295 replies to this topic

[xWiredx]

Used to? No, he -is-. The difference is he doesn't blindly support something in the face of mounds of contrary evidence.

[Goose]

Exarch Levin, on 23 December 2014 - 08:17 PM, said:

That's the go-to answer but I keep expecting at least 1 core to be at 100% utilization if that's the case and even at stock clocks I'm still not seeing max utilization.

sys_budget_soundCPU = 0
sys_MaxFPS = 60
d3d10_TripleBuffering = 1
d3d11_TripleBuffering = 1
d3d9_TripleBuffering = 1
 
e_GsmCache = 1
r_FogShadows = 0
 
q_ShaderWater = 0
 
r_silhouettePOM = 0
r_UsePOM = 0
 
sys_budget_streamingthroughput = [x * 1024]
sys_LocalMemoryGeometryStreamingSpeedLimit = x
sys_LocalMemoryTextureStreamingSpeedLimit = x
sys_streaming_max_bandwidth = x
 
r_WaterUpdateThread = 0
sys_TaskThread1_CPU = 0
sys_streaming_CPU = 2

You get "x" by typing "winsat mem" into an command prompt.

Don't forget to post results …

Edited by Goose, 23 December 2014 - 08:49 PM.

[Bill Lumbar]

xWiredx, on 23 December 2014 - 08:40 PM, said:

Used to? No, he -is-. The difference is he doesn't blindly support something in the face of mounds of contrary evidence.

Oh.... I see. Funny, he doesn't sound like a Amd Fanboy to me then. I think it goes much deeper then that, and seems he is Butte hurt that his favorite brand hasn't taken the crown back yet, or maybe its his Amd chip didn't hit the OC lotto. He claims that I have typed all these words, but haven't said a thing, or made a line in the sand on where I stand, or what I support.

I conceded that and agreed Intel CPU's put out better performance numbers in benchmarks, and fps in majority of games. Yet I haven't said where I stand?

I have never claimed my system can match yours in raw numbers it puts out, or beat yours, yet it seems it has been claimed I did. Is that what this is about for you guys that like to have the fastest things out your can afford? In fact, I stated that your machine has more power then mine when it comes to the fps it can push in game. Is the what the content of some of your replies stems from, do I need to take a pic of me bowing down to your machine and ever lasting quest to push the limits of fps in a game that is very well known for being poorly optimized for all hardware?

Seems like he has missed or dismissed most of what I have posted in replies here and in other threads regarding the topic. I believe its the latter of the two, but I could be on an Island here.

Is this what you all seek, would you sell you soul to obtain this? When is enough really enough? I wonder if this system would be enough to bring this game to its knees, or would it too fall to the beast of a game MWO has become?

https://www.youtube....e&v=RDDinmd0Xns


https://www.youtube....e&v=FI1ylg4GKv8

Edited by Bill Lumbar, 23 December 2014 - 09:47 PM.

[Exarch Levin]

More arcane wizardry with the config files? It's depressing that in a PC-only game we have to resort to bad-console-port style tweaks. Here are my results:

Run Time:					00:00:05.12
Memory Performance: 28393.32 MB/s
Total Run Time:		   00:00:06.27

Edit: So I'm to replace x with 28393?

So I used the RadeonPro program to force my GPU to max clocks constantly and my GPU usage was still spotty, 100% one second and 12% or so the next and my FPS was in the 30s-60s on DX11. Of course, with nothing else changed, when I spectate I'm getting 80+ FPS with multiple mechs and enough Clan LRMs to blot out the virtual sun on screen. That makes no sense to me and it proves that the CPU can render the game flawlessly. What is PGI/CryEngine doing wrong?

Edited by Exarch Levin, 23 December 2014 - 09:44 PM.

[Exarch Levin]

Wow. I went through and added those commands into my user.cfg and tested 'em out and now I'm getting ~60 FPS regularly, with occasional dips when entering a "new area" and/or panning. Momentary lag is absent so far and FPS is a lot smoother. Not spectating-level smooth but much closer than what it was.

That triple buffering seems to be working wonders.

[Smokeyjedi]

Goose, on 23 December 2014 - 08:46 PM, said:

sys_budget_soundCPU = 0
sys_MaxFPS = 60
d3d10_TripleBuffering = 1
d3d11_TripleBuffering = 1
d3d9_TripleBuffering = 1
 
e_GsmCache = 1
r_FogShadows = 0
 
q_ShaderWater = 0
 
r_silhouettePOM = 0
r_UsePOM = 0
 
sys_budget_streamingthroughput = [x * 1024]
sys_LocalMemoryGeometryStreamingSpeedLimit = x
sys_LocalMemoryTextureStreamingSpeedLimit = x
sys_streaming_max_bandwidth = x
 
r_WaterUpdateThread = 0
sys_TaskThread1_CPU = 0
sys_streaming_CPU = 2

You get "x" by typing "winsat mem" into an command prompt.

Don't forget to post results …

I started messing with this today.....for 3 drops.....saw no less than 55FPS (flailing torso lock to torso lock)in the heaviest of battles right after editing cfg file......I am stoked for more time to play..............to put this to more testing. Capture some framerates........

[Goose]

Exarch Levin, on 23 December 2014 - 09:31 PM, said:

More arcane wizardry with the config files? It's depressing that in a PC-only game we have to resort to bad-console-port style tweaks.

???

Exarch Levin, on 23 December 2014 - 09:31 PM, said:

Edit: So I'm to replace x with 28393?

Yes, and you multiply by1024 the first on.

Exarch Levin, on 23 December 2014 - 09:31 PM, said:

Of course, with nothing else changed, when I spectate I'm getting 80+ FPS with multiple mechs and enough Clan LRMs to blot out the virtual sun on screen. That makes no sense to me and it proves that the CPU can render the game flawlessly. What is PGI/CryEngine doing wrong?

Once you're dead, your CPU doesn't give a turkey who's got how many HP in their left shoe; It only has to render the play-by-play as given to it by the server(s) …

Exarch Levin, on 23 December 2014 - 11:28 PM, said:

Wow. I went through and added those commands into my user.cfg and tested 'em out and now I'm getting ~60 FPS regularly, with occasional dips when entering a "new area" and/or panning. Momentary lag is absent so far and FPS is a lot smoother. Not spectating-level smooth but much closer than what it was.

Did forget to mention "sys_budget_videomem = y" to you? I'm sorry; "y" is the meg on your card.

Exarch Levin, on 23 December 2014 - 11:28 PM, said:

That triple buffering seems to be working wonders.

Er: That's there in case you turn on v-sync, as v-sync has been know to backfire without triple-buffering and the one listed in the nVidia Control Pannel is strictly for OpenGL; I don't know my way around ATI cards well enough to say you don't need to set it.

No: All the rest of those lines are way moar important to upping fps …

[Goose]

Smokeyjedi, on 24 December 2014 - 11:45 AM, said:

… I am stoked …

I'm pleased to hear it, but I have some bad news: I don't actually have much current development work on quad-cores. :-(

See: From the crappy set of .docs I'd scrounged up, we have

;ca_thread0Affinity = type: int current: 5
;ca_thread1Affinity = type: int current: 3
;r_WaterUpdateThread = type: int current: 5
;sys_main_CPU = type: int current: 0
;sys_physics_CPU = type: int current: 1
;sys_streaming_CPU = type: int current: 1
;sys_TaskThread0_CPU = type: int current: 3
;sys_TaskThread1_CPU = type: int current: 5
;sys_TaskThread2_CPU = type: int current: 4
;sys_TaskThread3_CPU = type: int current: 3
;sys_TaskThread4_CPU = type: int current: 2
;sys_TaskThread5_CPU = type: int current: 1

as the default placement(s) for Teh Twelve Named Threads; As you can see, it uses a numberline of "zero" though "five".

ca_thread0Affinity is the problem childe, as it carps if it's not on a recent Intel spooling in the 4.3+ range, but there are at lest two things most everyone can do about it:

• sys_budget_soundCPU sets a minimum percent of the CPU any given thread has to leave open so the sound card driver can function; It's default if fifteen. Most of my systems have a real sound card, so they can get away with this set to zero; I have access to one box that uses a crappy on-board Reltek, and I seem to be getting away with zero there, too. But a setting of five or ten might be "safer."
  
• Move sys_TaskThread1_CPU someplace else! See how it sits on core 5, where ca_thread0Affinity is? BAD Just move it.

Now here's the hard part: What if you're not a hexacore? Well, it took me awhile, and a fair amount of whiskey, to puzzle that one out, and I feel bad about it talking so long; But early in Lorded's Benching Thread, it finally seeped though my think skull that the thread count just rolls over: When you i5 drivers see core #1 spazzing out, it's ca_thread0Affinity doing it, 'cause 5 minus 4 equals 1.

It's worth noting you cannot simply move Problem Childe itself: It simply ignores such orders.

Now: Remeber how I'd just said "move sys_TaskThread1_CPU"? If you don't have six cores to start with, that means you need to check the list of default placements, above, for all the other threads that need a better housing then hanging with "sharing means gim'me" ca_thread0Affinity. On a quad-core, for example, that means sys_physics_CPU, sys_streaming_CPU and sys_TaskThread5_CPU all need to get out. On the triple-cores, of which I have seen zero testing, you'd move sys_TaskThread4_CPU instead.

But you always move sys_TaskThread1_CPU …

Then we get down to the dual-cores:

ca_thread0Affinity = 0
ca_thread1Affinity = 0
r_WaterUpdateThread = 0
sys_main_CPU = 0
sys_physics_CPU = 0
sys_streaming_CPU = 0
sys_TaskThread0_CPU = 0
sys_TaskThread1_CPU = 0
sys_TaskThread2_CPU = 0
sys_TaskThread3_CPU = 0
sys_TaskThread4_CPU = 0
sys_TaskThread5_CPU = 0
 
ca_thread = 0
e_ParticlesThread = 0

Crytek thinks it's a bad idea, but all them zeros forces the OS, or maybe the microcode, to step in and start juggling threads dynamically; Any port in a storm …

Also, then last two lines actually reduces the number of thread in play, which I think helps.

But really, dual-cores need a way low framerate limit: 40~45 suggests itself, but there's always thirty …

Edited by Goose, 24 December 2014 - 02:03 PM.

[Bill Lumbar]

Goose, on 24 December 2014 - 01:47 PM, said:

I'm pleased to hear it, but I have some bad news: I don't actually have much current development work on quad-cores. :-(

See: From the crappy set of .docs I'd scrounged up, we have

;ca_thread0Affinity = type: int current: 5
;ca_thread1Affinity = type: int current: 3
;r_WaterUpdateThread = type: int current: 5
;sys_main_CPU = type: int current: 0
;sys_physics_CPU = type: int current: 1
;sys_streaming_CPU = type: int current: 1
;sys_TaskThread0_CPU = type: int current: 3
;sys_TaskThread1_CPU = type: int current: 5
;sys_TaskThread2_CPU = type: int current: 4
;sys_TaskThread3_CPU = type: int current: 3
;sys_TaskThread4_CPU = type: int current: 2
;sys_TaskThread5_CPU = type: int current: 1

as the default placement(s) for Teh Twelve Named Threads; As you can see, it uses a numberline of "zero" though "five".

ca_thread0Affinity is the problem childe, as it carps if it's not on a recent Intel spooling in the 4.3+ range, but there are at lest two things most everyone can do about it:

• sys_budget_soundCPU sets a minimum percent of the CPU any given thread has to leave open so the sound card driver can function; It's default if fifteen. Most of my systems have a real sound card, so they can get away with this set to zero; I have access to one box that uses a crappy on-board Reltek, and I seem to be getting away with zero there, too. But a setting of five or ten might be "safer."
  
• Move sys_TaskThread1_CPU someplace else! See how it sits on core 5, where ca_thread0Affinity is? BAD Just move it.

Now here's the hard part: What if you're not a hexacore? Well, it took me awhile, and a fair amount of whiskey, to puzzle that one out, and I feel bad about it talking so long; But early in Lorded's Benching Thread, it finally seeped though my think skull that the thread count just rolls over: When you i5 drivers see core #1 spazzing out, it's ca_thread0Affinity doing it, 'cause 5 minus 4 equals 1.


It's worth noting you cannot simply move Problem Childe itself: It simply ignores such orders.

Now: Remeber how I'd just said "move sys_TaskThread1_CPU"? If you don't have six cores to start with, that means you need to check the list of default placements, above, for all the other threads that need a better housing then hanging with "sharing means gim'me" ca_thread0Affinity. On a quad-core, for example, that means sys_physics_CPU, sys_streaming_CPU and sys_TaskThread5_CPU all need to get out. On the triple-cores, of which I have seen zero testing, you'd move sys_TaskThread4_CPU instead.

But you always move sys_TaskThread1_CPU …

Then we get down to the dual-cores:

ca_thread0Affinity = 0
ca_thread1Affinity = 0
r_WaterUpdateThread = 0
sys_main_CPU = 0
sys_physics_CPU = 0
sys_streaming_CPU = 0
sys_TaskThread0_CPU = 0
sys_TaskThread1_CPU = 0
sys_TaskThread2_CPU = 0
sys_TaskThread3_CPU = 0
sys_TaskThread4_CPU = 0
sys_TaskThread5_CPU = 0
 
ca_thread = 0
e_ParticlesThread = 0

Crytek thinks it's a bad idea, but all them zeros forces the OS, or maybe the microcode, to step in and start juggling threads dynamically; Any port in a storm …

Also, then last two lines actually reduces the number of thread in play, which I think helps.

But really, dual-cores need a way low framerate limit: 40~45 suggests itself, but there's always thirty …

I think its great what you guys are trying to achieve here and what you are doing....but I have to shake my head and laugh my ass off because any game that I have ever played in my entire life of gaming I have never had to go to such lengths to have it play with all the eye candy on, cranked up, and I just find this is a lot how trying to install Linux OS used to be on many rigs.... A complete pain in the ass! WTF!

[Smokeyjedi]

Bill Lumbar, on 24 December 2014 - 02:32 PM, said:

I think its great what you guys are trying to achieve here and what you are doing....but I have to shake my head and laugh my ass off because any game that I have ever played in my entire life of gaming I have never had to go to such lengths to have it play with all the eye candy on, cranked up, and I just find this is a lot how trying to install Linux OS used to be on many rigs.... A complete pain in the ass! WTF!

Took me about 5 minutes total to spread those CPU threads out across my 8 threads as best I could visualize.......heres hoping to tweak it enough to use >25% of any single core.

[Bill Lumbar]

Smokeyjedi, on 24 December 2014 - 02:55 PM, said:

Took me about 5 minutes total to spread those CPU threads out across my 8 threads as best I could visualize.......heres hoping to tweak it enough to use >25% of any single core.

Are you missing my point?

[Smokeyjedi]

Ingenuity....Or critical thinking skills ; see also Problem solving........tinkering, manipulating until I get the desired results. Its really not that big of a task, Your last post took more time to write than I did to tweak the cfg file........

There is a Huge difference between A System buyer(You) And a system builder(me)

*I think of the car guys who pay for horsepower with the skin of their knuckles-these guys know every bolt and screw on the car and could strip and assemble it blindfolded....vs some rich kid who drops $25K of dads money on some bolt in motor and scratches his head with confusion as to what the hell is actually going on............LOL

[Exarch Levin]

Quote

Er: That's there in case you turn on v-sync, as v-sync has been know to backfire without triple-buffering and the one listed in the nVidia Control Pannel is strictly for OpenGL; I don't know my way around ATI cards well enough to say you don't need to set it.

No: All the rest of those lines are way moar important to upping fps …

Which is what makes it intriguing as if I take the triple buffering lines out of my CFG file, while leaving in all the other additions, my FPS is actually worse (dropping into the low 20s) than it was before I started mucking about with the CFG file.

Quote

as the default placement(s) for Teh Twelve Named Threads; As you can see, it uses a numberline of "zero" though "five".

What happens if a "6" or a "7" is used on an octacore processor?

Quote

I think its great what you guys are trying to achieve here and what you are doing....but I have to shake my head and laugh my ass off because any game that I have ever played in my entire life of gaming I have never had to go to such lengths to have it play with all the eye candy on, cranked up, and I just find this is a lot how trying to install Linux OS used to be on many rigs.... A complete pain in the ass! WTF!

I concur with this. PGI has left it to the community to wade through Crytek's documentation and guess-'n-test in order to try to optimize their game for our systems.

Edited by Exarch Levin, 24 December 2014 - 04:46 PM.

[Flapdrol]

Any of you guys ever try the tool "process lasso"? Might be of some use with this kind of user.cfg stuff.

Anyway, I turned on msaa



gpu at max the whole round , cpu as well for some reason. Fps around 60, looks like I have perfectly optimised my system

[Goose]

Exarch Levin, on 24 December 2014 - 04:33 PM, said:

Which is what makes it intriguing as if I take the triple buffering lines out of my CFG file, while leaving in all the other additions, my FPS is actually worse (dropping into the low 20s) than it was before I started mucking about with the CFG file.

That's a new one …

Exarch Levin, on 24 December 2014 - 04:33 PM, said:

What happens if a "6" or a "7" is used on an octacore processor?

I'm shore the answer is "it reverts to default," but we've got Smokey, here, testing, so …

[Lordred]

Goose, on 24 December 2014 - 06:52 PM, said:

I'm shore the answer is "it reverts to default," but we've got Smokey, here, testing, so …

I do not know Goose, I am starting to consistently see MWO assign, (actively) all eight threads on my end since the Dec 22 patch.

I will continue to log and share what I am seeing.

This was a Viridian BOG match (with a substantial OC)

But I was getting consistent activity across all eight threads, where as before I would get 4 with a good load, 2 with a light load, and 2 ignored.

Edited by Lordred, 24 December 2014 - 07:05 PM.

[Bill Lumbar]

Smokeyjedi, on 24 December 2014 - 03:48 PM, said:

Ingenuity....Or critical thinking skills ; see also Problem solving........tinkering, manipulating until I get the desired results. Its really not that big of a task, Your last post took more time to write than I did to tweak the cfg file........

There is a Huge difference between A System buyer(You) And a system builder(me)

*I think of the car guys who pay for horsepower with the skin of their knuckles-these guys know every bolt and screw on the car and could strip and assemble it blindfolded....vs some rich kid who drops $25K of dads money on some bolt in motor and scratches his head with confusion as to what the hell is actually going on............LOL

Smokey, are you now claiming I am just a system buyer, and you are a system builder? I think your epeen proceeds you way to much Smokey. That is the difference between you and I. I think you failed to understand the point being made and I find your comparison about working on a vehicle and claim that it doesn't take any amount of real time to test out a gaming rig when tweaking and overclocking is laughable.

Btw, I work on my own vehicles, and have for the last 20 years. My cars don't go to a shop for any work, I do it myself and never have found myself scratching my head when I work on them. I also type around 75 wpm since the 7th grade... so your other claim I am guessing is also just hot air.

Edited by Bill Lumbar, 25 December 2014 - 07:09 AM.

[Goose]

Lordred, on 24 December 2014 - 07:02 PM, said:

I do not know Goose, I am starting to consistently see MWO assign, (actively) all eight threads on my end since the Dec 22 patch.

But I was getting consistent activity across all eight threads, where as before I would get 4 with a good load, 2 with a light load, and 2 ignored.

Le Mythical AMD patch, or just octacore patch?

I guess the trick now becomes to see if you can pick out ca_thread0Affinity from amongst the cores: If it's still within the second module, then modules now look like hyperthreading to the game …

[xWiredx]

FWIW, I just tried to manually dole out the threads as described above. Using the original, and using several different assignment configurations, I got the same result: lower performance. It was slightly more consistent, but way slower. I went from max fps in the 125-150 range to max fps in the 80-90 range. I went from averages in the 80s to averages in the 60s. That's a 25% drop! Taking that out of my user.cfg for sure. Actually, scratch that, I already did it. It doesn't seem like us Haswell-E people (all 2 of us...) will be needing this.

EDIT: I also see 8 cores pretty active, compared to the 6 when I had that in my user.cfg. I think the latest version might be winning.

Edited by xWiredx, 24 December 2014 - 08:44 PM.

[Smokeyjedi]

Goose, on 24 December 2014 - 06:52 PM, said:

That's a new one …

I'm shore the answer is "it reverts to default," but we've got Smokey, here, testing, so …

I was 2 matches in......1 public drop and 1 Invasion drop were completed.....Duty called and beers were drank..........I am liking what I did see, aiming to make a graph soon......And yes process lasso runs 100% of the time........for me, along with core parking program....

**EDIT** Its late but i ran a few tests matches in a jenner 7Fto make sure i was lrm'd and in faces....brawling at high speed.

I havent figured out how to get that beauty picture of MSI afterburners reading like ^^flapdrol did. so I used fraps to record frametimes......

The info I will release at this moment of beers and late night turkey is from MSI albeit not as pretty as flapdrols screenie?

CPU1(OS-0) managed a 91% max usage
CPU2 managed a 72% max
CPU3 94% max
CPU5 100%
CPU6 75%
CPU7 52%
CPU8 100% max usage....

This was resulting fraps file


I will post my cfg file............

- gp_option_ShowCockpitGlass=0 (off/on)
- r_DepthOfField=0 (off/on)
- r_HDRGrainAmount=0.0 (film grain amount)
- r_motionBlur = 0
r_MultiThreaded = 1
cl_fov = 70 (Default is 75)

sys_MaxFPS = 144
d3d10_TripleBuffering = 1
d3d11_TripleBuffering = 1
d3d9_TripleBuffering = 1

e_GsmCache = 1
r_FogShadows = 0

q_ShaderWater = 0

r_silhouettePOM = 0
r_UsePOM = 0

sys_budget_streamingthroughput = 21250048
sys_LocalMemoryGeometryStreamingSpeedLimit = 20752
sys_LocalMemoryTextureStreamingSpeedLimit = 20752
sys_streaming_max_bandwidth = 20752




r_WaterUpdateThread = 0

sys_streaming_CPU = 2
sys_budget_soundCPU = 0

sys_budget_videomem = 2096

ca_thread0Affinity = type: int current: 5
;ca_thread1Affinity = type: int current: 3
;r_WaterUpdateThread = type: int current: 5
;sys_main_CPU = type: int current: 0
;sys_physics_CPU = type: int current: 1
;sys_streaming_CPU = type: int current: 1
;sys_TaskThread0_CPU = type: int current: 3
;sys_TaskThread1_CPU = type: int current: 7
;sys_TaskThread2_CPU = type: int current: 4
;sys_TaskThread3_CPU = type: int current: 6
;sys_TaskThread4_CPU = type: int current: 2
;sys_TaskThread5_CPU = type: int current: 1

CPU 4 And 8 pegged 100% usage so I may just throw some of that weight off of them both to maybe CPU7 later when i can think straight, love those beers. hate how **** gets hard after a few......

Edited by Smokeyjedi, 24 December 2014 - 10:05 PM.