to provide a source of information that isn't just yourself.

This is the whole point.

[patrickjp93]

While I think you're in the right here, I don't think that's sufficient justification for insulting other members, regardless of how frustrated you are. If you have a problem with them, you can take it up with the mods, not fly insults in their faces. Rules exist for a reason.

Rules are guidelines. There is always the rare exception which proves the rule, and sometimes obsequious little gnats need to be squashed.

[patrickjp93]

And they're bringing it back with Pascal. There are other reasons why Tesla cards sell well... they're selling to an industry which is familiar with coding for CUDA and optimizing for CUDA. There have been problems, however, such as the low DP performance as well as the lack for a hardware scheduler. Everything, from Tesla to the Gaming GPUs are moving to Pascal in Q2 2016. Pascal will also introduce half precision (FP16) in order to handle tasks which don't require a high level of precision.

I'm waiting for Pascal and Greenland myself. Not bothering with Fiji or Maxwell 2.

You're right, that's how you calculate the theoreticals, but the problem is that a GTX 780 Ti doesn't hit its theoreticals in practice. Under DX11, neither did GCN. With DX12, GCN hits it's theoreticals with a high degree of efficiency because it can even execute tasks in between executions... that's Asynchronous Compute. Ensuring the CUs are never left idling. It can do so with a high degree of efficiency without relying on slow context switching.

Maxwell 2 relies on slow context switching when performing pre-emption... therefore it remains to be seen what sort of performance bump Maxwell 2 will get from Asynchronous Compute.

No GPU hits its theoreticals in practice. Nvidia still gets closer than AMD does, which is why the Teslas have always done better than the FirePros for HPC. CUDA has nothing to do with it. The best programmers in the world can convert CUDA to OpenCL, OpenACC, and OpenMP in short order. Nvidia just has the RWP whereas AMD completely lacks it.

[Dabombinable]

Rules are guidelines. There is always the rare exception which proves the rule, and sometimes obsequious little gnats need to be squashed.

Tsk tsk, some one doesn't seem to like being asked to prove something with a source other than himself.

[patrickjp93]

Does that mean that it should perform well as long as you have a strong CPU?

While lagging behind AMD if paired with a less strong CPU?

Yes and no. The bottleneck could also become the PCIe bus if you're trying to run 32 asynchronous loads instead of a unified single one as we do right now.

[patrickjp93]

Tsk tsk, some one doesn't seem to like being asked to prove something with a source other than himself.

I don't mind when it's a difficult/technical question for which the answer isn't available with 4 seconds of Google searching. @LukaP will tell you the same. If you're opening your mouth about technical features related to GPUs and don't even know the basics, you shouldn't be posting. You should be watching and learning. The technically literate in this field all know that equation by heart. The fact you don't and yet have the nerve to claim my mathematics aren't proof (equations cannot lie. Only the inputs to it can) is the most obscene thing here, not my balking at your shortcoming and frustration over it. I have a sterling reputation here and take it personally as an insult when arrogant little upstarts barely out of their diapers think they are entitled to hand holding while simultaneously trying to undermine the actual experts around them.

I provide cited evidence when the details aren't sitting in front of your face, such as IEEE papers (though you need a subscription or university connection to access those anyway). When the information is easily induced from the basics of the goddamn field guide for the world of PC gaming and you demand proof, it's only evidence you didn't finish your homework or blew through it without getting it. In short, stop wasting my time and learn before you think to challenge me.

[patrickjp93]

That includes you too.... for reasons you wouldnt admit.

This is the whole point.

I might not be a looker, but I'm a Hell of a lot more valuable to he gene pool than most around here.

[HKZeroFive]

Rant

Mate, this is an Internet forum. No need to show off your skill of the English language.

Just saying.

[Prysin]

I might not be a looker, but I'm a Hell of a lot more valuable to he gene pool than most around here.

not as much as you may enjoy imagining....but that discussion is for another thread.

[Tomsen]

This thread is becoming more and more a cringe-fest to be honest.
Narcissism is a bad thing, mkay..

[Dabombinable]

This thread is becoming more and more a cringe-fest to be honest.

Narcissism is a bad thing, mkay..

Ditto

[patrickjp93]

Mate, this is an Internet forum. No need to show off your skill of the English language.

Just saying.

Believe it or not that's how I speak in real life.

[Tomsen]

Believe it or not that's how I speak in real life.

I feel sorry for you..