** The Official Nvidia GeForce 'Pascal' Thread - for general gossip and discussions **

[humbug]

We would have to see it live in action through diagnostics tools to see what is really going on.

 

[D.P.]

So in other words the latest version with apparent working Async is doing what i stated. the driver just receives the async commands then runs the compute consecutively with the graphics as though the application had no async. It is certainly one way around supporting applications that enable async without the application needing another rendering path or disabling the effects that use async.

I reckon Nvidia did it this way in the end due to the context switching adding too much latency. That and with a few big games coming out that will be async heavy they want everything to just work.

Did anyone try the ones i posted? it gives you fps results.

I expect what Nvidia driver does is run the task serially by default if it is an application that is unrecognized. If the application is recognized and nvidia knows the async task is well coded for the architecture then it lets the Async code run as expected. Moreover, if it is a recognized application but not well designed for the Maxwell architecture then NVidia drivers will have a hand coded optimization that improves performance with ASYNC. This is why the Oxide developers could calim that the nvidia driver is now fulyl supporting ASYNC, because Nvidia's driver supports their application.

That might sound weak, but that is done all the time for normal drivers anyway. A game might heavily use a fragment shader that is just not well coded for a particular architecture or Nvidia knows of an optimization. They may tell the developer, but more often than not they will simply write a replacement shader and the driver will swap out the relevant shader.



The only reliable benchmark we have i Legends of Fable where the Nvidia Async part was at least as fast as on AMD, the reason due to Nvidia working closely to ensure the async code did not make too many context switches.

 

[Gregster]

We would have to see it live in action through diagnostics tools to see what is really going on.

I am surprised sites like PCPer and TheTechreport (I think it was those who did the FCAT thing first?) have not got something in place to test but I am sure something from somewhere will come about in the coming months.

 

[humbug]

I am surprised sites like PCPer and TheTechreport (I think it was those who did the FCAT thing first?) have not got something in place to test but I am sure something from somewhere will come about in the coming months.

I don't know but 'i think' a lot of the necessary tools don't exist yet, still too new a concept, people are still trying to get to grips with programming for it, from that come the diagnostic tools.

 

[Mauller]

You can already see if Async is working by using GPUVIEW, it comes with the Windows 10 SDK. just lookup how to use it, there is documents on the web.

essentially it will show commands being used in more than one queue in hardware.

 

[D.P.]

Isn't the increasing number on NVidia due to the queue size, seeing as Maxwell can do a queue depth of 32 and Fiji can do 128 (I think), if you asked the program to run 256 or 384 the AMD would step up as well.

Don't worry though NVidia will be paying developers to remove async from games, well according to DM anyway.

Yes, the compute part on the Nvidia results clearly shows the hardware working as expected, handling 32 concurrent threads, AMD"s can handle 128, but the 32 threads on Maxwel are more powerful so when doing a certain mount of computation the differences are much smaller and architecture specific which is faster-trivial to make a compute task where either nvidia or AMD faster.

The real question is how well compute + graphics works on maxwell, and the only real answer is "that depends". If a developer is not careful then they can make async code run very slowly on Nvidia hardware, if they are more careful then it will be very fast.




And at the end of the day the advantage of ASync compute depends entirely on how much load the GPU is under form graphics alone. Async is basically like Intel hyper-threading, there are weak compute threads that try to make use of spare resources. If you you have spare resources because the main graphics thread is fully leveraging the hardware then async will add very little. If the async compute threads are all doing the same tasks then they will keep running into each other and very quickly saturate.

 

[humbug]

RE: ^^^^

Async is basically like Intel hyper-threading

Thats the description i was looking for

You can already see if Async is working by using GPUVIEW, it comes with the Windows 10 SDK. just lookup how to use it, there is documents on the web.

essentially it will show commands being used in more than one queue in hardware.

No , i'm talking about diagnostic tools, detailed information about what exactly is happening where and when, not just that its doing something, often on the surface it can appear to be doing something but when you dig right down into the metal you see that its actually not.

 

[Mauller]

Ah i have sorted it now, the guy replied to me. To get these working you need visual studio 2015 installed (community should be fine) and windows 10 SDK. you then need to open the executable in visual studio as a project and it should run.

http://www.binaryvisionaries.com/dev/async.rar

Would have been nice if it was just a standalone build.

Edit - I just found the missing DLL. downloading and sticking it in the folder with the exe will let it run. http://download.dll-files.com/c3130cfb00549a5a92da60e7f79f5fc9/ucrtbased.zip?0XHhVAXMfT

 

[Mauller]

No , i'm talking about diagnostic tools, detailed information about what exactly is happening where and when, not just that its doing something, often on the surface it can appear to be doing something but when you dig right down into the metal you see that its actually not.

GPUVIEW already shows you what is happening in the hardware. how much lower do you want to get?

From what i gather you area talking about a full debug, but i doubt that would be allowed since it would need to be enabled in the code.

 

[layte]

Well I'm as confused as anyone right now. Here's a GPUView capture from running the Async test on my Titan X.

 

[drunkenmaster]

Well if it can support Async, I am fine with that. I am still waiting for you to show me where I said that the Titan was a compute card.

YOUR BASKET
1 x PNY Nvidia Tesla K20 Workstation Solution Graphics Card - 5GB - GDDR5 SDRAM £2999.99
1 x Gigabyte GeForce GTX Titan 6144MB GDDR5 PCI-Express Graphics Card £839.99
Total : £3,851.38 (includes shipping : £9.50).



I think it is cheap.

https://forums.overclockers.co.uk/showpost.php?p=23680665&postcount=108

They are binned K20 chips. This is why I feel they will be a limited quantity card. Maybe I am wrong and at the suggested price of £700, if they sell well, they could make more.

The 12.11's deffo put the 7970 ahead.

Okkkkkkk

Look at it from this angle. The compute card costs over £2500 (K20X) and Titan can do the job at a fraction of the cost. It isn't quite as quick as the K20X but very close and Nvidia spotted an opportunity to make the K20X into a gamers card. Nobody is forced to buy it and there are plenty of GPU's out that will do what most people need.

More justifying the price of Titan based on how it offers basically the same compute performance. It's funny how often you compare it and the price to a compute card, with multiple posts where you agree with people about it being a compute card... of course you had like eleventy billion posts deleted in all these threads for definitely not the reason of making it personal and insulting people all the time or anything, for completely random different reasons to that.

Compute performance doesn't interest me but these Titans should be perfect for GPGPU Cuda work.

Why would they be perfect for Cude/GPGPU work if they aren't compute cards, what makes them so perfect for that kind of work if not compute?

Originally Posted by humbug View Post
That, ^^^ some things AMD introduced into games like Sleeping Dogs, Dirt Showdown.... don't work well on Nvidia GPU's

The only reason for that is those features use GPU compute, which with Nvidia, the GTX 600 series especially, is not much good.

Agreed but I expect great performance from Titan

Hmm, why did you agree, ah right, because it's a compute card.

Going by the shortages on most of the main sites, I would say Titans sold very well. You have to remember also that Titan is a K20X card (not quite but almost) and it would save a lot of money for compute users etc.

So it seems to me you can't stop comparing it or bringing up that it is a compute card and that it seems cheap in comparison to the exact same card sold under different branding.

 

[Rroff]

Well I'm as confused as anyone right now. Here's a GPUView capture from running the Async test on my Titan X.

Its really best not to take some of the claims about nVidia and async at face value

 

[Gregster]

@DM - You even post where I said Compute doesn't interest me and I was quite correct in a cheaper K20 replicant card.

And I justify the price because I bought it (and 2 more). If that makes me wrong for self justification, then hang me.

 

[Mauller]

Just got the DX12 test working, you need to grab the DLL and then stick it in the folder with the exe for each test. then just check your scores between tests after letting them run for a few minutes.

Async DX12 examples with and without async http://www.binaryvisionaries.com/dev/async.rar

Missing DLL file http://download.dll-files.com/c3130cfb00549a5a92da60e7f79f5fc9/ucrtbased.zip?0XHhVAXMfT

I went from 34 ish to 40 ish fps with async on. using my 7870 XD

 

[Ayahuasca]

Well this thread is just utterly depressing isn't it, people going back months to find posts to 1 up each other, completely off topic. Admittedly Greg initiated it.

Agreed Rroff, don't know why people are trying to jump to conclusions on async already in a thread for an architecture that isn't released.

 

[Gregster]

Well this thread is just utterly depressing isn't it, people going back months to find posts to 1 up each other, completely off topic. Admittedly Greg initiated it.

Agreed Rroff, don't know why people are trying to jump to conclusions on async already in a thread for an architecture that isn't released.

Quite true and I will end it here and apologies for dragging up the past. I did say it with tongue in cheek but my bad and shouldn't have.

 

[FrenchTart]

It would be nice if people could calm down and stop bickering.

 

[drunkenmaster]

@DM - You even post where I said Compute doesn't interest me and I was quite correct in a cheaper K20 replicant card.

And I justify the price because I bought it (and 2 more). If that makes me wrong for self justification, then hang me.

What precisely is the point there? You didn't say show where you said DP was important to you and I never posted claiming you said that. You arbitrarily demanded for no reason that I show where you called(or implied over and over and over) that it was a compute card, I did. Now about where I had previously asked you to point out where I had insulted or attacked you?

Or where you quoted two of my posts that didn't in any way make a fuss about DP or lack of DP but only yours and others change in attitude over the pricing of Titan being cheap precisely because it was a compute card which incidentally all those quotes prove you stated... anything to say about that?

It's still the fact that you and was it lambchop decided that others were making a fuss over DP because we didn't know what it was. I posted directly to say I and other people never cared about DP, it was only that for Titan your excuse about it's price(again look at the last set of quotes) was that it was cheap FOR A COMPUTE card, and that when Titan X was expensive your attitude changed only to DP isn't important in relation to it's pricing.

THat you didn't need or care for DP performance personally for gaming is neither here nor there. YOu justified the pricing of Titan using it's DP performance, that is all I've said and you randomly confirmed by posting the two old posts of mine making that exact argument.

EDIT:- also when proven wrong you suddenly go on the defensive and it was tongue in cheek..... really

Always personal with you

Do me one biggie and show me where I said it was a compute card. If not, I will be reporting your post, as if you can't see any of that as insulting, I will let a mod decide.

Which part of those bits is tongue in cheek? ALso for the record that second post comes off as a threat as much as anything else... after consistently attacking my posts in recent months. Any time I post you instantly jump to accusing me of attacking others and insulting, yet when asked which parts are insulting... you deflect and randomly demand stupid quotes of things you said that are entirely unrelated as usual. When someone actually follows through(after someone comes to your defence) you pretend it was all a joke.

 

[Rroff]

It would be nice if people could calm down and stop bickering.

I wonder if whatever admin clicked the create button on this section had any premonition of just what they were unleashing.

 

[drunkenmaster]

It would be nice if people could calm down and stop bickering.

It would be nice if when Greg consistently attacks other posters, makes claims about insults and personal attacks while himself having a go people in every single thread, if the mods did something about it.