Why do people pay to be at the bleeding edge for a non-esports gaming computer?

[Me1z]

2 minutes ago, Paradine Sage said:

Blame the idiots who can't (or are unwilling to) code properly.

 

Good code written in this day and age damn well should take full advantage of however many cores are present. But people are willing to blame intel for their FPS lacking when they should be blaming the game maker.

 

There is near zero acceptable excuse for a 6 core not outperforming a 4 core at the same clock speed.

 

---

And you can multi-thread physics calculations in a number of ways, it just requires more effort in programming and most developers are quite lazy.

 

If you want evidence of that second point just look at how few implement Vulcan or Dx12 at anywhere near their capabilities.

Initially I thought the same way, BUT there are very good reasons why multi-threading isn't even attempted. Firstly, we are operating under the assumption that physucs engines make approximations to allow for user friendly gameplay, no one wants to wait 2 days to know if the bullet flying towards their space station ricocheted or punctured the armor. With the way that physics engines currently work, the closest you could get to multi-threading would be to assign each core to an individual entity's calculations, but what happens when these entities collide? Obvious choice would be for one entity's collision to take priority as the two communicate with each other, but when the durability of a single piece of sheet metal can decide the victor in an altercation, it is necessary for the outcome to be fair to both entities. That means you would need to assign a 3rd thread just to mediate between the other two. Then you're still limited to the single-threaded performance of the core the 'mediator calculations' were assigned to. Physics can't really be divided, because they are very long, complex equations that can only be worked on by a single process at once. Eventually, No matter what, you get to a point where calculations are limited by single threaded performance.

 

14 minutes ago, Paradine Sage said:

"I desperately need ave"age single-threaded CPU performance to increase by about 400% yesterday, and yet we're stuck with these garbage improvements in increments of 10-15% "

 

He's bitching that performance on intel's end has only ~10% in eighteen months. His reason for bitching is that he can't get the performance that he wants out of some applications and yet instead of bitching at those coding those applications he is bitching at the people who have had hardware on the market for years that can do what he wants.

 

So long as consumers keep blaming the hardware manufactures for performance issues the application producers have very little reason to take the time and effort to code their applications correctly, and I am very sad to say that it is far harder to keep improving hardware than it is to keep improving software.

You don't understand how physics calculations work do you? Division of calculations DOESN'T FUCKING WORK in the the time frame required for a playable videogame. It's not lazy programmers, its not more money than brains, its just LOGIC. Physics can't be split into multiple parallel processes in real-time. You can only divide on the axis OF time in the sense that a 4 minute simulation can be rendered in four 1 minute chunks. If you truly think that you've magically found a way to make this all work, then PLEASE go enlighten the developers at Keen, because I'm sure they'd love to hear it.

[Paradine Sage]

3 minutes ago, Me1z said:

Initially I thought the same way, BUT there are very good reasons why multi-threading isn't even attempted. Firstly, we are operating under the assumption that physucs engines make approximations to allow for user friendly gameplay, no one wants to wait 2 days to know if the bullet flying towards their space station ricocheted or punctured the armor. With the way that physics engines currently work, the closest you could get to multi-threading would be to assign each core to an individual entity's calculations, but what happens when these entities collide? Obvious choice would be for one entity's collision to take priority as the two communicate with each other, but when the durability of a single piece of sheet metal can decide the victor in an altercation, it is necessary for the outcome to be fair to both entities. That means you would need to assign a 3rd thread just to mediate between the other two. Then you're still limited to the single-threaded performance of the core the 'mediator calculations' were assigned to. Physics can't really be divided, because they are very long, complex equations that can only be worked on by a single process at once. Eventually, No matter what, you get to a point where calculations are limited by single threaded performance.

 

You don't understand how physics calculations work do you? Division of calculations DOESN'T FUCKING WORK in the the time frame required for a playable videogame. It's not lazy programmers, its not more money than brains, its just LOGIC. Physics can't be split into multiple parallel processes in real-time. You can only divide on the axis OF time in the sense that a 4 minute simulation can be rendered in four 1 minute chunks. If you truly think that you've magically found a way to make this all work, then PLEASE go enlighten the developers at Keen, because I'm sure they'd love to hear it.

To what degree do you want to model the universe?

 

The easiest method of multi-threading physics is to start with gross generalizations and then refine. For example, a human target gets shot by a shooter while standing in a cluttered room.

 

You have one thread figuring out physics with very bad accuracy and at a very simple level. Use single cubes as hit boxes, for example. This thread is constantly figuring out what the shooter (and any other object with a self-motivation flag) could do. If someone fires a gun then they can't hit what is directly behind them, or 90 degrees off to the right or left. This is also the controlling thread that manages the other processes.

 

Now anything that could be a target starts running it's own modeling assuming that it is the target, ceasing to do so but off loading any partially completed calculations to the cache (or RAM, depending on the type of game and the specifics). The gun object is constantly running the self updating physics calculation for if it is fired at that moment assuming no interference in flight by another object and as it moves through the world it offloads its XYZ coordinates and orientation along with the bullets arrival XYZ coordinates to RAM.

 

So the shooter pulls the trigger and, assuming that the gun has already been in that location and orientation at any time since the application started, the physics calculation for target coordinate isn't performed and instead that location is pulled out of the cached calculations and given to the room process. The room process then finds every object with a hit box that includes the target coordinates or any coordinates in a straight line path between the firing coordinates and the target coordinates and tells each one of them to update their physics calculations with the new information. Every other object around those starting targets is split off into its own separate thread by the room process (grouped assuming fewer available threads than objects) and set to figuring out what happens to it assuming that the given starting target is hit.

 

Once any starting target process calculates that it will be hit by the bullet, the time and coordinates of it being hit are served to the controlling room process and it starts ending every process that shows a "i'm hit" time later than another process, also canceling any process that assumed those potential targets were struck. Once the actual target is identified the room object gets back to work while the target's process continues to model the effects of the bullet on it.

 

The room object now figures out the maximum possible effect box of the "target" and tells everything within that area to start modeling what it would potentially effect if it was hit.

 

Once the "target" process finishes figuring out it's response to being struck that is fed to the room process to be shared out to every one of the potentially effected objects around it. Each of which then calculates whether or not it will be effected. If the object is effected then the whole loop starts over again. The chain ending once you reach an object that doesn't impact any other object.

 

In terms of total processing power you are wasting shit tons because you will just discard as irrelevant upwards of 90% of the physics calculations that you have performed but your real world performance shoots through the roof because you are simultaneously executing upwards of five times as many cycles per second.

 

One core at 5 Ghz is five billion cycles per second. A 6950x at 4 Ghz is 40 billion cycles per second. Or eight times as many. Does that mean that you will be performing calculations eight times as fast? Of course not. Assume that 90% of those excess cycles are totally wasted, that is still effectively an additional 3.5 GHz worth of single core performance, or upwards of 50% improvement. Just a 7700k going from one to four cores assuming 10% effective is equal to 1.5 GHz extra.

 

And you can tighten up the code to the point where it is much less than 90% wasted.

 

---

Then you have the fact that complete physics modeling is utterly wasted on video games played by humans. Our reaction times aren't good enough, our eyesight isn't good enough, our monitors aren't good enough, and our brains aren't good enough.

 

If you want to accurately model something like the impact of weather on the stealth characteristics of an F-35 then yeah, you need molecular level or better accuracy in your model. But if you want to, say, "accurately" model a dog fight between F-35's in a flight simulator then you can do about 95% of the work using meter level accuracy and then rest with accuracy in the ten centimeter or so range. In terms of player experience the errors will be totally and completely unnoticeable.

 

Honestly you would gain better realism results by bothering to program in accurate materials science figures and modeling (which again you can do at a level of detail that is grossly inadequate for professional engineering use). I've yet to see any game come even close to accurately portraying the impact of objects on one another.

 

---

The above is just one of a dozen or so different methods you can use to multi-thread physics calculations for video games.

 

And this is without getting into what you can and can't offload onto the graphics card, and to what extent you can offload onto any given GPU. 

[Me1z]

16 minutes ago, Paradine Sage said:

To what degree do you want to model the universe?

 

The easiest method of multi-threading physics is to start with gross generalizations and then refine. For example, a human target gets shot by a shooter while standing in a cluttered room.

 

You have one thread figuring out physics with very bad accuracy and at a very simple level. Use single cubes as hit boxes, for example. This thread is constantly figuring out what the shooter (and any other object with a self-motivation flag) could do. If someone fires a gun then they can't hit what is directly behind them, or 90 degrees off to the right or left. This is also the controlling thread that manages the other processes.

 

Now anything that could be a target starts running it's own modeling assuming that it is the target, ceasing to do so but off loading any partially completed calculations to the cache (or RAM, depending on the type of game and the specifics). The gun object is constantly running the self updating physics calculation for if it is fired at that moment assuming no interference in flight by another object and as it moves through the world it offloads its XYZ coordinates and orientation along with the bullets arrival XYZ coordinates to RAM.

 

So the shooter pulls the trigger and, assuming that the gun has already been in that location and orientation at any time since the application started, the physics calculation for target coordinate isn't performed and instead that location is pulled out of the cached calculations and given to the room process. The room process then finds every object with a hit box that includes the target coordinates or any coordinates in a straight line path between the firing coordinates and the target coordinates and tells each one of them to update their physics calculations with the new information. Every other object around those starting targets is split off into its own separate thread by the room process (grouped assuming fewer available threads than objects) and set to figuring out what happens to it assuming that the given starting target is hit.

 

Once any starting target process calculates that it will be hit by the bullet, the time and coordinates of it being hit are served to the controlling room process and it starts ending every process that shows a "i'm hit" time later than another process, also canceling any process that assumed those potential targets were struck. Once the actual target is identified the room object gets back to work while the target's process continues to model the effects of the bullet on it.

 

The room object now figures out the maximum possible effect box of the "target" and tells everything within that area to start modeling what it would potentially effect if it was hit.

 

Once the "target" process finishes figuring out it's response to being struck that is fed to the room process to be shared out to every one of the potentially effected objects around it. Each of which then calculates whether or not it will be effected. If the object is effected then the whole loop starts over again. The chain ending once you reach an object that doesn't impact any other object.

 

In terms of total processing power you are wasting shit tons because you will just discard as irrelevant upwards of 90% of the physics calculations that you have performed but your real world performance shoots through the roof because you are simultaneously executing upwards of five times as many cycles per second.

 

One core at 5 Ghz is five billion cycles per second. A 6950x at 4 Ghz is 40 billion cycles per second. Or eight times as many. Does that mean that you will be performing calculations eight times as fast? Of course not. Assume that 90% of those excess cycles are totally wasted, that is still effectively an additional 3.5 GHz worth of single core performance, or upwards of 50% improvement. Just a 7700k going from one to four cores assuming 10% effective is equal to 1.5 GHz extra.

 

And you can tighten up the code to the point where it is much less than 90% wasted.

 

---

Then you have the fact that complete physics modeling is utterly wasted on video games played by humans. Our reaction times aren't good enough, our eyesight isn't good enough, our monitors aren't good enough, and our brains aren't good enough.

 

If you want to accurately model something like the impact of weather on the stealth characteristics of an F-35 then yeah, you need molecular level or better accuracy in your model. But if you want to, say, "accurately" model a dog fight between F-35's in a flight simulator then you can do about 95% of the work using meter level accuracy and then rest with accuracy in the ten centimeter or so range. In terms of player experience the errors will be totally and completely unnoticeable.

 

Honestly you would gain better realism results by bothering to program in accurate materials science figures and modeling (which again you can do at a level of detail that is grossly inadequate for professional engineering use). I've yet to see any game come even close to accurately portraying the impact of objects on one another.

 

---

The above is just one of a dozen or so different methods you can use to multi-thread physics calculations for video games.

 

And this is without getting into what you can and can't offload onto the graphics card, and to what extent you can offload onto any given GPU. 

On your point about complete physics modeling being wasted on video games played by humans, I completely agree. In the specific case of Space Engineers, where each object is actually constructed of hundreds of thousands of sub-objects, which upon becoming a target of damage and which angle it will receive damage from, must then dynamically crush inwards or collapse, multiplied by how many sub-objects are simultaneously colliding, while also calculating altered trajectory and momentum as a result of each damaged object, and transmitting the information to each of the players, such accuracy would dramatically cripple performance and add little to the game. This also isn't a situation where the gun is likely to have ever been in that location or orientation at any prior point, each collision must be calculated from scratch. I think what you're getting at would be suitable for your typical video game, and if with modern hardware, what constitutes as your average video game is having performance issues, then yes you are a lazy programmer. The performance issues with this game aren't because each individual object collision is performance heavy, but because there are too many simultaneous sub-object collisions which each need to be calculated before moving onto the next group of sub-objects. This is because the nature of the game is to allow individuals to dynamically construct and destruct entities, which means each object is made of smaller objects. Most games have mid-range physics like you were describing, but this game has both micro and macro scale physics. Could you slightly improve performance with your method of having parallel predictive calculations? Yes, you could, but I don't think it would be by enough be considered a solution. There's still a point where the inital object's collision calculations are holding up the rest of the objects collision calculations beause you can't infinitely predict. If your definition of lazy is not programming in multi-threaded collision prediction, then sure they're lazy, I'll concede that to you.

[Paradine Sage]

45 minutes ago, Me1z said:

On your point about complete physics modeling being wasted on video games played by humans, I completely agree. In the specific case of Space Engineers, where each object is actually constructed of hundreds of thousands of sub-objects, which upon becoming a target of damage and which angle it will receive damage from, must then dynamically crush inwards or collapse, multiplied by how many sub-objects are simultaneously colliding, while also calculating altered trajectory and momentum as a result of each damaged object, and transmitting the information to each of the players, such accuracy would dramatically cripple performance and add little to the game. This also isn't a situation where the gun is likely to have ever been in that location or orientation at any prior point, each collision must be calculated from scratch. I think what you're getting at would be suitable for your typical video game, and if with modern hardware, what constitutes as your average video game is having performance issues, then yes you are a lazy programmer. The performance issues with this game aren't because each individual object collision is performance heavy, but because there are too many simultaneous sub-object collisions which each need to be calculated before moving onto the next group of sub-objects. This is because the nature of the game is to allow individuals to dynamically construct and destruct entities, which means each object is made of smaller objects. Most games have mid-range physics like you were describing, but this game has both micro and macro scale physics. Could you slightly improve performance with your method of having parallel predictive calculations? Yes, you could, but I don't think it would be by enough be considered a solution. There's still a point where the inital object's collision calculations are holding up the rest of the objects collision calculations beause you can't infinitely predict. If your definition of lazy is not programming in multi-threaded collision prediction, then sure they're lazy, I'll concede that to you.

Why in the world are they calculating the effects of damage by figuring it for each individual sub-object and their interactions with one another?

 

You figure out the bonding energy of each individual sub-object to one another when the thing is built in the first place and then make one single object. Then you model the impact event on the single whole object and include energy propagation throughout the object. Only split out the objects if the bonding energy is overcome and start calculating for those from there.

 

That would honestly give a much more accurate representation of reality than treating each sub-object individually actually.

 

Then you pick a sub object size that allows a level of physics accuracy that is "perfect" to the naked eye on modern hardware and push off everything smaller than that to the GPU to render.

 

You also need to figure in scaling. Given the title, I'm assuming that this involves macro scale construction is space. When distances are measured in tens of thousands of kilometers you can get away with very large sub-object's and only need to model with higher accuracy if the player zooms in

 

Then there is off-loading physics to the cloud to some extent. Every time physics calculations are done you upload them to a cloud database and then, whenever a "new" object is created you search the cloud database for that particular collection of sub-objects and download it to local storage so that you can negate calculations as much as possible. Although honestly I would be pre-calculating as much of this stuff as possible with the GPU. The issue with GPU physics calculating is the sync time between the GPU and CPU but you can take the finished object, offload the information to the GPU, and then have it calculate out the physics as much as possible during down time before dumping the finished process back onto the CPU and into memory for later use. You can also leverage the distributed processing power of your entire player base by essentially making folding@home for your physics problems to again just start modeling objects when cores are idle for your players and upload the results to your cloud to be pulled down by your players as needed.

 

----

But then the vast majority of commercial software sucks coding wise and is horrendously inefficient at using the available hardware. I can't wait until AI optimization makes it out of the lab and into commercial use. It will be so very nice when people can upload their code to the AI and it will rewrite it to do the exact same thing except with the highest possible optimization, even if that will make programmers even lazier. Sigh, ten more years or so.

[Trik'Stari]

11 hours ago, tmcclelland455 said:

You can run all of those on something like a 2600k or 2700k just fine lmao

Heavily modded Fallout 4 can be rather CPU intensive. Especially if you're like me and have 90+ mods, then drop The War of The Commonwealth that adds tens of thousands of spawn points to the game.

[Yoinkerman]

because hobbies

[Teddy07]

I want the newest PC tech but at the same time i know it´s stupid so i have to restrict myself. Still running my i7 4790k for some years.

[Sauron]

To be honest it doesn't matter in esports either. Most competitive games run very well on potatoes and professional players often use very low graphics settings to reduce on screen clutter.

 

While an i7 is not worth the premium over an i5 for games, if you already bought the best of the best for everything else you might as well - there is a difference and the premium is not that high compared to what you'd have to pay to go from a 1070 to a 1080 (for example).

[XenosTech]

30 minutes ago, Sauron said:

To be honest it doesn't matter in esports either. Most competitive games run very well on potatoes and professional players often use very low graphics settings to reduce on screen clutter.

 

While an i7 is not worth the premium over an i5 for games, if you already bought the best of the best for everything else you might as well - there is a difference and the premium is not that high compared to what you'd have to pay to go from a 1070 to a 1080 (for example).

I do this in CS: GO and every one I know that plays CS: GO plays it at the lowest possible settings and then cap the fps at around 70 or so. I noticed a night and day difference in my game play but due to hackers and shit that game is dying pretty fast now  

[Sir Asvald]

Well for starters, most of the streamers spend thousands upon thousands on a PC just to play one game. Now that is stupid. For example, as other people have mentioned, CS:GO is one the game that can be played on a potato. Yet they spend £2000+. This is ridiculous. Yet companies get away with it. 

 

I know it's stupid. But as they don't understand computer hardware, it does bother us. For example, I was called to build a PC for a family friend. They were willing to spend around £800-900. The problem was that he only plays one game, which is...... Minecraft. A game that can and could be played on a 5 year old PC. Yet they want to spend £900. I refused said it's a waste of money. £450 will do more than enough. 

 

Sometimes being in IT can be very annoying.

[LordXenu]

Why do people complain about what others do?

[flibberdipper]

11 hours ago, glunday said:

theres a difference in just running the game and running it with i textures and high fps well

You can do that on the 2500/2600K... Even though they're quite a few generations old, Intel hasn't done jack shit since they don't have any competition to push them.

[ZimFreak]

Not everyone is upgrading from last year's gen. Most people skip a gen or two between upgrades. And a lot of those i7-6700k vs i7-7700k threads are people who are:

A) just curious about how good the new thing is

or B) Trying to decide if it's worth it to buy the brand new thing vs a heavily discounted last gen thing.

 

That's what I did with my CPU and GPU. I wasn't searching for comparisons between the i5-6600k and i5-4690k because I owned a 4690, I wanted to know if it was worth it to forgo the "old tech discount" for the newest thing. I was actually upgrading from a 10 year old dual core pentium. And somehow I'm still not sure if I made the right choice haha.

 

[BuckGup]

Same can be said for any hobby

[Chris94_NOR]

I am on the 4th Intel CPU generation. I am not interested yet to go further. I will continue to stay on my i7-4790K. I have thought about buying a 6th generation CPU, but i won't. Don't see it necessary.

[Mira Yurizaki]

I want to make two points here:

 

First "eSports gaming" computers aren't probably bleeding edge either. Many eSports games are light as possible to start. Another point, why would you want to do anything that would harm performance like crank up the details? Do you need super soft shadows? AA? Ambient occlusion? Texture resolution so fine you can jam the camera against their skin and see in the pores? No. Set everything on low or medium, except the resolution and draw distance if available because they have merit, and go to town. You don't need anything more than an i5 and a GTX 1060 or RX 480.

 

Case in point, take a look at On Winning's recommended settings for Overwatch to play competitively. The detail settings are almost all low

 

Second, I found this last night while I was bored and looking up some details on Intel processors:

Spoiler

This tells me that Intel has been working hard on the front-end and some of those aspects have improved significantly. So really all this tells me is that software is now lagging behind and it needs to catch up. Or that we've reached some theoretical limit of Amdahl's law and we're refusing to believe.

[ARikozuM]

37 minutes ago, M.Yurizaki said:

 

Second, I found this last night while I was bored and looking up some details on Intel processors:

  Reveal hidden contents

This tells me that Intel has been working hard on the front-end and some of those aspects have improved significantly. So really all this tells me is that software is now lagging behind and it needs to catch up. Or that we've reached some theoretical limit of Amdahl's law and we're refusing to believe.

I think users and developers are too obsessed with the "if it ain't broken" mindset to allow for better software.

[Mira Yurizaki]

5 minutes ago, ARikozuM said:

I think users and developers are too obsessed with the "if it ain't broken" mindset to allow for better software.

And in some cases, reliable software is better than better performing software.

 

Would you rather have an OS that boots instantly and runs its stuff super fast, but is buggy as hell and routinely hard crashes, or an OS that can have an uptime of forever but takes a while to do things?

[Scruffy90]

21 hours ago, tmcclelland455 said:

You can run all of those on something like a 2600k or 2700k just fine lmao

Most folks seem to be under the notion that the bottleneck is high when in reality its like 2-5% tops.

[glunday]

6 hours ago, tmcclelland455 said:

You can do that on the 2500/2600K... Even though they're quite a few generations old, Intel hasn't done jack shit since they don't have any competition to push them.

you're generalizing, i upgraded my friends mobo and cpu from a 3770k and saw a huge improvement in cities skylines, space engineers, crises 3, and more. as i said before if you bothered to read. there's a difference between just running it and running it on high textures and high fps WELL.

[ARikozuM]

3 hours ago, M.Yurizaki said:

And in some cases, reliable software is better than better performing software.

 

Would you rather have an OS that boots instantly and runs its stuff super fast, but is buggy as hell and routinely hard crashes, or an OS that can have an uptime of forever but takes a while to do things?

Both. I don't mind using beta or experimental software. I like to use different forms of software to find bugs and reporting them. 

[Mira Yurizaki]

6 minutes ago, ARikozuM said:

Both. I don't mind using beta or experimental software. I like to use different forms of software to find bugs and reporting them. 

Well, I mean, direct that question to the general population.

[flibberdipper]

2 hours ago, glunday said:

you're generalizing, i upgraded my friends mobo and cpu from a 3770k and saw a huge improvement in cities skylines, space engineers, crises 3, and more. as i said before if you bothered to read. there's a difference between just running it and running it on high textures and high fps WELL.

Clearly there must have been something wrong with his machine because even my Pentium and 7950 can nearly max all of those out. Not even shitting when I say that my 7950 is ALWAYS the limit in Crysis 3, and only Cities gives it a problem.

[glunday]

6 hours ago, tmcclelland455 said:

Clearly there must have been something wrong with his machine because even my Pentium and 7950 can nearly max all of those out. Not even shitting when I say that my 7950 is ALWAYS the limit in Crysis 3, and only Cities gives it a problem.

but you're running an 7950 and he had a gtx 1080, the cpu bottlenecked the gpu in his config and it seems like the gpu is the bottleneck in your config

[flibberdipper]

31 minutes ago, glunday said:

but you're running an 7950 and he had a gtx 1080, the cpu bottlenecked the gpu in his config and it seems like the gpu is the bottleneck in your config

I highly doubt a 3770K would bottleneck a 1080 so badly that you'd see a "massive" improvement in Crysis 3.