Multicore optimisation tracker

Discussion in 'Player Support' started by donpost, Nov 27, 2012.

  1. donpost

    During the beta I took some data about how well PS2 was using my CPU and compared it to battlefield. I've taken another set today and I'll keep updating this thread after each optimisation so we can see if it's improved or not.

    Here is what I got during the beta:
    [IMG]

    And here's what I got today:
    [IMG]

    If you want to post up your own data all you need to do is run PS2, then press alt-f and turn the graphics down (probably wont need to), until it says [CPU]. Then you just need some logging software to get the data. I used AIDA64 extreme because it had a free trial.
    • Up x 7
  2. Crille

    That's some interesting observations, I'll be following this thread!
    • Up x 1
  3. Fishpoke

    I'll definitely take some measurements and post results, the more people that do this the sooner the game can be properly optimized hopefully!
  4. DonnyD.

    Been keeping track of this as well . Core 6 , yours not mine, is interesting to me you are spiking to 90% usage, for me I only go to ~78% on that main thread. I am not sure that the difference in clock speeds and minor changes between the 1090 and 1100 are enough to account for that. Some things you should add are what settings you are running at and exactly what video cards you have HD6700 series could be any number of cards, I think there are like 10 6700 series cards.

    EDIT: I am color blind and did not see your core 2 line there core 6 is not your main thread and why it is spiking to 90% usage I have no clue. For me I have 1 core ~78% max 4 cores hovering around 30% and one core around 50%.
  5. XRsyst

    What you're doing is well-intentioned but ultimately not very useful.

    Here are some of the considerations that detract from the usefulness
    • Volatile test environment - given that CPU load will change depending on what's going on around you, a measurement at any given point in time isn't very useful unless a developer can reproduce that moment in time. If not the developer isn't able to make changes and determine, quantitatively, how much of a change they have made and even if that change is beneficial.
    • Why would you want to use more CPU than it already is using? How is there a benefit there? All we can say looking at the charts provided is that the CPU is not what is bottle-necking your experience. Since none of your cores ever gets pegged at 100%, there are never any "wasted" cycles.
    • Your assumption is that the work PS2 is doing should be perfectly parallelizable like what BF3 is doing. While it's an interesting goal, as it's measurable, you should instead focus on a goal that matters like "Better FPS" "Less lag in battles". Those are likely not CPU bound by those that meet the minimum system requirements. (Also, keep in mind that BF3 might only have high CPU usage because it's coded poorly, not because it's coded well). Also games are notoriously poor at parallelization, espeically since they tend to be bottlenecked on resources that are not the CPU (Memory channels, graphics rendering, state synchronization between the two, etc...)
    • Up x 1
  6. Legitsu

    actually the more work you can offload onto the cpu the better
  7. XRsyst

    And what work should be offloaded to the client CPU? I'll venture if your system meets the minimum specs your bottleneck is either:

    • Graphics card, which has a computationally dissimilar workload to what a CPU is capable of efficiently handling
    • Network latency AKA interweb connection
    • Server capacity (lag caused by the PS2 main server)
    None of these issues can be handled by your CPU and these are the core issues that most people complain about. Given that the dev team has a finite set of resources, do you think they should spend those resources fixing problems that exist or problems that only exist on screen grabs of task manager?
  8. Legitsu

    none of the above actually
    when you have cases where you are seeing less then 70% TOTAL system load but are getting <40Fps then there is a problem someplace
    if you don't belive me install something like msi afterburner or ATT and watch the OSD
    on my system I see cases where the cpu is at less then 30% load and the gpu is at less then 70% load and I am getting about 40FPS tell me whats wrong with that picture
    and even the gpu being at 70% is way to high for how 'good` the game looks .....
  9. ExquisitExamplE

    I'll preface this by noting that I'm a complete novice when it comes to the technical end of computing, but essentially what I'm seeing here is that PS2 is not using nearly enough of the cpu as it could, and more particularly, does not make effective use of the other cores on multi-core processors. Feel free to correct me if I'm wrong here.

    I suppose this bodes well, because it means there is a lot of optimization yet to be done, and I should hope so, It's a bit vexing to get a constant 20-25 fps in combat situations.
  10. XRsyst

    If neither a single core, or the GPU is pegged, then the bottleneck is somewhere else, I would guess it's one of the following:
    * memory bus that's synchronizing the two (unlikely)
    * something to do with waiting to render a frame until it's received network info from the server
    * Loading textures from disk, across the bus to your graphics card main memory (this could be happening a lot depending up on their implementation. One only has to look to the Unreal 3 engine to see a crap implementation of this)
    * or what I think is the most likely is that your GPU is actually pegged, but it's not in the core processing, it's some other aspect (memory loading, some sub component)


    I don't program this sort of stuff on a daily basis, but I do deal a lot with parallelization and perf optimization so I know a bit about what I'm talking about.
    • Up x 1
  11. Legitsu

    AMD Phenom II 955 @ 4.0Ghz NB@2600 HT@2400(yea I know its nota 4.5Ghz 2500k) but seriously highest I have seen is 50% load
    no disk access during the situations mentioned
    6870 @ 940c/1250mem(256bit RB) [ I am only running at 1600x1200 so membandwith should not be a issue)
    GPU Z is fairly accurate when it comes to displaying the load of the GPU
    and you are not the only one thats been known to hack around in the innards of game engines
  12. Legitsu

    I am pretty sure the problem is with the way they handle the occlusion of non-view-able ents and other pre-render issues
    • Up x 1
  13. XRsyst

    So now I'm genuinely curious, how would them not handling occlusion well still cause the CPU and the GPU not to be maxed? Certainly the bottleneck is somewhere :)
  14. Deladin

    I understand what your trying to do, but SOE really just did not care about our performance problems as much as making money from the beginning. Now they are trying to play catch up, but are handcuffed because it would take sweeping coding changes to optimize properly.

    If they cared, they would have focused on utilizing multiple cores from the very beginning, as they knew this would be a CPU heavy game due to all the physics they have to process with large numbers of players. The scope of the game they wanted to portray and use as a selling point, far outweighed there ability to deliver.

    Second they would not have built a supposed revolutionary game on old and outdated Direct X 9, when DX11 is far more efficient, even if the graphical difference is not major in all games, it offers better performance. DX9 was released in 2002, and surely when the new Xbox system is released it will have DX11 support, meaning PC titles that are generally ported from consoles will not need to carry on with DX9 as a cheap means of porting the game, and Nvidia/AMD will likely just stop supporting DX9 soon after on their GPU's.

    It almost seem SOE PLANS on this game flopping after a short time, as they seem to put zero longevity into it, both in the coding/software side, and the game play side.

    Now the real issue is they just ignored AMD users all together. Intel users get pretty good performance compared to us. With my specs in my sig, and with all 8 cores unparked, I can only get between 20-50 FPS. Average is about 25-30 in large fights, dropping below ten in heavy heavy engagements.
    • Up x 1
  15. Rippy

    I will try and explain what the issue is and indirectly why more performance per core is generally better for gaming then having more slower cores even if the theoretical performance might be the same (Sandy Bridge/Ivy Bridge vs AMD FX).

    If you see 50-70% CPU load and are still CPU bound does not mean that your CPU is not fully utilized. What is represented by this number is the average CPU load over a certain amount of time and not "per frame".

    Now imagine a multicore CPU running several threads (lets say a dozen). Some work fine in parallel and some must run on just one core. Worse still some threads might have to wait for the result of others before they can begin. If you have a very "horizontal task" where certain threads wait for results or combination of results from threads before it, and some can't run across multiple cores you will see "50-70%" CPU load, but the whole task is going as fast as it can "and one frame pops out".

    In such cases you want to have the fastest per core CPU you can get your hands on.
    • Up x 1
  16. PoopMaster

    Similarly, my Phenom II 940 CPU is never used above 40-50 percent in the game.
  17. SilentStrike

    somthing funny to me that I have noticed, most of the people that have major issue with frame rate have top of the line computers
    I have 0 lag ...ever I do have momentary lockups (which I attribute to my video drivers as I have always had them in every game I play)

    my system

    amd phenom II x2 555 BE at 3.8ghz (3.2 stock)

    ati 5670 under clocked 720core 950 mem (stock is 800 somthing core and 1000 somthing mem)(for stability, I noticed if it goes above 40c the drivers crash i dunno lol)

    8gb giel ram overclocked running at 717 and 1433 8-9-9-25 timings (stock is 667 1066)

    120gb intel ssd for windows
    1tb for games and videos

    650watt 80+ silver psu

    MSI 890GXM-G65 motherboard



    now I do play with all settings on lowest settings except for render quality is at 100% field of view is at 100% and textures are on high
    the funny thing is my step dad tuns on an old dell that does not even meet min specs (video card) and 4gb value ram, and an old intel pentium D all on stock settings and he is able to play on all high settings with roughly 10-20 fps or all low settings with 0 lag as well

    everyone that I have seen with lag issues are people with I7's and 6 core amd's with high end nvidia and amd vid cards and 8-16gb high end low cas ram rofl

    thats why I made the joke to my friends that its almost as if the "scaling" is backwards

    I have no background in programming I am just a mechanic so I have no idea why its like this, just funny to me
  18. Zerulas

    Everyone has different standards of what they consider a good framerate.
    Some are happy with 20fps, some strive for 120fps on their 120hz monitors, and personally I'm aiming for 60fps.
    So yeah I'm still hoping the devs actually make some progress with their cpu optimization, as they've made very little since I first started playing 4 months ago. I pretty much get the same 20-100 fps and 40% cpu utilization now as I did 4 months ago.
  19. Clonosaurus

    My standards for a good framerate involves it basically being playable (probably staying above 20-25). I've never seen a game be this CPU dependent.
  20. SilentStrike

    for an FPS/flight combat I aim for at least 60+ as 30FPS just seems to lag for me
    for an rpg 30 is ok lol