X uses IPC which tests have proven on Linux is almost as fast as shared memory
Which is still magnitudes worse. Wayland latency is low tens of nano-seconds(passes around pointers), X is low micro to high nano(copies buffers via IPC/shared memory). HUGE difference.
Working with scalability problems helps one appreciate latency. Processing power keeps scaling, but latency keeps increasing relative to throughput. X used to be good for local machines back when latency was in harmony with throughput.
X won't be "replaced" any time soon by Wayland as they currently solve different issues. They do have overlap, but they're not the same.
My "City" is fewer than 20k people and we're getting symmetrical fiber internet 30/50/100/200 for $60/$100/$200/$300. No hidden fees, listed price is what you pay.
Even a county in Minnesota got 30/30 fiber internet for $100/month, and they have an average population density of 5/mi^2. The ISP covers slightly over 1,000mi^2 and the area has slightly over 5,000 total people(Under 2,500 families?). They did this fewer than 8 years ago and have already paid off the fiber install and are now making profit.
Incumbents are just trying to milk money from their old infrastructure. Seems to me that these small ISPs in the middle-of-no-where are supplying faster+cheaper internet.
They charge $10 per 50GB over, which is reasonable
Large ISPs pay based on 95th percentile. If you transferred 10TB during off-peak hours, it would be free for them. Depending on the time of day that you transfer those $50, the price could range from $0 to $1/mbit(rate that you transferred those 50GB). Since AT&T is both an ISP and Tier1 backbone, they probably have peering agreements everywhere and that 50GB was free.
You need to make sure that the upload rate is throttled on your router. QoS is almost useless if you don't throttle as most DSL/cable modems have HUGE buffers and you get nasty bufferbloat.
Charter was $60(plus $25 in fees) for 30Mb over here, then the local ISP announced 30/30 fiber for $60 in the coming year, so Charter dropped it down to $45 for 6month intro and 1 year contract for new customers. One month before the fiber went live, Charter started a $30 for 30Mb for 12 months for new customers with a 2 year contract.
I called up Charter and asked if I could get on that 30Mb for $30/month and they told me the $90/month($60+"fees") I was paying was a good deal.
Needless to say, I will be switching(once the light up my neighborhood) over to 50/50 fiber internet with no datacap and they claim to be 100% net-neutral with no traffic shaping QOS or anything, for the same price. Did I mention the local ISP doesn't have any "fees", listed prices include EVERYTHING.
GPON(2.5Gb) - Assuming all users on a 32 person node max the connection, about 70Mb/s
At this point the end user has an effective dedicated 70Mb/s(78Mb/s but rounded down for safety margin) to the ISP, the ISP can then purchase 1Mb/s of dedicated bandwidth for $1/month and even cheaper if peering.
For $70/month, the ISP can purchase 70Mb/s. Add the cost of infrastructure(connection fee), $15/month. For $100/month, one could expect 70Mb/s of dedicated bandwidth and a small margin of profit for the ISP. I'm sure there are other costs and the ISP needs enough margin to cover future expansion/etc, but it's not far-fetched to get decent amounts of dedicated bandwidth.
This is not a dedicated connection, but that doesn't remove the fact that there is a 1:1 ratio of max consumption to available bandwidth. I'm sure other bottlenecks will occur.
State non-profit co-op sells 1Gb dedicated fiber connections to Hospitals/Schools/Libraries for $300/month. And yes, they do get 1Gb effective. This is because the state University buys bandwidth in bulk, has massive peering agreements, and re-sells at non-profit rates; and these are the real-world prices to break even on a 1Gb dedicated line.
Before you break into the "laying fiber is expensive", real world recent studies have shown running fiber from the ISP to the house plus datacenter equipment is actually LESS expensive than the fiber receiver installed. Installing the fiber to Gb Ethernet converter at each customer's house makes up 60%-70% of the total costs.
With all programs opened, HD IO is closer to low 10s of KB/sec, not MB. Most of my IO is network traffic.
After a year of randomly benchmarking my SSD, having to reinstall Windows and 100GB of games a few times due to mistakes, it still is at 0% worn. At this relatively heavy usage rate, it will take more than 100 years to burn it out.
Don't base it on profit. Revenue or Assets or IPO value.. Whatever. Something that can't be manipulated without first ruining the company or causing criminal investigations for other reasons.
Social web sites are a communication medium. Are you saying that the phone company is responsible for hate speech going over its lines? Are you saying ISPs are responsible for illegal communications over its lines? Are you saying that US citizens are responsible for people using their voices to say hateful things through the air because air is publicly owned?
What if I say "I don't like politicians" and the company is so scared about "hate speech" that they remove my post out of fear to get fined. Suddenly a private company is behind held responsible to not only enforce removing hate speech, but defining what hate speech is as they must be conservative to make sure they don't lose any more money.
The government has indirectly removed my right to free speech by aggressively enforcing vague rules.
I think this post nails it. One must remember that Carmack has one who has worked with games and Valve is one that is a digital distribution system. Both points are valid, but they are different views of the same issue.
Carmack has never been in the position to have a remote chance of making a market in the sense that Value is attempting.
They did that because of the horrible Class Action laws the USA has. It doesn't do anything except funnel money from companies into the pockets of lawyers.
What Steam does give you is the right to use a 3rd party non-profit company that gets no money from Valve, to decide what kind of damages you got from Valve if you have a dispute. Valve is willing to pay up-to $10k in damages assuming that's what the 3rd party thinks is a fair value for the damage of losing games/etc.
The only "forays" I remember ID software making was paying other companies to port over quake engines years after they had already released and had the games were pretty well established on windows.
Treating it as a niche maket afterthought[...]
They didn't treat it as an "After-thought", they paid to port Quake to Linux so they could release the source under GPL and have it Linux ready.
In the beginning id sold games, but since Quake they've sold tech demos that advertised their game engines for other's to purchase. Having an engine have a fun game wrapped around it was just more money.
It was my understanding that the Windows Store will only allowed Metro apps. While you can advertise desktop apps in the store, the most you can do is have your advertisement link to your own website where you actually sell your desktop application.
Assuming Microsoft hasn't changed this, there is no benefit for most PC style games to still not use Steam since you still need someone to distribute your digital copy and Windows Store will only distribute Metro apps.
DX11 allows multiple cores to communicate with the GPU using a mostly lockless setup. It uses command queues. Each thread gets its own queue and the application registers the command queue with the context. When the main context threads pushes data(context switching), the drivers read from the other queues with no locking(no context switching).
This dramatically reduces context switching and allows the GPU hardware to pull the data and not the CPU to push the data. Very little locking overhead and scales nearly linearly up into the 16 cores/threads area. When 16 cores starts to become a bottleneck for video-games pushing graphics commands, then we can start to worry about an even better design.
I know AMD was looking into an event-based system where every program can register an arbitrary amount of queues and almost completely removes all CPU pushing data and lets the GPU pull data in an epoll/kqueue style manner. This would also be coupled with the GPU understanding and enforcing protected memory and working with the same memory addresses as the CPU. This makes it really easy to just pass pointers around as you don't need to translate the pointers for different memory spaces. AMD is working on this system for OpenSource and wants it to be part of Linux. w00t!
Warning: it has been a few years since I've read up on this info and could be slightly off, but the general ideas apply.
Too bad OpenGL doesn't support multiple threads(does on the back-end but not the front-end). This will eventually cause a huge bottleneck once games starts making use of multiple cores.
A prime example is Civ5. DX9/10can use about 6 cores of a 12core cpu. DX11 mode can make use of a bit over 11 cores and almost doubles the FPS. Entirely because a single core can only feed so many commands to the GPU. Once your hit that limit, you will not be able to increase your FPS without ether reducing the numbers of commands(reducing quality) or increasing the number of cores/threads.
First step first. Lets get Linux as a decent gaming platform, then worry about pretty colors.
Slashdot Mirror
You're lying and ignorant! cwhatIdidthar?
X uses IPC which tests have proven on Linux is almost as fast as shared memory
Which is still magnitudes worse. Wayland latency is low tens of nano-seconds(passes around pointers), X is low micro to high nano(copies buffers via IPC/shared memory). HUGE difference.
Working with scalability problems helps one appreciate latency. Processing power keeps scaling, but latency keeps increasing relative to throughput. X used to be good for local machines back when latency was in harmony with throughput.
X won't be "replaced" any time soon by Wayland as they currently solve different issues. They do have overlap, but they're not the same.
My "City" is fewer than 20k people and we're getting symmetrical fiber internet 30/50/100/200 for $60/$100/$200/$300. No hidden fees, listed price is what you pay.
Even a county in Minnesota got 30/30 fiber internet for $100/month, and they have an average population density of 5/mi^2. The ISP covers slightly over 1,000mi^2 and the area has slightly over 5,000 total people(Under 2,500 families?). They did this fewer than 8 years ago and have already paid off the fiber install and are now making profit.
Incumbents are just trying to milk money from their old infrastructure. Seems to me that these small ISPs in the middle-of-no-where are supplying faster+cheaper internet.
They charge $10 per 50GB over, which is reasonable
Large ISPs pay based on 95th percentile. If you transferred 10TB during off-peak hours, it would be free for them. Depending on the time of day that you transfer those $50, the price could range from $0 to $1/mbit(rate that you transferred those 50GB). Since AT&T is both an ISP and Tier1 backbone, they probably have peering agreements everywhere and that 50GB was free.
You need to make sure that the upload rate is throttled on your router. QoS is almost useless if you don't throttle as most DSL/cable modems have HUGE buffers and you get nasty bufferbloat.
Charter was $60(plus $25 in fees) for 30Mb over here, then the local ISP announced 30/30 fiber for $60 in the coming year, so Charter dropped it down to $45 for 6month intro and 1 year contract for new customers. One month before the fiber went live, Charter started a $30 for 30Mb for 12 months for new customers with a 2 year contract.
I called up Charter and asked if I could get on that 30Mb for $30/month and they told me the $90/month($60+"fees") I was paying was a good deal.
Needless to say, I will be switching(once the light up my neighborhood) over to 50/50 fiber internet with no datacap and they claim to be 100% net-neutral with no traffic shaping QOS or anything, for the same price. Did I mention the local ISP doesn't have any "fees", listed prices include EVERYTHING.
Since when has 300bps been considered slow?
P2P is dwarfed by Netflix alone
Hurricane Electric - $1/Mb advertised rate
GPON(2.5Gb) - Assuming all users on a 32 person node max the connection, about 70Mb/s
At this point the end user has an effective dedicated 70Mb/s(78Mb/s but rounded down for safety margin) to the ISP, the ISP can then purchase 1Mb/s of dedicated bandwidth for $1/month and even cheaper if peering.
For $70/month, the ISP can purchase 70Mb/s. Add the cost of infrastructure(connection fee), $15/month. For $100/month, one could expect 70Mb/s of dedicated bandwidth and a small margin of profit for the ISP. I'm sure there are other costs and the ISP needs enough margin to cover future expansion/etc, but it's not far-fetched to get decent amounts of dedicated bandwidth.
This is not a dedicated connection, but that doesn't remove the fact that there is a 1:1 ratio of max consumption to available bandwidth. I'm sure other bottlenecks will occur.
State non-profit co-op sells 1Gb dedicated fiber connections to Hospitals/Schools/Libraries for $300/month. And yes, they do get 1Gb effective. This is because the state University buys bandwidth in bulk, has massive peering agreements, and re-sells at non-profit rates; and these are the real-world prices to break even on a 1Gb dedicated line.
Before you break into the "laying fiber is expensive", real world recent studies have shown running fiber from the ISP to the house plus datacenter equipment is actually LESS expensive than the fiber receiver installed. Installing the fiber to Gb Ethernet converter at each customer's house makes up 60%-70% of the total costs.
1MB/s will only take 30 years!
256GB
assume 20% over-provisioning
307.2GB * 3000 writes = 921,600GB
921,600 * 1024 = 943,718,400MB
943,718,400 / 3600(hours) / 24(days) / 365 (years) = 29.925years
With all programs opened, HD IO is closer to low 10s of KB/sec, not MB. Most of my IO is network traffic.
After a year of randomly benchmarking my SSD, having to reinstall Windows and 100GB of games a few times due to mistakes, it still is at 0% worn. At this relatively heavy usage rate, it will take more than 100 years to burn it out.
Don't base it on profit. Revenue or Assets or IPO value.. Whatever. Something that can't be manipulated without first ruining the company or causing criminal investigations for other reasons.
Can't give one fine to one person and another fine to another (corporate) person.
I can and it does. Ever see Texas execute the heads of a company because said company caused people to die?
We should increase this to $50,000 immediately to prevent future abuses.
$50k is a bout much for smaller companies and nothing for large companies.
10% of gross yearly revenue post-tax and post deductions sounds much better to me, and the fine is non-tax-deductible.
Social web sites are a communication medium. Are you saying that the phone company is responsible for hate speech going over its lines? Are you saying ISPs are responsible for illegal communications over its lines? Are you saying that US citizens are responsible for people using their voices to say hateful things through the air because air is publicly owned?
What if I say "I don't like politicians" and the company is so scared about "hate speech" that they remove my post out of fear to get fined. Suddenly a private company is behind held responsible to not only enforce removing hate speech, but defining what hate speech is as they must be conservative to make sure they don't lose any more money.
The government has indirectly removed my right to free speech by aggressively enforcing vague rules.
NSFW
megatextures and honestly though were a kludge for a problem existing due to memory constrained systems
Exactly. Why can't more systems handle storing 48GB 128000×128000 textures in video memory? /sarc
I think this post nails it. One must remember that Carmack has one who has worked with games and Valve is one that is a digital distribution system. Both points are valid, but they are different views of the same issue.
Carmack has never been in the position to have a remote chance of making a market in the sense that Value is attempting.
They did that because of the horrible Class Action laws the USA has. It doesn't do anything except funnel money from companies into the pockets of lawyers.
What Steam does give you is the right to use a 3rd party non-profit company that gets no money from Valve, to decide what kind of damages you got from Valve if you have a dispute. Valve is willing to pay up-to $10k in damages assuming that's what the 3rd party thinks is a fair value for the damage of losing games/etc.
The only "forays" I remember ID software making was paying other companies to port over quake engines years after they had already released and had the games were pretty well established on windows.
Treating it as a niche maket afterthought[...]
They didn't treat it as an "After-thought", they paid to port Quake to Linux so they could release the source under GPL and have it Linux ready.
In the beginning id sold games, but since Quake they've sold tech demos that advertised their game engines for other's to purchase. Having an engine have a fun game wrapped around it was just more money.
It was my understanding that the Windows Store will only allowed Metro apps. While you can advertise desktop apps in the store, the most you can do is have your advertisement link to your own website where you actually sell your desktop application.
Assuming Microsoft hasn't changed this, there is no benefit for most PC style games to still not use Steam since you still need someone to distribute your digital copy and Windows Store will only distribute Metro apps.
DX11 allows multiple cores to communicate with the GPU using a mostly lockless setup. It uses command queues. Each thread gets its own queue and the application registers the command queue with the context. When the main context threads pushes data(context switching), the drivers read from the other queues with no locking(no context switching).
This dramatically reduces context switching and allows the GPU hardware to pull the data and not the CPU to push the data. Very little locking overhead and scales nearly linearly up into the 16 cores/threads area. When 16 cores starts to become a bottleneck for video-games pushing graphics commands, then we can start to worry about an even better design.
I know AMD was looking into an event-based system where every program can register an arbitrary amount of queues and almost completely removes all CPU pushing data and lets the GPU pull data in an epoll/kqueue style manner. This would also be coupled with the GPU understanding and enforcing protected memory and working with the same memory addresses as the CPU. This makes it really easy to just pass pointers around as you don't need to translate the pointers for different memory spaces. AMD is working on this system for OpenSource and wants it to be part of Linux. w00t!
Warning: it has been a few years since I've read up on this info and could be slightly off, but the general ideas apply.
Do they even know how to argue?
I like people who speak their mind, it lets me know who they are. No surprises, no backstabbing, you get what you see.
Too bad OpenGL doesn't support multiple threads(does on the back-end but not the front-end). This will eventually cause a huge bottleneck once games starts making use of multiple cores.
A prime example is Civ5. DX9/10can use about 6 cores of a 12core cpu. DX11 mode can make use of a bit over 11 cores and almost doubles the FPS. Entirely because a single core can only feed so many commands to the GPU. Once your hit that limit, you will not be able to increase your FPS without ether reducing the numbers of commands(reducing quality) or increasing the number of cores/threads.
First step first. Lets get Linux as a decent gaming platform, then worry about pretty colors.