New GeForce drivers with DX11 enhancements (and SLI improvements) are released

Roobubba

Someone on these boards mentioned this recently (I apologise, I forget who and can't be bothered to search!). NVidia just released the new geforce drivers with allegedly global (sometimes large) DX11 improvements.

I'm not able to test these for at least a couple of days, but thought I'd let everyone know

here's the info link: http://www.geforce.co.uk/whats-new/articles/nvidia-geforce-337-50-beta-performance-driver

halfofaheaven

Just in time as well, as I have a 144hz monitor on the way. I'll take any fps boost I can get.

Dictator93

If you have an SLI set up I do not recommend these drivers for NS2. The Shadow Rendering Fix is not included in these drivers (but rather will be in the udpate after this one).

Or at least, I imagine it will be in the driver post this most recent one (the Nvidia rep assured me they were working on a fix).

But back on topic, if you have SLI and want to play in DX11, still use 327.23 wit custom SLI bits for DX11 as pointed out in my thread. Sad they could not have the fix included in this driver branch though.

Mige

Not sure about the performance, but these drivers did fix my microlag problems no more 60ms frametime spikes and frametime usually is more flatter than ever.. AND I think input lag is better

joshhh

That sounds promising. I'll probably be checking these out tonight.

Cr4zyb4st4rd

After testing my old drivers (334.89) on both DX9 & DX11 There is a minimal improvement to DX11 on the new drivers (337.50) However, is DX11 worth using now? No.

I ran a loop around Veil with multiple hives/harvesters/rts etc dropped in the same locations for the following fps @ 1080p

NS2 dx9 (334.89)
Frames: 31188 - Time: 178812ms - Avg: 174.418 - Min: 120 - Max: 202

NS2 dx11 (334.89)
Frames: 28970 - Time: 179047ms - Avg: 161.801 - Min: 104 - Max: 193

NS2 dx9 (337.50)
Frames: 31054 - Time: 177312ms - Avg: 175.138 - Min: 117 - Max: 202

NS2 dx11 (337.50)
Frames: 29503 - Time: 178547ms - Avg: 165.239 - Min: 108 - Max: 192

The most obvious reason as to why you don't use DX11 NS2 comes when you look at how horrific frametimes are! Here's a graph comparing the frametimes of DX9&DX11 on the new 337.50 drivers

LOWER = BETTER
Less variation = less stutter



DX11 still has terrible frametimes in NS2, which makes it feel terribly slow to respond and it still hits massive stutters continuously, I noticed more stutters in DX11 on the new drivers than the old ones, although they're almost half the time old drivers hit 250ms+ frametimes new ones are average 150ms. They both still suck.

Specs:
4770k @ 4.5GHz
GTX680 2GB @ 1202Mhz/Mem 6200Mhz
16GB 2400Mhz
Game & OS on SSDs

TL:DR: Keep using DX9

Sammey

Read NVIDIA post
Immediatly went here
Read Cr4zyb4st4rd's post
Got sad

Dictator93

I will run a test too. Btw Cr4zy, what did you use to test and benchmark? I would like to useyour method and style for consistency.

Mige

Well my frametime graph (DX11 & EVGA Precision + public game) was more like your DX9.. good for me then

Cr4zyb4st4rd

Fraps to benchmark, excel to modify values to get actual frametime (because fraps adds to last frametime so just take the previous frametime away from current frametime and you'll get the render time for that frame) then plotted in Excel onto a Scatter graph (with straight lines) had to tweak it in photoshop a bit because excel didnt want to put the DX9 plot ontop of DX11 for some reason.

matso

There is actually a way to save and load "scenarios" in ns2 - basically, it saves where all non-player entities are and restores them on load. Checkout ScenarioHandler.lua for help. Wrote that back when I needed to setup benchmarks and overload the server for performance checking.

Anyhow, there is an old, old OLD scenario for summit that used to cause the server to croak back before ... oh, b180 and so. You can load it by starting a summit map, jump in on the marine side and then use the following console commands
"cheats 1"
"damage 0.01" (to let you survive all the hydra fire - the scenario was built pre-per-player hydra limit)
"scenload xl" (loads about 300 entities)
"p_logall" (starts the performance logger; also prints out the filename)
Run around one loop until you get back to the same place. You may notice a lot of hitching.
"p_endlog"
"p_logall"
Run around one more lap. You will probably not notice any hitching at all.
"quit"

Find the perfanalyser.py program in the "natural selection 2/utils/PerfAnalyser". Make sure you have installed python64. Run it, and find the logs in %APPDATA%/Natural selection 2/. Open the logs and browse the data.

On my old 3Ghz I7, GTX570 my framerate increased from 82 to 89 with the new driver. The overview for the runs looks something like



As can be seen, the first lap around the map causes all kinds of hickups for me - this was a freshly booted machine, and ns2 is on a non-ssd disk, which probably explains it; there are probably some kinds of data that needs to be loaded from disk while you are playing the game

Cr4zyb4st4rd

I should mention all my loops of the map while not perfectly the same were all run once before hand to make sure everything that should be loaded was. And all were fresh game starts/restarts with a reboot for the driver updates. All that hitching has always existed for me with DX11, maybe others are fine sounds like @Mige doesn't have issues and the only real hardware difference is 780/680.

All I know is my system likes DX9

I might test that stuff at some point matso, didn't know it was possible looks interesting.

Soul_Rider

I found that I reduced the micro stutters by turning down the polling rate of my mouse from 1000hz to 500hz in DX11. It really reduced frametimes.

I don't know why there should be a connection, but it was the best update I made to my gameplay in NS2

halfofaheaven

matso wrote: »

On my old 3Ghz I7, GTX570 my framerate increased from 82 to 89 with the new driver.

DX9 or DX11?

I don't like updating drivers.

Mige

matso wrote: »

As can be seen, the first lap around the map causes all kinds of hickups for me - this was a freshly booted machine, and ns2 is on a non-ssd disk, which probably explains it; there are probably some kinds of data that needs to be loaded from disk while you are playing the game. DX9 will be more stable to play than DX11 but these drivers did wonders to me and I think I´m back to DX11.

I bet that is the same old reinforced patch tweaks.. Lower the memory usage and load necessary files on the fly, so second run will be more smoother . Easy to test high textures and they will give more hickups than low (when alt tabbing you need to load same things again).

Cr4zyb4st4rd wrote: »

I should mention all my loops of the map while not perfectly the same were all run once before hand to make sure everything that should be loaded was. And all were fresh game starts/restarts with a reboot for the driver updates. All that hitching has always existed for me with DX11, maybe others are fine sounds like @Mige doesn't have issues and the only real hardware difference is 780/680.

Well many little things can give trouble to ns2 and only difference is that my cpu is now 4670k with the 780 (4 run to get better cpu, 4770k is waiting to buyer) Dunno what fraps do, maybe I should test it. But I can confirm that dx9 is still more stable, but I wanna join the dark side again (was using DX11 before reinforced)

Dictator93

OK: Ran some FPS tests with the latest drivers. BOth in DX9 and DX11

Rig:
GTX 570 SLI (1280mb VRAM)
Core i7 930 @ 4.2 Ghz
12 GB RAM @ 1600 Mhz

Settings: Textures Medium (High textures stutter due to that damn bug)
Everything to highest, PhysX GPU, no FXAA (ugly), AO High, etc...



Red is DX 11
Blue is DX 9

Me running around Biodome in the same path with very few entities (I will due an entity run later).

DX11 average is 117,4
DX9 Average is 117,0 (apparently exactly that amount)

They both dip with slightly different frame amounts at the same times except for one or two anomolous occassion. They basically perform the same on my RIG with low entity count.

I get a feeling though that DX11 will run better with more entities. Will try that out soon!

EDIT: ENtity RUN using Matso config on summit.



Colors reversed this time due to laziness.

Red is DX 9
Blue is DX 11

DX 9 average is 109,45
DX11 average is 106,9

Once again, not much of a difference between the two if at all.