• Acolyte of the Butterfly
    Acolyte of the Butterfly
    Posts: 109 from 2013/12/5
    Oh boy...


    I understand from your previous post(s) that AGP bandwidth may be limiting. As you stated, that seems a bit odd

    The game is software-rendered and the final bitmap must eventually be converted and pushed to VRAM every frame and then scaled up. That's why I chose TinyGL for that task, to offload the upscaling to the GPU to minimize bandwidth usage. On AOS4 I chose Compositing, on AROS MesaGL.


    I went back and read Trevor's blog and he doesn't say what resolution he is using so i assume the default windowed mode.

    Exactly, windowed, default resolution, dedicated save-game for beta-testers, mouse-cursor moved to the lower-right corner to have exactly the same conditions on ALL (! ALL !) platforms.


    At least I get "performance warnings" (bitmap doesn't have BMF_3DTARGET set)

    The surface and texture used are managed by TinyGL. It should know best what bitmap attributes to set and which not.


    Yep, that's a clever trick by Trevor. Tower57 runs faster in windowed mode than on fullscreen (that's why the FPS counter on his page favours OS4 over MorphOS).

    All such benchmarks should have been made under the same conditions mentioned above. And IMHO Trevor is not childish enough to cheat.


    In fullscreen mode FPS counter is stable at around 60 (I assume the game forces VSYNC in the fullscreen mode).

    Yes. Which is also why I told everybody who made performance measurements to use the default window mode.


    There are, which Daytona675x prepared and used for his benchmarks and obviously shared with Trevor.

    Yes, they were shared with all beta-testers, Trevor happened to be one of them.


    Where can I download it? Link?

    Nowhere, at least I didn't preserve them. And it doesn't matter, you can take any save-game you like and use that for comparison. The important part is that it's window-mode, the same save-game and the same settings on all devices. And that everything else is identical too (e.g. excactly the same scrolling position). Of course your own save-game will probably have its emphasis on other aspects of the game as the ones I used and thus lead to very different results. But the general "system X is around y-times as fast as system Z" should usually be the same. However, Trevor's benchmarks are old. It wasn't the final version 1 back then, IIRC.


    It depends what's going on in the game, the FPS varies wildly.



    I wonder if TinyGL is even being used on MorphOS. The moment the game is

    It is. I don't know how many times I stated in public how things work and why...


    The moment the game is launched my CPU load jumps to 100% and stays there. 100% of cpu load at 2.7Ghz?

    Yes. I need all performance I can get. As a reminder: the original PC version drops down to 15 fps in some areas on my Sony Vaio (and can't even keep steady 60 fps on my i7 all the time)... I optimized the absolute *hell* out of this, but yes, I need all the performance your antique system can deliver, sorry.


    The problem is the bizarre "benchmark" of it to promote OS4 superiority.

    You are surmising something that isn't there.


    The game happens to have been ported by someone who has also been contracted by A-eon to work on their 3D graphics stack.

    I am not involved into Warp3D Nova (in fact I was one of its biggest critics back then, but since then it became much better). I was hired by A-eon to implement OpenGL ES 2 and occasionally extend it for some $. And that's it with my relationship with them. I'm not even part of their development team.


    As was just announced, they will fly him out for Amiwest all the way from Europe even.

    No need for conspiracy or other wild theories. As so often with normal people, things happen because of pretty simple reasons. In this case here the situation is simply that Trevor's and my contact was rather close during T57-dev, because he was a very active tester (guess what, he's one of the guys who helped testing the MOS version too). Simply because of kindness and likely because one came to know and respect the other, he asked me some weeks ago if I'd like to come over to AmiWest. I said yes, which resulted in him bringing my name on the table of the AmiWest comitee. And those guys, not aeon, decided to select me this year, thanks. For your notes: AmiWest != Aeon. And AmiWest != AOS4. Yes, believe it or not: not everybody is an bigoted militant Amiga flavour fanboy spreading FUD or other crap.


    I do not think it is far fetched to assume that the version for Hyperion Entertainment's OS is the most optimized one compared to the AROS and MorphOS ports.

    That's probably the most insane acusation I ever heard in Amigaland. Complete nonsense, of course. Last not least because I actually spent most time with the MOS version, thanks to the rather limited joypad support through sensors.library.
    You know, pal, in contrast to you and some other guys here I'm known for *not* being a fanboy of one particular Amiga flavour, that's why I even supported AROS, although less than a handful of people actually use that version of the game (and I knew that it would be this way beforehand).
    It's really extremely far fetched that I should prefer one version over the other / optimize one more than the other.
    Besides that it's also technical nonsense: the game-logic incl. the software renderer is practically identical on all Amiga systems (AROS differs a tiny bit, because I didn't use asm there at all).


    Of course, if you want to gauge the credibility of the published benchmark numbers, all you need to remember is that results on AROS were not included...

    No need to come up with yet another pointless conspiracy theory. In case of Trevor the simple explanation is: he had no AROS system at hand, guess what... Yes, the simple truth can be so boring.
    And in my case: I published lots of my own benchmarks at different places in public, incl. those of my own AROS system.
    Really, man. Stop spreading bullshit, thanks.


    Also, minimum requirements: MorphOS >= 3.9, >= 600 MHz, Gfx-card with TinyGL support. Something is off the road here :)

    That should be enough to be playable using the 16bit version. Even my MOS G4 733 can handle that good enough. I suppose you read the readme etc. If nothing helps: bad luck, sorry.
    And before somebody asks again: no, there is no demo version, no, I'm not the rights holder, no, I'm not involved in sales.


    The original game uses hardware rendering but needs 1 GiB or so VRAM. That's the reason Daytona675x gave for switching to software rendering.

    Not that much, but definitely too much for my lowest end targets, yes :) The whole gfx system was designed in a way that made it impossible to use hw-acceleration on most systems (exception: W3D Nova and maybe some fat AROS systems). And even if: my software-renderer runs faster than the original OpenGL-renderer on Windows... And the quality is higher (the lighting calculations are much more exact because, well, they are calculated and not based on a too-small fixed light-sphere-texture).


    Okay, but then why port such a hungry and obviously bad programmed game in the first place?

    Because I did not know that it would be coded so sub-optimal when I said "yes" to port it some year before. I thought it would be pretty optimal code. I was wrong. But instead of simply canceling that project I spend half a fucking year optimizing / rewriting crucial parts of it.


    I blame the concept of advertising and selling a game for "Amiga NG" which then does barely work on the fastest machines available

    It runs (too) fast on fast Amiga machines. Actually faster than on the many PCs :)
    But yes, I'd also have liked it more if it had been fast enough in the beginning and if it had been the "some weekends"-job I though it would be :P


    Ok, the term "only working" would have been more correct.

    Actually I'm pretty happy that it's well playable even on what I considered my min. specs. And regarding the specs: I actually managed to LOWER them compared to my estimation during the Kickstarter campaign when I didn't know the source yet :)


    In any case, scrolling is smoother (!) on G4 1,67Ghz MorphOS 9800XT 256MB ram 2xAGP than on a Windows7 Core2Duo@2.3Ghz 4GB Ram, and 9700GT 512MB, which is remarkebly impressive (ok, on MorphOS I run it on 1280x1024 but nevertheless Daytona's work is amazing!).

    Thanks :) One reason why the Amiga version is *really* smoother is because it is.
    The original game's logic is not decoupled from the physical fps, so if the frame-rate drops below 30 then internal calculations go wrong, interpolations are calculated falsely (incl. scrolling interpolation), shots start going through enemies, etc. and everywhere.
    The Amiga version is decoupled from the fps.
    Which in turn means that the Amiga version has to do much more work than the PC version per frame (unless everything runs fast enough; so low-end systems are the ones with the extra burden). Plus software-rendering.
    Btw.: besides all the stuff I wrote on the web in the past regarding technical aspects / issues, the game itself contains an easter-egg, namely an not-so-hard-to-find (incomplete) list of some of the optimizations / changes I made with the biggest impact.

    Anyway, enough.
    Cheerio! And to all the conspiracy-theory guys: get a life.

    [ Editiert durch Daytona675x 06.09.2018 - 19:08 ]
  • »06.09.18 - 17:45
    Profile Visit Website