Multi-Core Optimization and Nvidia SLI/AMD Crossfire Support

[Versus-PC_Official]

I know this topic has been beaten over and over again, however, i haven't seen it seriously talked about in a while, and with the massive progression in the computer industry over the last couple of years, i really believe this game could benefit from a small "overhaul" so to speak. I know it wouldn't be easy, and would require a handful of work from JLV, but from a Desktop Builder's perspective, these updates would allow a much more broad range of systems to be able to run MXS, and possibly allow cheaper systems to achieve the same results as more costly systems are currently achieving.

For example, you can get two GTX 970's for about the same price as a GTX980Ti, typically a bit less, and in most new games, the two GTX 970's in sli would outperform the Single GTX 980Ti. Plus, you would effectively be getting an extra 2GB of VRAM with the SLi setup as well.

On the other side of things, as displays progress, SLI setups will be mandatory to achieve 144Hz + in 4K with desirable graphics settings, so it would be a bummer to be limited by the game itself.

On the CPU topic, i feel like MULTI-CORE optimization is key, considering the number of high core, multi thredead, Performance CPU's are on the market. Hyperthreading/ Nvidia Threaded optimization already causes some issues, like stuttering, freezing, and FPS drops, and attempting to stream with OBS can be a nightmare if your system isn't VERY powerful.

So, long story short, JLV, how many organs does the community have to sell on the black market to bribe you for a little (kinda big) update.

Oh, and the icing on the cake would be some form of active shadow below the bike, so they dont look like theyre "floating"

Ps. Please don't take this as a complaint, or demand, as the game is awesome as it is, and i am beyond appreciative of everything you and the community have done to keep it progressing forward.

[Ktmrdr11]

Id gladly sell my organs for this game.

[Wahlamt]

I think we should have multi core support before optimization. As far as I know the game only uses 1 cpu core.

[Versus-PC_Official]

I think we should have multi core support before optimization. As far as I know the game only uses 1 cpu core.

I guess multi-core support was implied with multi-core optimization. It is currently single core only.

[jlv]

I'm not a fan of alternate frame SLI setups. They do a terrible job of supporting textures generated on the GPU which is critical for the terrain texturing. The worst part of alternate frame rendering is you get an extra frame of latency. Unless it doubles the frame rate it's actually getting a worse response time compared to a single card.

Not sure if I'm misunderstanding you on the VRAM thing, but the textures are duplicated on each card, so 2 cards with 2 GB each can still only hold 2 GB of textures.

As far as multithreading, there are some things that I could parallelize but the biggest bottleneck is I generate too much traffic on the graphics bus. I need redo the terrain tessellator so it keeps more terrain data on the GPU instead of resending it every frame. It doesn't sound as sexy as multicore but it'd make a bigger difference.

[Versus-PC_Official]

I'm not a fan of alternate frame SLI setups. They do a terrible job of supporting textures generated on the GPU which is critical for the terrain texturing. The worst part of alternate frame rendering is you get an extra frame of latency. Unless it doubles the frame rate it's actually getting a worse response time compared to a single card.

Not sure if I'm misunderstanding you on the VRAM thing, but the textures are duplicated on each card, so 2 cards with 2 GB each can still only hold 2 GB of textures.

As far as multithreading, there are some things that I could parallelize but the biggest bottleneck is I generate too much traffic on the graphics bus. I need redo the terrain tessellator so it keeps more terrain data on the GPU instead of resending it every frame. It doesn't sound as sexy as multicore but it'd make a bigger difference.

Understandable regarding the SLI situation, i was under the impression there would be a way to optimize to where it would deliver better results, however from the data you gave it definitely would not. Only real hands on experience with SLI setups i have is on the Overclocking and Benchmarking side of things, no actual in game use.

And that makes sense regarding the CPU/GPU usage, as this game is definitely heavy on the cpu! It would be nice to take advantage of all of the high performance cards out on the market nowadays.

[Versus-PC_Official]

I'm not a fan of alternate frame SLI setups. They do a terrible job of supporting textures generated on the GPU which is critical for the terrain texturing. The worst part of alternate frame rendering is you get an extra frame of latency. Unless it doubles the frame rate it's actually getting a worse response time compared to a single card.

Not sure if I'm misunderstanding you on the VRAM thing, but the textures are duplicated on each card, so 2 cards with 2 GB each can still only hold 2 GB of textures.

As far as multithreading, there are some things that I could parallelize but the biggest bottleneck is I generate too much traffic on the graphics bus. I need redo the terrain tessellator so it keeps more terrain data on the GPU instead of resending it every frame. It doesn't sound as sexy as multicore but it'd make a bigger difference.

Understandable regarding the SLI situation, i was under the impression there would be a way to optimize to where it would deliver better results, however from the data you gave it definitely would not. Only real hands on experience with SLI setups i have is on the Overclocking and Benchmarking side of things, no actual in game use.

And that makes sense regarding the CPU/GPU usage, as this game is definitely heavy on the cpu! It would be nice to take advantage of all of the high performance cards out on the market nowadays.

[Versus-PC_Official]

Also, i could be incorrect about this, but with alternate frame rendering, while there is an additional frame of latency introduced, i don't believe the sli setup would need to be rendering twice as fast to gain any advantage.

If we look at the mimimum-maximum range of frame times, assuming the single card is running at 30fps and the sli setup is running at 60fps you have this:
Single Card: 1-2 frames (33.33ms) of delay at 30FPS = 33.33ms-66.66ms
SLI Setup: 2-3 frames (16.67ms) of delay at 60FPS = 33.33ms-50.00ms
SLI has the same minimum latency, a lower maximum latency, and it's delivering 60fps over 30fps. But pertinently, there's no extra latency in the end being added here.

Now assuming the single card is capable of the same speed as a sli setup, in that situation there is latency added because of the way it is processed, assuming my math is correct here.
1-2 frames (16.67ms) of delay at 60FPS = 16.66ms-33.33ms
2-3 frames (16.67ms) of delay at 60FPS = 33.33ms-50.00ms

[Versus-PC_Official]

Oh, last little thing i forgot to include in the above post, the latencies i used up there were assuming best and worst case for each card setup ie. the frame being done rendering just prior to polling, or just post polling. Where the single card can have 1-2 frames of latency because of that, where sli can only have 1-1.5 frames of latency, as the second card will start rendering the next frame half way through the first if the first card finishes rendering the frame just after the polling.

This makes way more sense in my head than it does in the damn English language, so my apologies if i am making zero sense.

[jlv]

That's a good point on the latency. The best case latency is the time to render one frame, but the worst case is the time to render one frame plus the inverse of the total frame rate. So 2 cards capable of 45 FPS each and 90 FPS together would get the same worst case latency (1/45 + 1/90) as a single card running at 60 FPS (1/60 + 1/60). To get the same average latency, you'd need two 50 FPS cards running together ((1/50 + 1/50 + 1/100) / 2 = (1/60 + 1/60 + 1/60) / 2). So not quite the double that I incorrectly stated before.

[Versus-PC_Official]

That's a good point on the latency. The best case latency is the time to render one frame, but the worst case is the time to render one frame plus the inverse of the total frame rate. So 2 cards capable of 45 FPS each and 90 FPS together would get the same worst case latency (1/45 + 1/90) as a single card running at 60 FPS (1/60 + 1/60). To get the same average latency, you'd need two 50 FPS cards running together ((1/50 + 1/50 + 1/100) / 2 = (1/60 + 1/60 + 1/60) / 2). So not quite the double that I incorrectly stated before.

Right, i genuinely wasn't trying to be an ass there, i was just trying to understand. And looking at your math compared to some benchmarks, it looks like that second card is enough of a bonus to be able to grab a very small advantage. I agree that SLI is NEVER the best setup, however in some situations, price to performance in terms of overall FPS it can be hard to pass up.

Id much rather see what you were talking about previously regarding more of the processing happening on the gpu however, as it sounds like the SLI benefits might not be worth the effort.

Just another question, how difficult would doing a live shadow below the bike be, could it be as rudimentary as some form of png with transparency under the bike, or am i way off the mark here?

[Versus-PC_Official]

I really need to remember i can't edit my posts here, and stop posting before i get all of my thoughts out. So assuming real life is neither best nor worst case scenario, but a mix of the two, it actually looks like there may be a substantial benefit.

And what i was referring to with VRAM was even the most base GTX900 series gpu's have 4gb vram, opposed to the 2gb as previous generation standard.

[jlv]

For shadows I think I'd go straight to shadow buffers.

[Versus-PC_Official]

For shadows I think I'd go straight to shadow buffers.

So, back to my original question, what kind of bribe would it take to get a little refresh for current gen systems, and maybe take a little advantage of performance available by current hardware, and add a couple of graphics options like shadows.

For example i hate to bring it up, because the riding/physics aren't even in the same league of MXS, however i do feel like MX Bikes in some ways is headed the right direction for the future, in terms of graphics capabilities, as well as things like terrain deformation.

Again, thanks for all of the work you've done for FREE, this was the best $40 I've probably ever spent.