[BufferPools] PoolSize=0 the holy grail of FSX performance...

[bojote 23]

DPC Latency checker appears best with a BP setting of 1000000.

Good, then DPC LC its a tool with enough valid information to you, I respect you use of the tool and the way you interpret its results.. its valid to you, based on your hardware, because.. as I already stated, I can't seem to stress the buses enough on my PC to even change latency under heavy load situations. Will it make DPC LC less valuable? of course not! :) will I question the validity of your testing because your 'choice' of tool? no way!

[bojote 23]

however given the above change in VRAM handling we perhaps need to be careful and specify which OS we are using and taking this into account when looking for a reason or condemning other folks results.

Hi Andy, thanks for pointing that out. People here don't 'condemn' results, but 'testing methods'. But don't let that EVER put you down :) it's difference of opinion, to which we are all entitled, so no big deal.

[Noel 4,563]

Good, then DPC LC its a tool with enough valid information to you, I respect you use of the tool and the way you interpret its results.. its valid to you, based on your hardware, because.. as I already stated, I can't seem to stress the buses enough on my PC to even change latency under heavy load situations. Will it make DPC LC less valuable? of course not! :) will I question the validity of your testing because your 'choice' of tool? no way!

For what it's worth when I set BP=0 I get alot more spikes of high DPC latency, up into the red zone, and as I say, for some reason which I only vaguely follow setting BP=1000000 (that's only 1 mb!) I get the lowest DPC latency averages and less spikes. By the time I get up to 10mb it's gets really ugly! I do have to say I get the overall best subjective perf around 1mb, but even that is not totally objective. Before reading this thread I was at 5mb, and performance seemed quite good.Noel

[bojote 23]

For what it's worth when I set BP=0 I get alot more spikes of high DPC latency, up into the red zone, and as I say, for some reason which I only vaguely follow setting BP=1000000 (that's only 1 mb!) I get the lowest DPC latency averages and less spikes. By the time I get up to 10mb it's gets really ugly! I do have to say I get the overall best subjective perf around 1mb, but even that is not totally objective. Before reading this thread I was at 5mb, and performance seemed quite good.

Noel, is there a way for you to influence the latency average valuebesides changing BP? for example, assuming BP=0 can you make latency change reducing screen resolution, antialias, or texture size? by the way, are you using 4096 textures by any chance? I want to replicate your results and completely saturate my bus, othersiwe latency values will remain unchanged in my case.

[Guest Nils79]

Is there a i5i/i7 user who is affected of the DPC Latency issues?It seems to me that mainly C2Q user are affected.Thanks@NickNThanks, Nick! But i lost confidence in my system and i will replace it. So i'm not sure how meaningful my tests are for others.

[yankeeromeo 2]

Hello,BTW, which difference between High and real time- mean Priority setting- ?Cheers,

[bojote 23]

Latency issues are the result of a saturated BUS, do we all agree on that?Latency issues are a 'consequence' of high volumes of data going through a bus, do we all agree also on that?(evidently, the above statements assume NO driver problems or IRQ conflicts) just a lean, clean healthy system)if BP=0 increases your latency on the BUS if this a 'bad' thing?an easy analogy which I'm sure everyone will understand, you have a 10MB network card, connected to a 1000MB Network switch and you try to pump (through the card) 1000MB of data.. what happens to latency on your END? it goes sky high! what happens if you replace your 10MB with a 1000MB network card? no more latency! why? because you are not 'saturating' the BUS any more.. so, was pumping 1000MB through the card a bad thing in the first place? BP=0 stresses your sim, no doubt about it, so if it increases your latency, is this bad?

[Dirk98 354]

Hello,BTW, which difference between High and real time- mean Priority setting- ?Cheers,

In my case the best is above average (in my language O/S), not even High. The High starts giving anomalies immediately, blocking out some inputs, ESC, etc. The above average works great for me, smoothing out the flow.Dirk.

[Noel 4,563]

Noel, is there a way for you to influence the latency average valuebesides changing BP? for example, assuming BP=0 can you make latency change reducing screen resolution, antialias, or texture size? by the way, are you using 4096 textures by any chance? I want to replicate your results and completely saturate my bus, othersiwe latency values will remain unchanged in my case.

I can try some other things and let you know.Yes, I am using a texture max load value of 4096, but mainly only in the Super MD-80 McPhat liveries have the textures of this size. I am using 1024 cloud textures much of the time. My guess is that my bus is just short of completely saturated, but that is just a guess. Do you have a good grasp of exactly what the bus is being tasked to carry, so I can get a better sense of whether or not mine is saturated? I am using a quad core Penryn on an x48 mainboard. I do have settings up quite high as you can see in my sig. Even so, I get really good performance, even with my new heavy the PMDG 747. Love that thing! Almost went for the MD-11, but alas at $29 from BB who could argue with that. I also picked up the J41, and am not overly impressed so far. I am a sucker for really rich sounds and so forth, and this bird seems kind of unimpressive to me, though I am alone I think in this opinion. Very nice textures tho. I am really hoping PMDG would do a 757, but don't think that is coming. The 737 I had in FS9, and may pick it up with the new one comes from PMDG. May eventually do the MD-11.I like to use a batch file to load different fsx.cfg's for different situations. For the 747, I have dialed back clouds to medium density and water to x2 low, and with those changes get quite excellent performance. The J41 I will have to play with because as it is now it is more of a hog than the 747 is. When I bought my QX Quad processor, I did it in the hopes of running FSX good enough until something beyond i7's came along, which I am clearly being able to do. Thanks to Nick for nudging me into better memory and mainboard. I don't feel a need to get anything more potent like the i7's or more as I am getting very good perf right now. The bigger screen would be nice, but I may wait til the big upgrade in a couple more years.Do you think that because gpu-z shows only ~378mb of video memory being used, that a 1GB card is very much overkill for FSX on my platform? I have been loath to move to 1980 x 1200 monitor mainly because of concerns over other games/sims I use that are more heavily GPU-dependent for performance. It will represent at least a 20% deterioration in overall perf in those games I believe. I'm guessing with a screen rez of this size I will then see more video memory being used.

[Guest Nick_N]

Latency issues are the result of a saturated BUS, do we all agree on that?Latency issues are a 'consequence' of high volumes of data going through a bus, do we all agree also on that?(evidently, the above statements assume NO driver problems or IRQ conflicts) just a lean, clean healthy system)if BP=0 increases your latency on the BUS if this a 'bad' thing?an easy analogy which I'm sure everyone will understand, you have a 10MB network card, connected to a 1000MB Network switch and you try to pump (through the card) 1000MB of data.. what happens to latency on your END? it goes sky high! what happens if you replace your 10MB with a 1000MB network card? no more latency! why? because you are not 'saturating' the BUS any more.. so, was pumping 1000MB through the card a bad thing in the first place? BP=0 stresses your sim, no doubt about it, so if it increases your latency, is this bad?

No, it isn

[bojote 23]

My guess is that my bus is just short of completely saturated, but that is just a guess.

Ok based on your information, here are the facts:You said that either HIGH BP values or 0 don't work well for you.For your specific setup, a value of 1.000.000 works perfectly fineNick discovered some time ago that 'increasing' BP values, lowered latency averageYou said that Video Ram utilization remains 'unchanged' (for you) regardless of BP setting.The above are facts, right? well.. there is TWO things going on: First, what your Video Card stores in its video memory HAD to get there via the PCI bus, the PCI bandwidth is limted to X ammount of MBs per second. If you approach or saturate that ammount, where will be an interruptin on the 'flow' of information going on in the BUS, this 'interruptions' can be easily measured checking how all devices in the PCI bus (which is shared) are fighting to get a piece of the pie. The way you measure this is using a Latency tool that can tell you IF devices are struggling with the flow OR if they are happily sending data back and forth without having to 'wait' for other devices to do so.In my specific case, I don't see any variations in Latency.. why? because (on my setup) it appears its not an issue (based on my current configuration) so, again, in MY specific case a latency tool is not going to help me determine where a bottleneck exists, so I have to resort to other means to measure th impact of my specific tweaks.

Even so, I get really good performance,

there is a 'sweet spot' in your case... which is the point where your PCI devices are happily sending data without saturating the BUS, you discovered this by yourself, using a value for BP of 1000000

Do you think that because gpu-z shows only ~378mb of video memory being used, that a 1GB card is very much overkill for FSX on my platform?

Nope, it simply means you are not pumping too much textures into it, which leaves a lot of room, however, think of the following analogy: Imagine filling a 747 fuel tank using straws vs using huge wide hoses. The wide hoses will fill it faster and there is a LOT of liquid flow going on on that hose!!!So, what if you 'increase' your textures? will it work because you still have room in video memory? NO! because FSX works in a per-frame basis, it will pump information through the buses!! if your buses are little straws then you'll experience high latency values, corruption, and weird things.Don't know if this explanation works for you.. I tried my best on using plain descriptive terms for the benefit of all.

[bojote 23]

It can vary how much time the PCIe bus has on the bus as compared to storage or sound that may be sharing that resource as most motherboards share resources between devices including the video card. That share and how efficient it is per system is based on all the devices and their drivers in play.

Every system is different I also said in my post 'assuming you don't have IRQ or driver conflicts' which assumes a healthy system. So, latency tool 'could' be showing LOW latency values on a specific setup, but show no issues of any kind on another, do you agree?

[Guest Nick_N]

Ok, so you like facts? you want proof? you got it!! and you can check this for yourself if you have the right hardware.First, go get yourself a copy of GPU-Z, the latest version shows (in real time) HOW MUCH Video memory a Direct3D application is currently using.[bufferPools] PoolSize=0 Will tell FSX NOT to reserve ANY video memory at all. However, this doesn't mean FSX will not use it... IT WILL!!! and more effeciently because it does so at the driver level.In fact, Setting a PoolSize value (other than 0) makes FSX manage video memory, which might sound the logic thing to do.... however, this 'management' doesn't occur at the DRIVER level.. its a 'hack' by the FSX development team so FSX will not exhaust all the video memory available. this 'hack' is a static memory pool that needs to be constantly erased... guess what happens when FSX swaps memory in this static pool?? you guess it.. STUTTERS!! plus, the penalty of having CPU cycles devoted to this instead of having the driver handle it!!Setting PoolSize=0 allows FSX to utilize Video Memory at the DRIVER level, however, there is a problem with this approach. If you have a video card with LESS than 1GB this tweak will make your FSX crash and/or unstable!! why? because FSX doesn't know (or care) how much video memory you have... it will simply use ALL available video memory via the Direct3D driver until it is totally exhausted. FSX will crash or become unstable when it uses more than 65% or 70% of your available video memory. So even a 1GB Video card might not be enought unless you lower AA, AF and/or screen resolution.You can confirm everything I'm saying by doing your own tests, but you need a >1GB Video card and also the GPU-Z utility to measure Video Memory utilization. This will confirm my findings.I've been using PoolSize=0 since Oct 2008, it has always puzzled me why others don't see any benefit when using this tweak.. I think I found the answer.

Very misleading... and the follow-up that went with it also misleadingthe only fact in that post was that BP is changing the playing field.. what it was changing you really did not know

BP=0 will not work based 'just' on the fact that you have a 2GB! I agree, probably I misled people into believing this was the only requirement? then, I'm wrong, I should have also mention BUS speed to be a huge factor as Nils, Noel and UlfB demostrated!

thank you for correcting that... :(

Every system is different I also said in my post 'assuming you don't have IRQ or driver conflicts' which assumes a healthy system. So, latency tool 'could' be showing LOW latency values on a specific setup, but show no issues of any kind on another, do you agree?

Have you ever stopped to consider that YOUR system and the others that show a response to low BP are the ones that are not playing nice on the bus and systems like UlfBs, mine and other ARE?Perhaps it is your systems that are 'unhealthy' with this application in use as you put it! (not saying that is true either)That is why I got into this topic to begin with. I think it is GREAT that you have pushed the setting for people to try and that people are finding ways to circumvent their issues and get things in order but what I won

[Guest Nick_N]

Most people do not know about the XP 32/64 SP2 issue whereby the HAL may not update correctly if the user applies XP 32/64 SP2 using the web or SP2 download package. The issue is PCI.SYS and NTFS.SYS (SP2 versions) in many situations through using the SP2 web patch may not function correctly unless the user installs XP x32/64 with SP2 burned on the Windows install disk, the full install. No workaround except to pull pcie.sys and ntfs.sys SP1 versions from the Windows install disk and replace them which means the SP2 install is not complete and only a temp fix at best. Because of that, there are going to be reports of perf issues with those OS

[bojote 23]

This is the kind of thing I don