AnandTech AMD Dives Deep On Asynchronous Shading

[Gloomy]

What's with the "64+" number, surely they know how many queues their hardware exposes??

8 ACEs, 8 queues per ACEs.... that's 64. But where is the "+" coming from? Does the GCP also schedule compute? How many queues?

 

[Rvenger]

Yet AMD gets trounced in DX12 performance, so much for having 8 ACEs, 64 queues.

nvgpu, stay out of this thread. Your posts are just here to bait the others and will not be tolerated. If I see you again in this thread, it will be an immediate vacation for baiting others.

-Rvenger

 

[nvgpu]

I will post in any thread I want with facts to back me up and even if you permanently ban me it doesn't affect me at all.

Keep on abusing your mod powers just to feed your oversized ego though. Same with every other mod here with oversized egos giving out infractions and bans, they just can't accept the truth backed with facts.






You were told to stay out of that thread and given a reason why.




esquared
Anandtech Forum Director

 

[Hitman928]

I will post in any thread I want with facts to back me up and even if you permanently ban me it doesn't affect me at all.

Keep on abusing your mod powers just to feed your oversized ego though. Same with every other mod here with oversized egos giving out infractions and bans, they just can't accept the truth backed with facts.

You have a very loose interpretation of the word facts. Anyways, hope you enjoy your vacation though I wonder how many names we might know you by before all is said and done.

 

[stahlhart]

You have a very loose interpretation of the word facts. Anyways, hope you enjoy your vacation though I wonder how many names we might know you by before all is said and done.

And you stay out of it, unless you want to join him.
-- stahlhart

 

[Abwx]

What's with the "64+" number, surely they know how many queues their hardware exposes??

8 ACEs, 8 queues per ACEs.... that's 64. But where is the "+" coming from? Does the GCP also schedule compute? How many queues?

...............

 

[kagui]

so the 64+ is context switch, i was under the impression that only carrizo have it

 

[DarkKnightDude]

https://www.youtube.com/watch?v=v3dUhep0rBs

AMD made a more simplified video explanation about Asynchronous Shaders on their channel.

 

[geoxile]

https://www.youtube.com/watch?v=v3dUhep0rBs

AMD made a more simplified video explanation about Asynchronous Shaders on their channel.

AMD needs to fix their marketing. They seem to have left out the part where multiple queues can execute concurrently by sharing GPU resources.

 

[VirtualLarry]

AMD needs to fix their marketing. They seem to have left out the part where multiple queues can execute concurrently by sharing GPU resources.

I can't help but imagine if x86 CPUs couldn't also be designed to work this way, rather than the individual, but seperate cores.

 

[Gloomy]

I can't help but imagine if x86 CPUs couldn't also be designed to work this way, rather than the individual, but seperate cores.

...Hyperthreading?

 

[Dug]

Easily. People talk about Nvidia's perf/watt advantage but at what cost? There is no free lunch and Nvidia has clearly skimped on features to bring down the transistor count.

5.6-8 billion is skimping for a single core?
ejus I got to get out more often.

 

[kagui]

5.6-8 billion is skimping for a single core?
ejus I got to get out more often.

is skimping if you compared to the previus arch, sacrificing DP to hit better FPS, to most that doesnt matter but for some of us does matter,i like to use the full chip, compute + graphics, anyway that is offtopic

 

[Silverforce11]

Still haven't updated their chart. Its presented as if GCN 1.1/1.2 can only do 1 + 8 when its actually 1 + 64 queues.

If its presented as actual queue engines, and each ACE is counted as 1, then the author needs to clearly state the queues that each engine can handle simultaneously.

As it is, it implies that Maxwell 2 is far superior in DX12 graphics & compute than GCN 1.1/1.2 which isn't true at all, but the reverse.

 

[Abwx]

Still haven't updated their chart. Its presented as if GCN 1.1/1.2 can only do 1 + 8 when its actually 1 + 64 queues.

If its presented as actual queue engines, and each ACE is counted as 1, then the author needs to clearly state the queues that each engine can handle simultaneously.

As it is, it implies that Maxwell 2 is far superior in DX12 graphics & compute than GCN 1.1/1.2 which isn't true at all, but the reverse.

Yet he s claiming in the article that it gives an advantage to AMD, so it s not like he s ignorant of the thing, and several members pointed him the mistake at the page bottom.

Anyway not professional at all to say the least.