still settling in

[MADe]

couple of sorties, in a crowd, perfect!

so then my problem was a heat issue?????????

so then some of the basic profiles with the app, it what, tunes the card properly?????

if my issue was heat, then AH3 has NO glitch where I'm concerned.

v-sync disabled
all nvidia settings max, auto thread disabled, triple buffer disabled, pre-rendered 1, DSR 2560 x 1440
AH max all sliders, ALL. post light enabled no options selected, reflection disabled

ave 80f/s in a crowd, 1100+/- VRAM

I will add in the post light selections 1 at a time, I will leave games AA off and reflections off. I hate the reflection and nvidia doing the AA.

Its beautiful, Have I said that.
 

[MADe]

heat issue
solved
the vid card performs better with the Gigabyte Xtreme Game utility. It must work with the cooling better than nvidia driver alone.

I thought I'd be doing myself a favor by not junking up the pc with stuff like this utility. I had not been watching gpu temps cuz the pc is running in the low 30's and so new.
 

Now, At certain altitudes the trees flicker in the white , towns same, all very white shades seem to scintillate constantly.
Whats this?
 

[Pudgie]

Applications do not have a choice as to which cores get assigned to which threads.  That is handled in the operating system thread manager.

Natvely, the game will use up to 3 cores.  That does not take into account all the other parts of DirectX which may be threaded, such as sound, data loading to the video card and so on.

Thanks, Skuzzy.

That's good to know.

 

[MADe]

test 1 and 2 following Pudgie lead.




https://msdn.microsoft.com/en-US/library/ee377058(v=bts.10).aspx

these test done with the reg tweak describe for threads only.

done to see frametime diff if any,      i think there is but...........frametimes seem more consistant with no affinity assignments
the OS thread is maxed.

[MADe]

test 3

[MADe]

I enabled vsync to adaptive in profile no other change.

I started up AH and assigned it to cores 7, 8, 9. I then went thru every other process listed on the list and disabled core 7, 8, 9. Now I was not able to adjust every process but I got most.

It froze up multiple times but kept going. 
1st graph I use as a base line, I was just sitting a runway for 5 min.
2nd was the freeze, temp was acceptable but total system burp, cards, cpu. Frametime shot thru roof as the gpu completely went 0% usage.

You can really see AH using 3 cores. Crowd it starts to fluctuate.

for what its worth

Interestingly when windows left alone, the 1st core is pegged, frametime has a +10ms variance, with tonights setup cores 7, 8, 9, spread a load evening amongst themself, other cores did a little and frametimes were slightly steadier BUT IT FROZE.

[MADe]

today,
i disabled HT in bios. Now just 6 cores.
I adjusted additional worker thread setting to 6 from 12.
v sync disabled
AH given above normal process priority

Like yesterday, I selected 3 cores and isolated them as much as I could by adjusting affinity assignment options for every listed process. I chose 3, 4, 5 cores, removed affinity assignment from all stuff and allowed AH to run on only the 3. Major difference is that yesterday I chose all HT cores for assignments, very bad. Today will be all primary cpu cores!

I used ah with 3 cores, with 2 cores, with 1 core.
With each selection I went to same map, same area and sat the run way as a baseline reading. I went to arena crowded areas.

I dog fought, cruised, died, killed. I would stop and check the graphs at high point moments, kills, deaths, freeze glitches which were few.

By far and away the game runs best on 2 CORES only. The frametime responses appear to flatline the best, with variations occurring with view changes, objects entering vicinity, usual. What appears to me to happen is when the threads, dominant core shifts, it spikes the gpu usage to 0 and frametime 1000ms+.

I do not begin to understand the trigger threshold when 1 core gets maxed and a second is added but that crux point leads to a system glitch, or burp. Using 3 cores makes for more burps as windows tries to balance or pass things around I guess. Either way the burps are more pronounced because there is more core changing. Graphs show as a core dips its second core spikes as well as gpu taking hit, all at same time.

Now if I set up nada and just go with it it appears that the game latchs onto same core as windows OS, core 1. In game you can see the core constantly pegged so there some core handoffs happening and the frametime variation is doubled, with occassional system spikes.

I believe this was/is Pudgie's point. I roll with Nvidia so it seems to happen both.

It does feel better, the game, with it confined and all alone on 2 cores. They both get used more than 50% and the handoffs seem a lot less.

[Pudgie]

Hi MADe,

I read and saved the document that you linked then went into registry and found the 2 registry settings as follows:

AdditionalDelayedWorkerThread s was already set in registry at 0x10 (16) so nothing to do there.
AdditionalCriticalWorkerThrea ds was found in registry set at default (0x0 (0)) so I changed this to the recommended setting of 0x10 (16). This 1 I find critical as from understanding the importance of these worker threads to process time-critical work thus the stack is kept active in system mem for availability (with the fact that I possess a LOT of system mem onboard (16Gb in Quad Channel config) relative to my box's usage so this taking up mem space is a moot point to me (also checked all this in Task Manager and noted no significant mem usage amount change...less than 200Mb).

Rebooted my box and all came up well.....Windows shows to have picked up a little since doing this (little snappier and crisper in operation to me). While I was in the registry I also made a few cosmetic changes to give the appearance of snappier performance (reset MenuShowDelay from the default of 400 to 150 and set up NoLowDiskSpaceChecks then set to 1...disables this so Windows doesn't waste time checking my SSD's for disk space as I do this myself quite often).

Went flying under this "new" Windows OS config and all went very well using 3 CPU cores w\ AHIII Dx11 but the CPU core usage distribution was very low.....game ran well but this bothered me so I went and reset the CPU core affinity back to 2 CPU cores for AHIII Dx11 and retest.....now all was looking much better to me and the game ran very well on my box so I'm gonna stick w\ 2 CPU cores for AHIII Dx11 (3 CPU cores works very well w\ AHIII Dx9). Here is a snippet of my box running AHIII Dx11 using 2 CPU cores and reconfig'd OS provided below:

I encountered my 1st screen pause since updating to Patch 12 during this run (note the FT spike on left side of graph) but it didn't last long....less than .25 secs and resumed normal operations. Never saw another 1 since this 1 rest of the night. All ran well on my box.

Continuing to test.........................

 

[Skuzzy]

All you are doing when you disable or enable cores is showing how efficient or inefficient the operating system thread manager is.

Also, when you reduce the core count, you reduce the likelihood of bus contention happening.  Bus contention is the thing where multiple cores need external access (memory, I/O...) and they have to wait in turn as only one core at a time can access external resources.

Now, with reduced cores, also means threads have to be run in time slices which can cause all the threads to be run slower.  A thread only needing to access the cache of the core it is running on can run full speed in parallel with everything else.  That thread would be slower to run if it had to wait for a time slice to run.

[Drano]

I was not gonna OC, but that 113C temp from afterburner reading got me piqued. I mean dam card has 3 fans on it by design, case has mucho airflow...

I use AB to overclock and like it a lot. Plenty of tweaks and monitors. If you're gonna overclock at all you really have to be on top of the cooling otherwise you risk toasting your card. Did you try messing with the fan curve tab? There you can set up a curve of fan speed relative card temp. Just drag and drop the points as you like and check the box to enable it. AB will monitor your card's temp and ramp the fans up or down accordingly. I believe that curve is flat by default which may be why your temps went so high. 113 is pretty brutal! I get askeert if mine get to 80! But at the same time you don't need the fans howling away just to surf the net. The curve is handy for that too. Without looking at it I don't believe it goes below 30%.

 I'm looking at an MSI card that will actually kill the fans completely under minimal loads for quieter operation. Really good coolers on the MSI cards.



Sent from my XT1585 using Tapatalk

[MADe]

All you are doing when you disable or enable cores is showing how efficient or inefficient the operating system thread manager is.

Also, when you reduce the core count, you reduce the likelihood of bus contention happening.  Bus contention is the thing where multiple cores need external access (memory, I/O...) and they have to wait in turn as only one core at a time can access external resources.

Now, with reduced cores, also means threads have to be run in time slices which can cause all the threads to be run slower.  A thread only needing to access the cache of the core it is running on can run full speed in parallel with everything else.  That thread would be slower to run if it had to wait for a time slice to run.

Bus contention, this must be those spikes I noticed. The gpu would suffer at these contention points yes?
 The game runs the best for me on 2 cores, it feels better. Maybe a placebo effect....
but the graphs bear out something..........I think getting it off core 1 and isolating it does help as well.

[Skuzzy]

Bus contention occurs when one core screams, "I need the bus!" and another core yells, "I need it too!".  It is then up to the hardware priority who wins.  One gets the bus, the other goes into a wait state (consider it asleep as it literally has to stop everything) until it can get the bus.

[MADe]

let me ask this.

Are core assignments like a cascade effect?
ie: as demand increase things are diverted to next in line, so to speak.

I understand when you say I'm just moving things around. But I have also isolated the chosen cores from almost all other things.
I also wonder if a core takes precedence 1st, I know windows grabs the cpu 0. I know that when AH loads it uses cpu 0, pegging that core to 100%.
The graphs where I did no affinitization show pegged cores and lots of widely variated measurements. Other extreme which I used affinitization showed game cores not pegged and measurements more flat lined. Not perfect mind you but much less glitchy.

I really think its this thread contention is a possible issue for lesser pc's. Mine just bulls thru it, but I can recover, others????????????????
 

[MADe]

today

HT enabled
worker thread tweak to 12

affinity set for cores 3 and 9 for AH only.
If I understand correctly, I enabled core 3 and importantly its hyper thread companion 9 for AH only. I flew.

bus contention, skuzz new word to google you gave me...............

Unfortunately the arena was under 100 pilots by the time I got home but I got a baseline pic, sitting the tower. The other was right after getting waxed over an enemy field.
Note the lack of spikes, there is variation but no show stoppers, if it is thread contention making the graph spike, then on 1st impressions, mating a core with its HT core appears to have less thread contention.

Now I have to stick with this arrangement for a period of time to see if..........

graphs are best yet, very consistant, clean even.

[Pudgie]

Here ya go, MADe.......................

I finally got around to running your test according to your directions w\ a small deviation..............
When I went to kill CPU core 3 & 9 on the remaining running processes Win 7 SP1 said nada to that......denied all attempts.
I also forgot to go into registry and reset AdditionalDelayedWorkerThread s to 12 and AdditionalCriticalWorkerThrea ds to 12 (had reset these in my later on from the recommended 16 to 8 to give a couple more than the number of CPU cores available....6 at the time) so the results were using 8 instead of 12.

Just ran AHIII Patch 13 Dx11 w\ this configuration......snippet of MSI AB graph below.

Pretty much got the same results as you did in spite of the differences.............

FYI.......................... .