I believe it would require a complete paradigm shift. Exactly. That's what needs to happen. Even if intel and AMD manage to release a 1Ghz processor, the thing will just be wasting clock cycles.
Processors are plenty damn fast today. But they are useless if they don't have information to process. If we could process things without making the CPU wait, then we would see some pretty snappy performance. --
The reason ram is so slow now is because its so cheap No, the reason RAM is cheap is because it is slow. There's a difference.
Do you understand how a cache works? The speed of the cache is not what works the magic. That's only part of it. Caches work on the principles of temporal and spacial locatity. RTFM.
The CPU spends hardly any time whatsoever accessing memory outside the cache, speeding up main memory to core speeds wouldn't speed the computer up very much at all. There would be no need for a cache if the processor could read and write to main memory at core processor speeds.
What I'm saying is that I'd rather see main memory speed up and the elimination of cache. I don't want to turn main memory into cache. That won't work.
You are right about disabling cache slowing down your system. If you had no cache and your RAM ran at core processor speeds, your system would be faster than a system with core speed cache and low speed RAM.
Sure, pricing is the key point here. But the research dollars could be better spent. Processors are plenty fast today. I/O subsystems are the areas that need our attention. If core speed RAM came out, geeks like us would be the first in line to buy it. Everything is high priced when it first hits the market. Prices would drop. --
Well, I'm sure we'll have 800mhz ram in a few years, of course CPU speeds will be around 3 gigHz. The only way to get (cheap) ram speeds that fast is by stopping CPU development. Anyway it doesn't really matter, because almost all Memory access come from out of the CPU catch. (If you have a PC, try disabling the l1 and l2 catches in the BIOS, see how fast your system is without them). Speeding up raw main memory won't get you that much of an increase. And what would you rather pay for, a $25 chip that speeds up a few routines, or $10,000 worth of ultra high-speed ram?
Sorry to be anal about this, but it is "cache".
I understand they are trying to make a few specific things faster, but my point is work on making the RAM run at processor core (which for the most part, happens to be what cache RAM runs at) speed. What I'm thinking of is eliminating the cache and make memory run at cache speed. The reason that cache exists is because main memory is slow. If main memory is very fast, there is no need for cache.
And, quite frankly, unless my $25 can speed my system up noticeably, I won't spend it.
Why would cheap RAM speeds stop CPU development? I could see increase RAM production lowering CPU production, but not stopping development. --
What's the point of trying to speed up memory allocations? Why not just make the hardware faster?
I'm not giving everyone the right to go out and write inefficient code. It just seems that this research will not get very far. By the time they have something working, memory will be faster than their hardware solution. Sure, just use their hardware to make things even faster. Only if they get the cost down very low.
Barjne looked at some specialized machines. They just weren't cost justified. I would have to think that is the case here.
I'd rather see research into fabricating memory that ran at core processor speeds. That would speed up every memory access, not just the malloc()s and free()s.
Congrats on beating MTV to the 1,000,000 video mark!
Some interesting things: What story had the most trolls? What story had the most comments? What was the first story? How much disk space do the stories and comments currently use? --
Credit where credit is due, just in case ESR is lurking.
I remember ARV coming from the movie Revenge of the Nerds, when the two main nerds (Anthony Edwars and the Other Guy) were in the riding in the car to go to college. They were calculating their ETA based on speed and two ARVs. --
You're just shifting the work around a bit. The same work has to be done, regrouping free space, you're just trying to defer it until a convient time. I can agree with that.
I don't see where your argument about placing memory for new() on the stack holds up. The point of dynamic object creation is to allow the program to create an aribitrary number of objects and then free() them when done with them.
Trivial example that really makes the stack argument fall apart: Let's say I have 500 objects new()ed. I'm assuming that the first object created would be on the bottom of the stack, and the last object created would be on the top. I now delete the first 499 items. I now have 499 * object size bytes of memory on my stack that I can't use becuase I still have the 500th item on the stack. I can't recover the memory because it's on the stack.
Strange considering most of the problems seem simple enough to fix... Get off your duff, download the source, and start hacking. If it is so simple, fix it. --
If we are to take the "growing pains" argument to RL, however, we find that it loses some validity. When I join (or form) a group of people that meet to do similar things, such as a book club, if one or two of the members start berating or browbeating the other members, I'm not going to call that "growing pains." I'm going to point out that those people are being rude, and probably ask them to leave.
Right. But, you're putting things into the physical perspective. It took time for book clubs to form. First, books had to be invented. Clubs also had to be formed. Rules of discussion had to be formed. Let's pretend that you invent the book, the club (a gathering, not a weapon), and the discussion on the same day. Sit a group of people down, give them these three things, and you've got trouble. The people will eventually work out a protocol of reading, grouping, and discussing, but it won't happen on the first day. It won't happen on the second day. It will take time.
The internet (in the widely available consumer model ) has been around for only 5 years or so. You can't expect people to develop norms and protocols that quickly. There are some commonly identified protocols/responses/objects (spam, flames, newbies, etc.) that exist. In the book club situation, you ask the combatants to leave. This is the established protocol. Of course, another person may handle things differently by hitting the combatants. There will alway be people that don't hold to the norms.
You also suggest a method for dealing with flamers. I fully agree. But this is not the norm yet. Part of my original argument was that e-mail (I should extend that to all e-communication) is too quick. People don't stop to think. Thinking needs to become part of the protocol, and it currently is not. Give it some time, and I think it will change.
Human society has had centuries to develop. The internet has had 5 years. --
The real problem with online communication (e-mail,/., etc.) is that it is too quick. It's pretty easy to fire off an e-mail instead of print off a memo or go and talk to someone. The "thinking time" is removed. If you have to get up and go somewhere to complete your communication, you have time to think, and you usually tone down your message. Typing a message and clicking "send" removes the "thinking time".
I've been in a company where we went from having no e-mail to having e-mail. There was a period where e-mail was flooded with knee jerk responses and offers for free kittens. After the learning curve, people got the idea, and e-mail because usefull, not just a buch of junk. I think the internet, given the huge number of people, will need a great deal of time for the signal to noise ratio to settle down. There will always be newbies and downright rotten people to mess things up, but there is hope for the newbies.
We were all newbies once. We got better.
Remember, this whole internet thing has been around for just a short time. It took many years for societies to develop. This internet society needs to have its growing pains also. Things will get better, but given the vast number of people involved, it is going to take a few decades for things to settle out. --
Slashdot Mirror
Just imagine... it could be the next generation of "punch the monkey" ads.
Oh. Punch. This whole time I'd been...well, you know. He was a bad monkey.
--
"Give peace a chance."
Don't build a computer to test nukes, build a computer to help cure cancer. Let's use our high end processing power for playing chess and doing good.
Okay, I'm probably OT, but I was listening to Mr. Lennon last night, and sometimes his lyrics just ring so true.
--
I believe it would require a complete paradigm shift.
Exactly. That's what needs to happen. Even if intel and AMD manage to release a 1Ghz processor, the thing will just be wasting clock cycles.
Processors are plenty damn fast today. But they are useless if they don't have information to process. If we could process things without making the CPU wait, then we would see some pretty snappy performance.
--
The reason ram is so slow now is because its so cheap
No, the reason RAM is cheap is because it is slow. There's a difference.
Do you understand how a cache works? The speed of the cache is not what works the magic. That's only part of it. Caches work on the principles of temporal and spacial locatity. RTFM.
The CPU spends hardly any time whatsoever accessing memory outside the cache, speeding up main memory to core speeds wouldn't speed the computer up very much at all.
There would be no need for a cache if the processor could read and write to main memory at core processor speeds.
What I'm saying is that I'd rather see main memory speed up and the elimination of cache. I don't want to turn main memory into cache. That won't work.
You are right about disabling cache slowing down your system. If you had no cache and your RAM ran at core processor speeds, your system would be faster than a system with core speed cache and low speed RAM.
Sure, pricing is the key point here. But the research dollars could be better spent. Processors are plenty fast today. I/O subsystems are the areas that need our attention. If core speed RAM came out, geeks like us would be the first in line to buy it. Everything is high priced when it first hits the market. Prices would drop.
--
Well, I'm sure we'll have 800mhz ram in a few years, of course CPU speeds will be around 3 gigHz. The only way to get (cheap) ram speeds that fast is by stopping CPU development. Anyway it doesn't really matter, because almost all Memory access come from out of the CPU catch. (If you have a PC, try disabling the l1 and l2 catches in the BIOS, see how fast your system is without them). Speeding up raw main memory won't get you that much of an increase. And what would you rather pay for, a $25 chip that speeds up a few routines, or $10,000 worth of ultra high-speed ram?
Sorry to be anal about this, but it is "cache".
I understand they are trying to make a few specific things faster, but my point is work on making the RAM run at processor core (which for the most part, happens to be what cache RAM runs at) speed. What I'm thinking of is eliminating the cache and make memory run at cache speed. The reason that cache exists is because main memory is slow. If main memory is very fast, there is no need for cache.
And, quite frankly, unless my $25 can speed my system up noticeably, I won't spend it.
Why would cheap RAM speeds stop CPU development? I could see increase RAM production lowering CPU production, but not stopping development.
--
What's the point of trying to speed up memory allocations? Why not just make the hardware faster?
I'm not giving everyone the right to go out and write inefficient code. It just seems that this research will not get very far. By the time they have something working, memory will be faster than their hardware solution. Sure, just use their hardware to make things even faster. Only if they get the cost down very low.
Barjne looked at some specialized machines. They just weren't cost justified. I would have to think that is the case here.
I'd rather see research into fabricating memory that ran at core processor speeds. That would speed up every memory access, not just the malloc()s and free()s.
--
I do. I really do. I Blame Canada for this!
--
Let me know when Signal11's Karma hits 10K! Or has it already?
--
Congrats!
Congrats on beating MTV to the 1,000,000 video mark!
Some interesting things:
What story had the most trolls?
What story had the most comments?
What was the first story?
How much disk space do the stories and comments currently use?
--
An old friend of mine gave me a Kinesis. He got it from a friend who couldn't get used to it. He couldn't get used to it either, so he gave it to me.
It took me about three days to get back up to speed on the thing, but once I did, I wouldn't give it up.
And, as an added bonus, mine doesn't have a windows key on it.
They just rock. Pricy, but worth it.
--
Newsflash! People can drink "pure water"!
"Hydrogen is so clean burning that what comes out of the exhaust pipe is pure water," Melis said. "You can drink it."
--
Credit where credit is due, just in case ESR is lurking.
I remember ARV coming from the movie Revenge of the Nerds, when the two main nerds (Anthony Edwars and the Other Guy) were in the riding in the car to go to college. They were calculating their ETA based on speed and two ARVs.
--
...a bathroom factor.
Basically, can you get anything out of the book during an ARV (Average Restroom Visit)?
--
I'm offtopic, but...
Thanks for SFC. That game just rocks. Kudos to the whole team. If I could, I would buy beers for all of you.
Wonderful job of capturing the feel of SFB.
--
SFB on my PC! Finally. This is the game I've wanted for 15 years.
A Linux port would be so nice.
--
Oh yeah. The promise of Diablo II is keeping win98 alive on my system. Well, that and Starfleet Command.
--
I snagged 1.0.0 as soon as I could get it. I also snagged the Aureal drivers last weekend.
:P
Everything is working fine. Odd. I'll have to check which version I have when I get home.
Or perhaps I've been listening to the same songs for so long that I'm starting to hear them when I sit down in front of the computer.
--
You're just shifting the work around a bit. The same work has to be done, regrouping free space, you're just trying to defer it until a convient time. I can agree with that.
--
I don't see where your argument about placing memory for new() on the stack holds up. The point of dynamic object creation is to allow the program to create an aribitrary number of objects and then free() them when done with them.
Trivial example that really makes the stack argument fall apart:
Let's say I have 500 objects new()ed. I'm assuming that the first object created would be on the bottom of the stack, and the last object created would be on the top. I now delete the first 499 items. I now have 499 * object size bytes of memory on my stack that I can't use becuase I still have the 500th item on the stack.
I can't recover the memory because it's on the stack.
Am I missing your argument? Help me out here.
--
Strange considering most of the problems seem simple enough to fix...
Get off your duff, download the source, and start hacking. If it is so simple, fix it.
--
This is great! Instead of waiting for Cavedog to write Total Annihilation 2, we'll be able to play for real.
--
Hey Havoc! Is that your Dad that was pissing BillG off? That's pretty cool.
--
If we are to take the "growing pains" argument to RL, however, we find that it loses some validity. When I join (or form) a group of people that meet to do similar things, such as a book club, if one or two of the members start berating or browbeating the other members, I'm not going to call that "growing pains." I'm going to point out that those people are being rude, and probably ask them to leave.
Right. But, you're putting things into the physical perspective. It took time for book clubs to form. First, books had to be invented. Clubs also had to be formed. Rules of discussion had to be formed. Let's pretend that you invent the book, the club (a gathering, not a weapon), and the discussion on the same day. Sit a group of people down, give them these three things, and you've got trouble. The people will eventually work out a protocol of reading, grouping, and discussing, but it won't happen on the first day. It won't happen on the second day. It will take time.
The internet (in the widely available consumer model ) has been around for only 5 years or so. You can't expect people to develop norms and protocols that quickly. There are some commonly identified protocols/responses/objects (spam, flames, newbies, etc.) that exist. In the book club situation, you ask the combatants to leave. This is the established protocol. Of course, another person may handle things differently by hitting the combatants. There will alway be people that don't hold to the norms.
You also suggest a method for dealing with flamers. I fully agree. But this is not the norm yet. Part of my original argument was that e-mail (I should extend that to all e-communication) is too quick. People don't stop to think. Thinking needs to become part of the protocol, and it currently is not. Give it some time, and I think it will change.
Human society has had centuries to develop. The internet has had 5 years.
--
The real problem with online communication (e-mail, /., etc.) is that it is too quick. It's pretty easy to fire off an e-mail instead of print off a memo or go and talk to someone. The "thinking time" is removed. If you have to get up and go somewhere to complete your communication, you have time to think, and you usually tone down your message. Typing a message and clicking "send" removes the "thinking time".
I've been in a company where we went from having no e-mail to having e-mail. There was a period where e-mail was flooded with knee jerk responses and offers for free kittens. After the learning curve, people got the idea, and e-mail because usefull, not just a buch of junk. I think the internet, given the huge number of people, will need a great deal of time for the signal to noise ratio to settle down. There will always be newbies and downright rotten people to mess things up, but there is hope for the newbies.
We were all newbies once. We got better.
Remember, this whole internet thing has been around for just a short time. It took many years for societies to develop. This internet society needs to have its growing pains also. Things will get better, but given the vast number of people involved, it is going to take a few decades for things to settle out.
--
What about a Beouwolf cluster of these?
--