• Guest, The rules for the P & N subforum have been updated to prohibit "ad hominem" or personal attacks against other posters. See the full details in the post "Politics and News Rules & Guidelines."

[VC] NVIDIA GeForce GTX 1060 Specifications Leaked, Faster than RX 480

Page 12 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

sontin

Diamond Member
Sep 12, 2011
3,273
149
106
Every GPU has multiple hardware schedulers. Otherwise a developer would need to the job on an application level.
You can read about it in nVidia's whitepaper for Fermi:
https://www.nvidia.com/content/PDF/fermi_white_papers/NVIDIA_Fermi_Compute_Architecture_Whitepaper.pdf

Coming from YOU who wrongly argued for months that Maxwell has real Async Compute... it's priceless man. You keep it up.
Great argument. :thumbsup:
Any proof that nVidia has lied about Pascal's Async Compute improvements?
 
Feb 19, 2009
10,457
5
76
Every GPU has multiple hardware schedulers. Otherwise a developer would need to the job on an application level.
You can read about it in nVidia's whitepaper for Fermi:
https://www.nvidia.com/content/PDF/fermi_white_papers/NVIDIA_Fermi_Compute_Architecture_Whitepaper.pdf
I hope you realize Fermi was the last NV GPU architecture which has a full hardware scheduler.

NV has moved a portion of it's function to software, ever since Kepler to save on power.

ps. I'm not the one who has to prove it, it's not my job to defend NVIDIA. The conclusion is obvious for all to see, do you see DX12 games where Pascal runs faster than DX11?
 
Mar 10, 2006
11,715
2,010
126
I hope you realize Fermi was the last NV GPU architecture which has a full hardware scheduler.

NV has moved a portion of it's function to software, ever since Kepler to save on power.

ps. I'm not the one who has to prove it, it's not my job to defend NVIDIA. The conclusion is obvious for all to see, do you see DX12 games where Pascal runs faster than DX11?
What NVIDIA did was move the portion of the scheduler that scheduled instructions within a warp to the compiler, but scheduling of warps themselves is still done in hardware.

So, this leads to the following question: do the ACEs schedule the execution of threads within a given wavefront (AMD's equivalent to NVIDIA's warps) or are they scheduling at the wavefront level?
 

sontin

Diamond Member
Sep 12, 2011
3,273
149
106
I hope you realize Fermi was the last NV GPU architecture which has a full hardware scheduler.
Go back and read again what Anandtech wrote:
However based on their own internal research and simulations, in their search for efficiency NVIDIA found that hardware scheduling was consuming a fair bit of power and area for few benefits. In particular, since Kepler’s math pipeline has a fixed latency, hardware scheduling of the instruction inside of a warp was redundant since the compiler already knew the latency of each math instruction it issued. So NVIDIA has replaced Fermi’s complex scheduler with a far simpler scheduler that still uses scoreboarding and other methods for inter-warp scheduling, but moves the scheduling of instructions in a warp into NVIDIA’s compiler. In essence it’s a return to static scheduling.
http://www.anandtech.com/show/5699/nvidia-geforce-gtx-680-review/3

And now read nVidia's whitepaper and look for the "warp schedulers" and what their job is.

NV has moved a portion of it's function to software, ever since Kepler to save on power.
Has nothing to do with Multi-Engine. Multi-Engine is about scheduling "high level" workload. nVidia reduced scheduling of "low level" workload.

ps. I'm not the one who has to prove it, it's not my job to defend NVIDIA. The conclusion is obvious for all to see, do you see DX12 games where Pascal runs faster than DX11?
What has DX12 performance to do with Multi-Engine? Warhammer DX12 patch is 30% slower than the DX11 path. Maybe its more a developer thing...
 
Last edited:

Yakk

Golden Member
May 28, 2016
1,574
272
81
I'm curious to see more information on Async & priority based Async Compute. Essentially if a GPU can run fast enough without using an Async scheduler it can surpass a slower GPU using an Async scheduler. Doesn't mean the faster GPU is using its resources efficiently, just brute forcing it's way more quickly and it still have a tonne of unused resources.
 

ASK THE COMMUNITY