News:

Do you need help?
Simutrans Wiki Manual can help you to play and extend Simutrans. In 9 languages.

general survey on the players play Simutrans

Started by Frank, November 28, 2008, 06:55:37 PM

Previous topic - Next topic

0 Members and 2 Guests are viewing this topic.

DrSuperGood

QuoteBut also when the computer idles, to be able to provide stutter free, say 4k, video, and simple 3d all must be able to handle greater peaks, which limits power saving through undervolting, increases leak currents due to larger dies, requires larger heatsink areas, greater max currents from batteries. When dealing with baseline power of a few Watt this matters.*
Clock rates dynamically drop when not running performance intensive applications or running in an energy saving mode (power management). This provides the biggest power savings since power consumption is related to clock rate squared (this is because of the inter-track capacitance aka the cause of leakage). Sure they will never be super energy efficient due to the size and design (they are performance parts) but then again power savings can be made elsewhere of equal magnitude.

Additionally even if running at full clock speed a processor will use less power idling than it does doing something (any instruction other than idle). This is because most of the processor is in a steady state and switching the transistors causes a flow of current (so I2R losses).

As such the energy cost for performance is minimal. It is often better to build a laptop capable of performing when required rather than one which is slightly more energy efficient but has a much lower maximum performance. If a person wants to watch full 1080p video (something not unreasonable in this day and age) then the laptop should have those resources available to do so. When the resources are not used their energy overhead can be minimized to acceptable levels.

Cooling has other issues not related to performance. Laptops mostly have under designed cooling as they are not designed to run at full performance for more than a few minutes at a time. If someone is playing a game (modern demanding games especially) on a laptop and reports crashes than 9/10 times it is due to the laptop overheating due to insufficient cooling.

Your backlight power issue can also be minimized by manually setting your backlight brightness. Setting it dimmer (which most laptops allow one to) decreases the power it uses and so a 5 watt source might end up considerably less. In really bright situations it may even be possible to disable the backlight all together, a feature that was used by early portable LCD devices so that they did not need a backlight at all (something that was hard to make originally). There also exists alternative display technologies such as OLEDs which can provide better energy efficiency as it only illuminates pixels that need light but these currently are too costly for the average consumer.

Ideally laptops should have more intelligence for controlling their processor resources. Specifically the full power of the processor should be more limited in availability to only under exceptional performance loads and even then in small bursts to prevent overheating. The technology is either not there yet or so obscure that people seldom use it.

The problem with Simutrans is that of its design. It produces a single heavy thread that utilization depends on the output display area in pixels. Many laptops have high density displays (eg some MacBook Pros and many tablet devices) and rely on multi-core processors for energy saving (most modern portable devices, run many tasks in parallel at a slower clock speed for power saving). Running Simutrans on such a device full screen runs the processor at full clock speed doing a lot of work continuously on a single core. Obviously what one views makes a difference but simple actions like panning the window can waste a lot of processor time. This problem has also been reduced thanks to multi-thread rendering functionality (if enabled). As such it can be quite energy efficient and has no problems running on most portable devices.

However for those people who want to get a little more battery life out of their device when playing Simutrans I suggested a possible solution a while ago. There should be some functionality to allow for "upscaling" of a lower resolution output to the full screen size. When playing in full screen mode this could be achieved with non-native resolutions supported by the display (and letting the display driver handle the scaling). When playing in windowed mode some form of scale factor could be used with an OpenGL (or something else) accelerated up scaling. This could be used to reduce the number of pixels filled when playing as well as handle high density display scale problems (too small to resolve what is happening).

sdog

#71
QuoteYou're not supposed to see individual pixels. That's the point of high-res.
If there are no conditions under which individual pixels can be seen, than it is certainly over-engineered. Individual pixels only have to be unnoticeable under under usage conditions plus a decent margin.

Example for badly overengineered: Build a car that can drive very well at close to 200km/h for markets where the maximum speed is 130 km/h. Side-effect, because such a car looks cool, the whole fleet gets over-motorized and too heavy because of market forces and irresponsible consumers.

QuoteI remember when we used to look at desks. Those usually were, and still are, wider than [one metre]. Often much wider. Sure, one tended to focus on just one part, but I guess that's the idea with big screens as well.
To keep a good overview when the screen size increases would require to move further out. That is perhaps not such a bad thing, possibly reducing eye strain. However, the pixel density can also be lower while keeping the same angular resolution, i.e. just enough that individual pixels are not discernible. Increasing the resolution further quickly provides only diminishing returns. Unless one gets to the point where not the whole screen ought to be perceived at one time, traditionally done by multiple monitors.


Quote from: DrSuperGood on April 16, 2015, 06:25:12 PM
Clock rates dynamically drop [...] Sure they will never be super energy efficient due to the size and design (they are performance parts) but then again power savings can be made elsewhere of equal magnitude.
That is exactly my point.

Quote
As such the energy cost for performance is minimal. It is often better to build a laptop capable of performing when required rather than one which is slightly more energy efficient but has a much lower maximum performance. If a person wants to watch full 1080p video (something not unreasonable in this day and age) then the laptop should have those resources available to do so. When the resources are not used their energy overhead can be minimized to acceptable levels.
1080p is the lower end today, and i think a very sensible resolution. The next step in resolution is 4k, and i argue it costs more than it provides. I tend to differ in the first part. Saving one or two Watt of peak power might often be quite relevant, just the difference between passive and active cooling. The shift to completely closed passively cooled notebooks is perhaps the most important step in notebook design presently.

Quote
Cooling has other issues not related to performance. Laptops mostly have under designed cooling as they are not designed to run at full performance for more than a few minutes at a time. If someone is playing a game (modern demanding games especially) on a laptop and reports crashes than 9/10 times it is due to the laptop overheating due to insufficient cooling.
Only a badly designed machine. It ought to go into thermal throttling. But it is exactly the point that increasing possible peak power has a whole train of secondary design costs attached. Therefore it is preferable to be cautious about the max TDP when designing a system.

I think that is also one of the reasons the Core M is so beneficial to notebook design, the whole series has a TDP of 4.5 W.

Quote
Your backlight power issue can also be minimized by manually setting your backlight brightness. Setting it dimmer (which most laptops allow one to) decreases the power it uses and so a 5 watt source might end up considerably less. In really bright situations it may even be possible to disable the backlight all together, a feature that was used by early portable LCD devices so that they did not need a backlight at all (something that was hard to make originally). There also exists alternative display technologies such as OLEDs which can provide better energy efficiency as it only illuminates pixels that need light but these currently are too costly for the average consumer.
I don't know what you are referring to exactly. The backlight is a major power sink of course, and not much dependent on screen resolution. It indeed takes the lion's share of my battery, at this time 5.5 W of 9 W, while mostly idling.

Quote
Ideally laptops should have more intelligence for controlling their processor resources. Specifically the full power of the processor should be more limited in availability to only under exceptional performance loads and even then in small bursts to prevent overheating. The technology is either not there yet or so obscure that people seldom use it.
Well, that is not a 'should' but an is. There is also not much to do, the OS and frequency governor. I've just pulled the plug to go from ondemand to powersafe, both my cores are more than 95% of the time in C7 state. 0.1% of the time at 1.5GHz (3 GHz max). And this is a 5 year old cpu, and running under Linux that is not yet as well optimised for power saving as other OSes are. I think lack of proper controll of processor resources is a problem of the distant past.

Quote
The problem with Simutrans is that of its design. It produces a single heavy thread that utilization depends on the output display area in pixels. Many laptops have high density displays (eg some MacBook Pros and many tablet devices) and rely on multi-core processors for energy saving (most modern portable devices, run many tasks in parallel at a slower clock speed for power saving). Running Simutrans on such a device full screen runs the processor at full clock speed doing a lot of work continuously on a single core. Obviously what one views makes a difference but simple actions like panning the window can waste a lot of processor time. This problem has also been reduced thanks to multi-thread rendering functionality (if enabled). As such it can be quite energy efficient and has no problems running on most portable devices.

However for those people who want to get a little more battery life out of their device when playing Simutrans I suggested a possible solution a while ago. There should be some functionality to allow for "upscaling" of a lower resolution output to the full screen size. When playing in full screen mode this could be achieved with non-native resolutions supported by the display (and letting the display driver handle the scaling). When playing in windowed mode some form of scale factor could be used with an OpenGL (or something else) accelerated up scaling. This could be used to reduce the number of pixels filled when playing as well as handle high density display scale problems (too small to resolve what is happening).
Well that is a completely different beast. Blitting 2d is also not a very typical application on notebooks, else one would return to single core, single thread designs. In particular a mid 1990s approach scaled up some orders of mangitude just because modern hardware also provides more single thread performance.

Ters

Quote from: sdog on April 16, 2015, 11:21:53 PM
If there are no conditions under which individual pixels can be seen, than it is certainly over-engineered. Individual pixels only have to be unnoticeable under under usage conditions plus a decent margin.

I wan't suggesting that there is a goal to make pixels smaller than the Planck length, just so small that they can't be individually seen by the naked eye. A 20" 4K screen might be enough, I haven't tried.

Quote from: DrSuperGood on April 16, 2015, 06:25:12 PM
The problem with Simutrans is that of its design. It produces a single heavy thread that utilization depends on the output display area in pixels. Many laptops have high density displays (eg some MacBook Pros and many tablet devices) and rely on multi-core processors for energy saving (most modern portable devices, run many tasks in parallel at a slower clock speed for power saving). Running Simutrans on such a device full screen runs the processor at full clock speed doing a lot of work continuously on a single core. Obviously what one views makes a difference but simple actions like panning the window can waste a lot of processor time. This problem has also been reduced thanks to multi-thread rendering functionality (if enabled). As such it can be quite energy efficient and has no problems running on most portable devices.

Although it's reported that multithreading the rendering does help, there is still an issue that the bus might not be dimensioned for moving full screen images at 25 Hz in addition to all the other data Simutrans processes. (Actually, when panning around the map, it's more than two full screen images, because Simutrans always renders a complete image to the back buffer with some overdraw, then, in this case, needs to blit the entire thing to the front buffer.)

sdog

Quote from: Ters on April 17, 2015, 05:07:00 AM
I wan't suggesting that there is a goal to make pixels smaller than the Planck length, just so small that they can't be individually seen by the naked eye. A 20" 4K screen might be enough, I haven't tried.
Planck lenght displays!
Angular resolution is just 4 arc minutes, or more convenient unitless 2.9e-4 (used to be rad). With small angle approximation tan x = sin x = x at 1 m distance nothing smaller than 0.3 mm is discernible.

A typical notebook display is about 11" wide (12.5" diagonal, 16:9) that is about 30 cm. Dividing it into 1000 pixels is already sufficient. I.e. there is no gain to even go from the WXGA resolution  (1280) to FHD (1920). It is still ok, because it is very much a standard resolution and a lot of things fit natively, and it doesn't hurt to have quite a bit higher resolution. The next resolution step to 4K means 80 μm pixels. At 1 m they fill 1/5 of the human angular resolution.
Quote
Although it's reported that multithreading the rendering does help, there is still an issue that the bus might not be dimensioned for moving full screen images at 25 Hz in addition to all the other data Simutrans processes. (Actually, when panning around the map, it's more than two full screen images, because Simutrans always renders a complete image to the back buffer with some overdraw, then, in this case, needs to blit the entire thing to the front buffer.)
Well, we get away with it because modern hardware is so extremely powerful. Are there secondary applications that sample low res up to the native resolution? Might be more sensible than actually dealing with the problem yourself.

Ters

Quote from: sdog on April 17, 2015, 04:55:25 PM
1 m distance nothing smaller than 0.3 mm is discernible.

My display is more like 0.5 meters aways, at least when using laptop on my lap. When at my desk, the distance varies with sitting positure. My laptop is also 15.6". It's also just 1366 pixels wide, which is good, because that gives it almost the same DPI as my external 22" HD monitor. I think both could benefit from better resolution (not so much for text due to cleartype, although cleartype doesn't cope as well when screen is rotated), although splitting current pixels in four as in 4K might indeed be overkill.

(PS. A rotated HD screen looks ridiculously narrow.)

Quote from: sdog on April 17, 2015, 04:55:25 PM
Well, we get away with it because modern hardware is so extremely powerful.

When I hadn't set up hardware accelerated video decoding on my Linux box, it couldn't handle HD video. Sure, it wasn't the most powerful CPU, but from what I read, doing HD video without GPU help isn't a good idea, even though the software decoder is multithreaded.

DrSuperGood

QuoteI don't know what you are referring to exactly. The backlight is a major power sink of course, and not much dependent on screen resolution. It indeed takes the lion's share of my battery, at this time 5.5 W of 9 W, while mostly idling.
Adjusting display brightness usually adjusts the power going to the back light. This is why many laptops automatically dim when disconnected from a charger. At very bright settings you might notice black contrast being reduced as the LCD can only filter a certain fraction of the light out to form black.

Do note that a laptop and notebook are two entirely differently designed computer systems. A laptop is generally bigger and is designed for performance at times. Notebooks generally are smaller and designed for battery life. This could be seen with the sort of processors they used with laptops generally using revised versions of desktop processors while notebooks uses special low energy processors. In this day and age the line between tablet and notebook is becoming increasingly blurred and fundamentally they are very similar.






Combuijs

Bob Marley: No woman, no cry

Programmer: No user, no bugs