Nvidia or AMD for competitive gaming ?

[stirner]

That would explain that... Is the cursor position update rate in Windows equal to polling rate? What is the simulation interval (framerate) of a desktop environment without VSync?

[flood]

idk
i think it's just software rendering and there really isn't a concept of frames.

[SImonsays]

Actually i said that im experienced about playing CS and CS GO,have ability to play with high ranks(have 1100+350 hours CS GO) and said that i try every guide for lowering input lag.

I had to register for this topic.

You are either inconsistent(Very likely when you say you have 1450 hours of csgo? Or is that 350 of csgo?) or you don't know how peekers advantage works.
Supreme is fairly easy to achieve, it's time to move away from MM and move into leagues and get a team and you will see how much harder it gets.

[lexlazootin]

I had to register for this topic.

You are either inconsistent(Very likely when you say you have 1450 hours of csgo? Or is that 350 of csgo?) or you don't know how peekers advantage works.
Supreme is fairly easy to achieve, it's time to move away from MM and move into leagues and get a team and you will see how much harder it gets.

Yea ecay, you're just shit. join the real leagues and learn about peakers advantage.

lol, you serious simon?

[flood]

well that escalated quickly

[RealNC]

The Windows desktop mouse cursor is hardware rendered, unless you drag something.

To test it: put the mouse cursor on a window title bar. Press the left mouse button and hold it. You will see the mouse cursor flickering for a tiny bit. It's VERY subtle. But you can see it if you watch carefully. This is an artifact of windows replacing the hardware mouse cursor with a software one in an almost seamless way, in order to avoid having the mouse cursor be ahead of the pixels you are dragging across the screen (because the drag is v-synced, while a hardware mouse cursor is not, so they would get out of sync with the mouse cursor leading the way and the dragged window falling behind.)

[stirner]

I knew the cursor was treated differently in the rendering process, but I didn't think about the possibility that proprietary drivers sync the cursor to vblank while default don't.
Do you know if the desktop is always vsync'd or only with aero? I guess the explorer/interface simulation steps are tied to the clock resolution (15.6ms/10ms/1ms/.5ms).

[shish_KEbOB]

In regards to what shish_KEbOB said about pre-rendering, it does sound reasonable that NVIDIA compromises response for raw frame rate performance by utilising more data queuing than AMD. That would at least be in line with their "frame pacing" technique that compromises response for frame frequency performance.

The reason I say it context queue looks better on nvidia is that in all games I tested where you can control render ahead limit (or however the specific game calls it) packet queuing looks exactly like nvidia implementation.
Only source engine games behave differently, with "mat_queue_ mode" just not having a setting for limited queuing, it's either on (unlimited) or off (amd style).

We are talking about games made way before this whole frame pacing issue and it behaves the same way weather it's AMD or nvidia.

Regarding the dreaded frame-pacing, it seems like AMD worked on their drivers to be more "nvidia like" after the anandtech article, I wouldn't really bet on there being much difference.

Inspector is miles ahead of RadeonPro,

Radeon pro is miles behind and will stay there since the guy who made it is not working at it anymore.
Funny thing is that it includes the infamous "flip queue 0" setting which GPUView reveals is simply Driver default, which sometimes manages to bug into "last specified value".
A pretty cheap populistic move in my eyes. Sometimes the community just decides to believe in things regardless of the truth.

[stirner]

Might come down to roughly the same with both... now that I know why the cursor feel changes so drastically I am less skeptical about the drivers. It is however a fact that benchmarking and all that concentrates exclusively on raw framerate performance and a bit on frametime performance on the side. Output latency is not a selling point. Compromising the latter in favour of the former has been done and does seem reasonable from that perspective. Although I am optimistic that will change with the heavy investments that are happening around virtual reality and innovative means of input, where overall I/O response becomes much more of a factor.

mat_queue_mode handles how many processor cores are utilized by the engine. Do you mean when pre-render is set to 1 and queue mode utilizes all cores available for threading (mat_queue_mode 2), each thread can store one pre-rendered frame? Would explain induced input lag with multicore rendering, but it does seem unlikely.

I read about the "flip queue 0" or "pre-render 0" myth as well. The processor has to pre-render at least 1 frame for the GPU to even have anything to work with, so pre-render 0 cannot exist.

[shish_KEbOB]

mat_queue_mode handles how many processor cores are utilized by the engine. Do you mean when pre-render is set to 1 and queue mode utilizes all cores available for threading (mat_queue_mode 2), each thread can store one pre-rendered frame? Would explain induced input lag with multicore rendering, but it does seem unlikely.

Actually it handles both number of threads and pre-render queue.
0 is single thread with no packets on top of presents (same as multicore rendering off in the menu), 1 allows other packets on top of the first present (sorry but with current hardware I cant get two presents, gpu time is just too short) , while 2 is multithreded with unlimited queuing.