Jump to content
Bohemia Interactive Forums

dnk

Member
 View Profile  See their activity

  • Content Count

    578

  • Joined

  • Last visited

  • Medals

Posts posted by dnk

  3. Posted

    Steam means I can't lose a game. It also is a very nice and easy service for keeping games up to date and maintaining data file integrity (which I can f with on occasion, and it helps being able to one-click fix an installation whereas before it was a full uninstall/reininstall hassle). I don't love it, but I use it for most everything now, and just accept the online requirements as they come since it doesn't cause me issues 99% of the time.

  5. Posted

    Do note that the primary thread only uses one core, and that's the thread that's bottlenecking the game so hooorribly. Simulation-AI-render, all on one thread, one core. They certainly have multithreaded it and moved less significant things (eg PhysX) onto other cores, but the big stuff is still single-threaded. It's also why a headless client that takes the AI off the main thread can greatly improve performance. That's a hack that we shouldn't still be having to use to get true multithreaded performance out of this engine, though. They need to make that hack obsolete.

  10. Posted

    I was referencing specifically Arma's boot times as "got more money than you need", didn't realize we'd expanded out into general computing. If you mean Windows and general computing, then yes it's worthwhile. For a long, long time hard drives have been the slowest component and increasing their speed would show the best increase in "responsivity", program loading, etc, assuming RAM and CPU were up to spec already. I'm looking forward to my full reinstall tomorrow quite eagerly. That was the primary reason for the purchase. Adding a lot of perceived smoothness to Arma is the cherry on top (and it does seem to be quite better subjectively, even if it's not so great an improvement objectively).

  13. Posted

    Get a faster/better CPU, otherwise you could reduce object/terrain quality or VD, that would take some of the load off of the CPU. I mean, 5KM VD is massive, especially with the higher object quality settings. Do note that you're not expected to be able to run Ultra settings (and 5km is ultra as it gets) in this game with hundreds of AI. Until the devs rewrite the engine to make AI run in parallel on multiple cores, you're stuck.

    As you increase the number of AI, you increase the amount of computation required on the main thread running on its single core, and this thread is the limiting factor for FPS for most players (and for you in this scenario, clearly). The longer this thread takes each frame, the slower the frame, because this is the bottleneck for your system. Reducing IQ or VD will help because this same thread also handles draw calls, which are decreased with decreasing scene geometry (lowered object/terrain IQ and VD).

    I don't know why you need more than 100 active AI, but if that's what you want, well, that's the deal. Drop your settings, drop the number of AI, or drop some money on a high-end overclocked processor.

  14. Posted

    The FPS are definitely smoother IF you're using high quality textures/models. If not, it doesn't do much. It depends heavily on your IQ settings and actual system if it's a must-have or not. Someone with a GTS 250 like I had before... not a must-have. I'm still not convinced it was worth $100 (only talking about Arma and not general computing) compared to how much the rest of my system cost - the $130 I spent on the 650 Ti-B yielded vastly superior improvements than the SSD, and I'd imagine a new i7 4xxx would likewise yield better than 3x the improvements of the SSD (at 3x the price). Even more RAM probably would, though that's a bit more of a gray area.

  15. Posted · Edited by DNK

    @Sniperwolf572 RE: Scenario and Processing

    For the original tests, I ran the following scenario;

    http://i482.photobucket.com/albums/rr181/davidk594/arma32014-02-2020-04-40-65_zps7dc9bfe1.png (365 kB)

    For the previous tests, I would run through this mission one time before benchmarking at the lowest IQ settings, then ran each after that in sequence without performing any flushes or program restarts. For the newest tests below, I ran through one time to preload everything, then one time again to benchmark (this is why the newest comparison test has a smaller difference, since both the HDD and SDD had already loaded almost all the textures to VRAM on the first run-through, so there wasn't much to do on the second run-through).

    The markers are the 7 different locations that the script setpos'd the player's chopper to every 7 seconds. The player enters the chopper as a passenger, then switches to 3rd person view, then hits zoom in. Then the chopper flies around Kavala. After each setpos, I would quickly pan 180-degrees and back. In Kavala, I kept the camera centered on the hospital, with a couple specifically timed occasions when I would do a quick 180-degree pan.

    For benchmarking, I used FRAPS' benchmark with frametimes selected. After finishing, I opened it in a spreadsheet and took the delta time for each frame (FRAPS sets them as cumulative, which is not very useful for a graph). Then I put the HDD and SSD data together and made a chart from them. For the "ordered" charts, I just copied the values, pasted them as numbers (not formulas), and used data->sort on each, then graphed them.

    Really, it must be another hardware issue, because the SSD offers no real FPS advantage I've seen.

    NEW TESTS

    Rig:

    (this has changed!)

    Intel i5 3350P 4c @3.3GHz (stock) 32-32-256-6M caches

    Gigabyte B75M mATX board (current BIOS)

    Gigabyte 650 Ti BOOST 2GB GDDR5 192-bit 900/1500MHz core/memory

    2x2 GB OCZ 1066MHz 7-7-7-16 DC DDR3 (moderate latency, low bandwidth)

    WD 7200RPM 1TB HDD SATA3

    vs

    Corsair Force LS 120GB SSD SATA3 (current firmware)

    FIFTH TEST

    This was a copy of the old scenario with the new GPU and new settings.

    Settings:

    HIGH + 2500VD

    Texture: very high

    Objects: high

    Terrain: high

    Shadow: ultra

    Particles: high

    Cloud: high

    PIP: off

    HDR: standard

    Dyn Lights: ultra

    VD: 2500/2500/200

    1600x900

    16xAF 8xAA "Low" HDAO ATOC all trees+grass, caustics on

    old_test_zps3b4c3ade.jpg

    Black is SSD, pink is HDD.

    Here, I preloaded all the textures/objects by running through the mission once before testing. As you can see, this reduces the SSD's edge over the HDD considerably, especially with a new 2GB VRAM video card (which never goes past 1.2-1.4GB of usage but whatever, it's more) to store it all in. Still, there's considerably less hanging after each sudden setpos movement for the SSD.

    In Kavala, the two are fairly similar, though the SSD has fewer lag spikes overall and what it does have are half to a quarter as bad as the HDD.

    Average FPS of SSD vs HDD: 31.6 v 31.2

    SIXTH TEST

    Settings:

    HIGH + 2500VD (same as above)

    Texture: very high

    Objects: high

    Terrain: high

    Shadow: ultra

    Particles: high

    Cloud: high

    PIP: off

    HDR: standard

    Dyn Lights: ultra

    VD: 2500/2500/200

    1600x900

    16xAF 8xAA "Low" HDAO ATOC all trees+grass, caustics on

    Scenario: player takes a 2x time speed chopper ride from Pyrgos, around the island, and drops in to Kavala, where he runs around a set course, quickly panning and zooming in constantly. This is meant to approximate a plane ride and typical infantry combat movements in a major city. This was done right after the prior test, so most textures/models should have been preloaded already. If that was not the case, I would expect the HDD to perform significantly worse as it did on tests 1-4.

    new_test_zps56449c4b.jpg

    Black is SSD, pink is HDD

    new_test_ordered_zps663f25c4.jpg

    Black is HDD, Blue is SSD

    First, other than a very few major lag spikes, neither drive has a clear "low end performance" problem here. Additionally, the HDD and SSD have very similar stutter performance while flying over the island - no clear winner there, though most textures/objects had been preloaded in both cases. However, the HDD clearly struggles a LOT more in terms of stutters in the last half, when the chopper gets to Kavala and the player gets out and runs around on foot. Here, the SSD clearly offers far smoother gameplay.

    Total frames above 40ms for SSD and HDD: 0.3% vs 0.6%.

    Worst frames for SSD (in ms): 104, 103, 83, 81, 74, 71, 63, 62, 61, 56...

    Worst frames for HDD (in ms): 456, 179, 104, 103, 102, 96, 90, 89, 89...

    The SSD's frequence of stutters is perhaps half as much as the HDD, and the severity also is roughly half. And this was in a test that was as biased in favor of the HDD as possible, with all textures being preloaded. We've seen how badly the HDD does when the bias is shifted in the SSD's favor, so it's only fair.

    Avg FPS of SSD vs HDD: 47.7 v 47.6

  18. Posted

    4GB is all I need for this game, and before people jump up and say, "no, it can utilize blahblahGB of RAM!" recognize that I watch my RAM usage after finishing, and I've yet to see total physical usage go past 3GB out of 3.95 available. The RAM is low bandwidth, but I'm fairly sure the VRAM streams direct from disk, so that shouldn't matter. The CPU is middle-of-the-road (but limited for Arma, yes) - for any other FPS it'd be more than adequate for 60FPS. The GPU has now been replaced by a 650 Ti Boost (yay!).

    I wouldn't expect many players to build a system just around this game, when so many other games prioritize the GPU over the CPU and more RAM over a faster hard drive. Now, I'm totally undoubtedly CPU limited with just a mediocre 650 Ti Boost, like 60-70% core usage. Given the lack of graphical polish, there's just nothing for the GPU to do but dawdle and wait for a better graphical game to get played. This isn't really news, though, and I wholly expected it.

    Anyway, I've started testing with the new GPU and much higher settings. Report coming soon. First thing I've noticed is that the GPU doesn't use more than like 1.2GB of RAM (it's 2GB total), meaning I have almost a full GB there too that's not being utilized, so instead the VRAM keeps going up and down as memory management does minor purges and loads. Ancient engine is ancient.

    I'll also do another test once I've moved the Windows and pagefile over to the SSD to see if that does anything.

  19. Posted

    ^ exactly. This is why I never fully trust FPS numbers from Arma benchmarks, especially when done just once. There's always like a +/-10%. It might actually be like 4% in the SSD's favor, hopefully further testing will get it more accurate.

    I have noticed that with all the reduced stuttering, I can increase my texture quality from low to high without a significant performance impact (due to said stutters), which is a huge IQ improvement since I would get huge hangs and long frames before on the WD with anything other than low (very high still hangs too much, but with another 1GB in the GPU it probably wouldn't, but right now there's too much memory management going on with just the 1GB).

    MP seems a lot smoother now, but that might just be my brain playing tricks, trying to justify the thing. Unfortunately with MP, there are so many more variables to control that it's damn-near impossible to do a benchmark.

  20. Posted · Edited by DNK

    Got a new SSD and figured why not do some benchmarking. I created a simple fairly script-less mission on Altis with 120 AI. To specifically test the SDD, I made the first part of the mission have the player setpos across the map to 8 different locations every 7 seconds, while I panned left and right after each move (the first pan right would load everything, the return pan left would look back over what was loaded). Then I had the player flown over Kavala for a couple minutes as I kept my view squarely on the hospital, with a couple quick pans out at specific locations. It should be easy to see on the graphs when the mission switched from the setpos script to the flyover.

    Rig:

    Intel i5 3350P 4c @3.3GHz (stock) 32-32-256-6M caches

    Gigabyte B75M mATX board (current BIOS)

    BFG GTS 250 1GB GDDR3 256-bit 738MHz/1100MHz core/memory (current drivers) (yes, this is a horrid GPU that's 12hours away from retirement)

    2x2 GB OCZ 1066MHz 7-7-7-16 DC DDR3 (moderate latency, low bandwidth)

    WD 7200RPM 1TB HDD SATA3

    vs

    Corsair Force LS 120GB SSD SATA3 (current firmware)

    Expectations:

    I expected minimal differences in FPS overall, with far fewer and lower stutters on the SSD (which would be the reason for whatever average FPS differences were seen). I expected the SSD would have far lower max frametimes than the HDD, and the SSD would do far better on the scripted setpos part of the mission with little to no pop-in.

    Results:

    Basically what was expected.

    In each graph, the green is the SSD and the red is the old magnospinner.

    The first graph is the frametimes, the inverse of FPS - higher is bad. 100 = 10FPS, 500 = 2FPS. The second graph is all the frames in order from fastest to slowest, again high = bad.

    FIRST TEST

    Settings:

    LOW + 2000VD

    Texture: low

    Objects: low

    Terrain: low

    Shadow: ultra

    Particles: low

    Cloud: low

    PIP: off

    HDR: low

    Dyn Lights: low

    VD: 2000/2000/200

    1600x900

    Nothing but full AF for the last tab

    low2000times_zps408f4542.jpg

    low2000ordered_zps172d5aa1.jpg

    In overall FPS neither is the clear winner; actually, the spinner looks a bit faster, but that's almost certainly variance at work. Already, though, the SSD is showing fewer stutters, especially in Kavala, but also far faster load-in for the setpos sudden moves. Still, overall they're basically the same.

    SECOND TEST

    Settings:

    LOW + 3500VD

    Texture: low

    Objects: low

    Terrain: low

    Shadow: ultra

    Particles: low

    Cloud: low

    PIP: off

    HDR: low

    Dyn Lights: low

    VD: 3500/3500/200

    1600x900

    Nothing but full AF for the last tab

    low3500times_zpsa4adc826.jpg

    low3500ordered_zpsf092943d.jpg

    Now, here's a stark difference when it comes to the setpos sudden mass loads. The SSD does a much better job of streaming in all the data quickly, with far less stuttering. There's still quite visible pop-in, but the SSD clears it within 1-2 frames each time, while the WD struggles frame after frame to stream in all that crap. In Kavala during the flyover, though, both drives are basically the same, with the SSD getting hit hard by a sudden turn in the middle.

    THIRD TEST

    Settings:

    High + 2000VD

    Texture: high

    Objects: high

    Terrain: high

    Shadow: ultra

    Particles: low

    Cloud: high

    PIP: off

    HDR: standard

    Dyn Lights: low

    VD: 2000/2000/200

    1600x900

    Nothing but full AF for the last tab

    high2000times_zps7ab2a874.jpg

    high2000ordered_zps9ea109ff.jpg

    Here, there's clearly a bigger difference in average FPS in Kavala (and the WD just fails at the setpos portion), though this may be random luck. While the SSD still loses a few frames occasionally, the WD is clearly constantly struggling to keep up, and it should show up in lowered average FPS by a smidge as well.

    FINAL TEST

    Settings:

    High + 3500VD

    Texture: high

    Objects: high

    Terrain: high

    Shadow: ultra

    Particles: low

    Cloud: high

    PIP: off

    HDR: standard

    Dyn Lights: low

    VD: 3500/3500/200

    1600x900

    Nothing but full AF for the last tab

    high3500times_zpsc5db391c.jpg

    high3500ordered_zps75e97379.jpg

    Here, there's not a clear overall FPS winner, but the SSD crushes the WD in the fast loading part. I'm going to go a bit more indepth, since this is where we should see the biggest difference. The longest frame for the WD is 540ms (less than 2FPS), with 15 being under 5FPS, very noticeable stutters, if not breaks in the game. By comparison, the SSD has only 3 frames under 5FPS (all just under that), and the 15th longest frame is 120ms (which is the same as the 40th longest for the WD).

    Looking solely at the Kavala flyover, ignoring the quickload tests, there's a clear favorite in the SSD:

    Time under FPS for SSD, for WD, ratio

    10FPS - 0.03% - 0.3% :: 10:1

    15FPS - 0.6% - 1.9% :: 3:1

    20FPS - 17% - 21% :: 1.25:1

    33FPS - 91% - 95% :: ~1:1

    AVG SSD = 42.9FPS

    AVG WD = 44.7FPS

    Conclusion

    Well, there's random luck at play. It's doubtful such a long, AI-heavy (for me) scenario would yield perfect results, and with just 1 run-through we're likely to get a noisy result. But it's clear that if there's any overall FPS gain from an SSD, it's miniscule. The real difference is in reducing stutters and stutter amplitude and in quick loading (the latter isn't very important usually, though). Going from having serious stutters 2% of the time to just a quarter as much is going to feel better, and having massive game-stopping stutters all but disappear is pretty huge since those happen all too frequently on the WD. However, I doubt it's worth the pricetag of an SSD just for that since the $100+ could be better spent on faster RAM (see this thread), saving up for a better CPU or GPU, or you know, hookers and blow.

    Still, if you were considering an SSD for other uses or just general computing, it probably will be worth it to put Arma on the drive.

  21. Posted

    You've got a 4.9GHz here, which the vast majority of players don't have. It also doesn't look like there's much of a difference between 1600 and 2133.

    I mean, the difference in latency between the fastest and slowest is a 38% reduction, which goes with a 17% reduction in avg frame times and 20% for minimum frames (meaning roughly a 50% ratio between RAM speed increases and FPS increases, which is still significant enough to justify a new purchase for players with old slow RAM).

    However, it seems like once you reached a certain threshold for your system (1600/C10) your minimum frame times didn't improve with anything faster, while your averages continued on improving. Now, if you have a much slower processor (most players), that plateau point for min FPS might be considerably lower, and the avg FPS increases might be less than here (50% of RAM speed increase) as well.

    So would a 3.9GHz or 2.9GHz require even 1600MHz RAM for best performance, or would 1333 or even 1066 work just fine for each, respectively?

×