Isn't the entire point to be like HyperThreading and interleave instructions on a clock-by-clock (ns) basis and not some 0.1ms crap?
pascal is just maxwell on speed
pascal wasn't suppose to exist, nvidia was suppose to go from maxwell to volta
therefore, pascal async support is still ghetto rigged
>>54582062
Simultaneous Multi-threading (called hyperthreading by intel) is meant to allow for the simultaneous execution of 2 instructions from 2 threads on the same core during the same clock cycles.
This is simply driver controlled temporal multi-threading with some hacks thrown in for good measure.
>>54582062
Nobody knows till 1080 NDA ends tomorrow and benches come out.
>>54582062
There is still no async shaders. The compute engine is still worse than gcn 1.0.
Looks pretty ghetto to me.
>still falling for tbe async meme
any game developer that thinks resources should be split between a graphics and compute task instead of dedicating 100% of resources to one task before moving onto the next is a fucking moron and doesn't understand how to write good parallelized software and gaymes.
>>54583729
both of those slides are from AMD presentations. 'switching overhead' is a problem on both GCN and kepler/maxwell/pascal.
>>54583843
You are retarded.
>>54583893
>decades of design methodology should be thrown out the window to appease indian engineers who shit out rebrands and have less than 18% market share
nice try raja
>>54583948
Doesn't change your retardation.
>>54583969
>Doesn't change your retardation.
present an argument that proves my supposed retardation then. have you even written a game against dx12 or vulkan yet? because we literally lost ~30% performance on both nvidia and amd cards in our testing when we ran graphics and compute concurrently rather than running our compute shaders between frames.
something tells me you most likely have no real world experience and are just regurgitating AMD marketing memes
>>54584000
>we
Yes, it still has no hw async
http://videocardz.com/59962/nvidia-geforce-gtx-1080-final-specifications-and-launch-presentation
Who gives a shit, benchmarks will show how well the cards actualy perform and thats what matters in the end
>>54583843
because every task utilizes every resource completely and evenly for the entire duration of its execution?
> entire point of async compute is to use ALUs for compute while graphics shaders are bottlenecked by tessellators, ROPs, mem stalls, or whatever
>>54583161
>adoredtv memes
no its not maxwell on speed. its very different but its not a fully new architecture like volta will be.
>>54585011
>its very different
how is the core of the GPU actually different?
just seems like they added a little more color compression and the multi-projection while leaving pretty much everything else identical.
the preemption granularity looks to make VR finally not shit on Nvidia, but it's not exactly doing what people consider async compute on PS4/xbone/radeons.
>>54582062
Preemption can increase the performance, because it bridges the latency time of the graphics memory. So there is nothing bad here.
>>54583843
>dedicating 100% of resources to one task before moving onto the next
>better than doing 2 things at the same time