1. Thread posting restricted: Only the BeamNG team members can post new threads in this sub-forum in order to prevent spam. Posting is available on unlocked threads.

Development screenshots #1 [Closed]

Discussion in 'Microblogs' started by gabester, Sep 5, 2012.

Thread Status:
Not open for further replies.
  1. JDMClark

    JDMClark
    Expand Collapse

    Joined:
    Aug 10, 2013
    Messages:
    1,347
    Well yeah but its game optimizing. They can push the game as hard as possible while having it optimized for lots of different PCs.
     
  2. moosedks

    moosedks
    Expand Collapse

    Joined:
    Nov 4, 2012
    Messages:
    1,113
    New kaveri apu's don't seem like a huge leap from the older ones. Seems like they're faster (or more optimized?) but clocked lower so the cpu part benchmarks almost the same as the richland apu's. GPU is supposed to be better though. Lower TDP than older ones also.

    Apu's are a bit dumb though. The Gpu is still crap for serious gaming and if you want to get the best out of it you need to buy high speed ram which costs a buttasston. Better off getting a dedicated graphics card, cpu, and some 1333 or 1600 ram.
     
  3. metalmuncher

    metalmuncher
    Expand Collapse

    Joined:
    Aug 6, 2012
    Messages:
    257
    I think the appeal to someone like estama is the possibility of using HSA and hUMA in his programs, or stuff like the huge cache on iris pro. Anything is probably better than his current rig, raw performance is probably not as important as the special features you get from an APU type system.
     
  4. TheAdmiester

    TheAdmiester
    Expand Collapse

    Joined:
    Aug 7, 2012
    Messages:
    579
    On the plus side at least you're the go-to guy for low end optimisation testing.
     
  5. Bubbleawsome

    Bubbleawsome
    Expand Collapse

    Joined:
    Aug 5, 2012
    Messages:
    1,887
    While we are talking about optimization I'd like to point out that the original core i series is still not getting good fps. I have a 4/8 lynnfield running 3.2 to 3.6 ghz and it has a problem with more that 3 cars. :/
     
  6. Cwazywazy

    Cwazywazy
    Expand Collapse

    Joined:
    Dec 1, 2012
    Messages:
    1,245
    With the stock Moonhawk I get about 25FPS on my CPU at 4.2GHz (I get 19-20 on stock speed) Does this mean I would get maybe 35-40FPS with the update?
     
  7. redrobin

    redrobin
    Expand Collapse

    Joined:
    Aug 21, 2012
    Messages:
    606
    4 cores and 8 threads. That's purely an optimization problem. I saw Gabe's specs somewhere and I do believe that he's got an original Core i as well, coupled with a couple GTX 500 series cards. Granted... I think Gabe had his overclocked, now that I think about it. For me, in AMD land, the game becomes unplayable at about 8 cars on a 6 core system. I think Gabe said at one point that it should be a car per core, so in theory, you SHOULD be able to run 8 cars without a problem. But, again, it's an optimization problem.
     
  8. n0ah1897

    n0ah1897
    Expand Collapse

    Joined:
    Jan 31, 2013
    Messages:
    625
    IIRC it was one core per physical core. I also remember them saying that hyper-threading doesn't help much either.
     
  9. Bubbleawsome

    Bubbleawsome
    Expand Collapse

    Joined:
    Aug 5, 2012
    Messages:
    1,887
    HT gives around a 30% gain. I should be able to handle 3 cars without a dump to sub-20s though. The moon hawk and h45 by themselves smash the poor CPU into the low 30's. I think it is poor STP though, the 4770k would double my power.
     
  10. SixSixSevenSeven

    SixSixSevenSeven
    Expand Collapse

    Joined:
    Sep 13, 2013
    Messages:
    6,958
    YOu cant limit it like that in code. Software can only spawn threads. Windows then manages threads itself and will shift them around cores as it sees fit. Nothing stopping it from going onto a hyperthreaded "core". BeamNG spawns 1 thread per car. On a single core system they will all be processed on the single core the same way you already have a good 100 or so threads running on your machine already, on a 16 core machine (they exist, as do higher counts) there is room for the threads to spread out.
    Windows has a tendency to group all of the windows services onto core 0 and then start applications on core 1. From there it will manage them based on actual need.
     
  11. n0ah1897

    n0ah1897
    Expand Collapse

    Joined:
    Jan 31, 2013
    Messages:
    625
    Yeah, I just thought that while back one of the devs said it was one car per physical core.
     
  12. logoster

    logoster
    Expand Collapse

    Joined:
    Sep 5, 2012
    Messages:
    2,084

    thats because they DID, in a livestream, in chat
     
  13. estama

    estama
    Expand Collapse
    Developer
    BeamNG Team

    Joined:
    Aug 7, 2012
    Messages:
    268
    Exactly. I mostly care about new architectures. From that point of view, Kaveri's architecture is very interesting to me. But on the other hand, the Data Mining guy in me loves the huge cache of Iris Pro. Iris Pro should be marketed as a purebred DB machine.

    From an optimization point of view, i value my current setup. Because everything is a bottleneck it is very easy to see if something improves or not the performance, without expensive profile tools. The optimizations we do are mostly targeted towards a lowest common denominator, with a preference to Intel CPUs. So if some optimization improves the performance on AMD architectures but decreases the performance on Intel architectures, we take it out. I had to throw away a 5% optimization on AMD architectures because it negatively affected Intels.

    From a performance point of view, and based on the benchmarks that we have done, the fastest machine (heavily overclocked) that we have a benchmark from, is ~8x faster than mine (which is the slowest of them all). This is because our main bottleneck, on the physics side, is memory bandwidth, rather than CPU speed.

    Concerning hyperthreading, it doesn't help that much heavily optimized code like ours. Hyperthreading is opportunistic. It helps wherever the main thread "stalls". In our case the main threads do not stall that much. So only real cores and aggregate cache size matter the most.

    Dedicated GPUs aren't that interesting to me. Having a huge amount of processing power, at the other side of the mountain (PCI bus) is self defeating. APUs bring all this power a lot closer. I can imagine a Kaveri+ architecture being able to considerably change the rules of the game. In an architecture like this, the CPUs would be used to prepare and orchestrate the data flows to the GPU processing units. As it is right now, and IMHO, Kaveri's CPUs are too weak to be able to support the GPU side.

    - - - Updated - - -

    Well, we need to spend all these hours optimizing, because hardware is always getting faster, *but*... The hardware guys don't go blindly optimizing for everything. They see what software is getting sold (specific AAA games) and optimize for these cases. They know what software the review sites will use as benchmarks, and it doesn't involve high frequency physics simulation...

    With the latest optimizations, the delta from my system (which was released in 2008), from the fastest current systems, is ~8x. Based on a relaxed Moore's law (the performance of systems doubles every 2 years), the delta is right (or higher) of where it should be (2^((2014-2008)/2) = 8). Without the optimizations, trying to make the most of systems that aren't optimized for our processing needs, the performance delta would have been considerable lower.
     
    #1953 estama, Jan 15, 2014
    Last edited: Jan 15, 2014
    • Like Like x 1
  14. Bubbleawsome

    Bubbleawsome
    Expand Collapse

    Joined:
    Aug 5, 2012
    Messages:
    1,887
    So memory overclocks help here? That's different.
    I also assume that haswell-e with DDR4 will smash everything, being an 8/16 with DDR4 and all. o_O
     
  15. AirSKiller

    AirSKiller
    Expand Collapse

    Joined:
    Aug 7, 2013
    Messages:
    229
    Yeah, some optimizing must be done, that's for sure. What I was saying is that taking it to an extreme is very time consuming and often brings little extra performance delta. And, in my opinion (repeat, MY opinion haha) when you have the option to make it run better or prettier, prettier is better because, 'With time, slow becomes fast, but ugly, will always always be ugly'' (''AirSKiller's Quotes' 2014'' xD)

    - - - Updated - - -

    Memory bottleneck is not solved (most of the times) by simply slap extra MHz's (did the test, my memory running at 1330 and 2400 (can't go higher) brought about 5-10% extra fps's in BeamNG, only :/)
     
  16. n0ah1897

    n0ah1897
    Expand Collapse

    Joined:
    Jan 31, 2013
    Messages:
    625
    But there isn't really much that Estama can make prettier. He codes the physics side. Sam and Gabe make it look prettier, and they are definitely doing a good job at that :D
     
  17. AirSKiller

    AirSKiller
    Expand Collapse

    Joined:
    Aug 7, 2013
    Messages:
    229
    Yes, I know that and they really are! Hell, all of them are doing a great, awesome job! :O hahaha
     
  18. simon48

    simon48
    Expand Collapse

    Joined:
    Aug 8, 2013
    Messages:
    203
    The CPU and GPU work in parallel so you have to optimize for both. Making the physics run faster doesn't give you anymore FPS to give to graphical improvements.
     
  19. AirSKiller

    AirSKiller
    Expand Collapse

    Joined:
    Aug 7, 2013
    Messages:
    229
    Yes I knoooowww, please don't all start stating that now :l I was just giving out my general opinion, it was not strictly about BeamNG.
    But 'prettier' in physics terms could mean running them at 2500x a second instead of 2000 like now. (If it changed anything, I don't know)
     
  20. pf12351

    pf12351
    Expand Collapse

    Joined:
    Feb 24, 2013
    Messages:
    396
    IRL, that warning is in a jeep cherokee. I would expect that in this but not in a jeep! forester09warning1.JPG
     
Thread Status:
Not open for further replies.
  1. This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
    By continuing to use this site, you are consenting to our use of cookies.
    Dismiss Notice