man I remember having to calculate cycles, back when I use to do assembly programming on my amiga and MSX. As fun as this was, I'm glad we don't have to worry about things like that anymore :)
I was looking at some of my old 6510 assembler code earlier this evening, it has comments added to each instruction to describe the cycles used, including things like overlapping cycles (6502 and its derivates have a simple pipeline) and special conditions (crossing page boundaries requiring an extra cycle, 'stolen' cycles due to video or dma etc). The code is intended to run in cycle exact lock step with the painting of the screen (actually writing directly to the dac to implement a kind of software video controller). Like you, I'm glad to not have to do that anymore, but it was fun to do, and fascinating to read back (how on earth did I have all the info for that before the internet age.. much of it gleamed from diagrams and lots of experimentation)
One of the things people miss: The signal is electricity. All the actions are powered by this. The digital signal is really high voltage-low voltage-high voltage...etc. The electronics are designed to do something when the voltage goes up, being powered by the voltage jump. They then drop into their new state when the voltage drops. They then do the next thing when the voltage jumps up again. The clock is more than just something to synchronize work, it powers it. The timing is critical, since the pulse creates a cascade of work, which all has to complete before the next pulse. That is why you have a limit on how fast any particular hardware works.
Nostalgia Nerd Whats the theoretical highest clock rate that could ever be achieved as known physics allow them today and what is the highest possible clock rate based on the certain theoretical physical properties that may or may not exist.....kind of like science fiction...like is it theoretically possible something might be faster than light could be used instead of electricity. Could you have electrons travel shorter distances in or outside the cpu using space bending things like tiny wormholes and such....also how small could transistors get using the same logic? What about cpi?.....how high can that get?...sorry for all the questions I just really want to know.
Maybe you could do a future episode on Charles Babbage's machine, one of the first mechanical computers in the 19th Century. linguapress.com/advanced/babbage.htm en.wikipedia.org/wiki/Charles_Babbage#Analytical_Engine
@@jawuku3885 An episode on unit record equipment (one of the predecessors to computers) would also be cool, if you could manage to get your hands on preserved working copies.
12 years ago CPU's hit 3GHz. Today, we've barely gone over 4GHz. Lots of cores and efficiency in today's chips but I think we've hit the clock limit. We'll be stuck with 4-5Ghz chips for the next decade, at least, and only in high-end chips. Above 5GHz and the amount of heat generated is not worth the minimal performance gains compared to multi-cores and modern instruction sets.
If you look at overclocking, the record is something like 8.7 ghz. Ironically, the highest speeds are accomplished with cpu's specifically from 2006 and 2009. The best anyon's done with a more recent cpu is about 7.5 ghz. That sounds impressive, but who knows what the cooling looked like. (liquid nitrogen most likely - which, while effective, boils off in a matter of minutes, meaning you have to manually keep adding more, or risk destroying your computer. - great for showing off, not so impressive for actually trying to use the system) Aside from that, power use is a concern. There were cpu's that you could reasonably overclock by 50-60% or more of their designated speed and it'd be quite usable, with a good cooler, but power usage could double or triple to get that extra performance out of the chip...
Yeah. I recently upgraded from a generation 1 i7 to generation 5 i5. Even with less threads the new instructions and features give outstanding results in heavy tasks like video encoding or emulation compared to my old chip. I think in last decade still these upgrades also came with boosts to clocks, though also boosted clocks were back then significant, but now we have moved to other areas to get significant performance boosts and the clock rate is less significant really. There is still room for better and better optimization of multiple cores too, as it is not yet all encompassing especially in gaming.
diablomix > Above 5GHz and the amount of heat generated is not worth the minimal > performance gains compared to multi-cores and modern instruction sets. Because of Amdahl's Law single core performance is king. If possible, a 5 ghz cpu is much better than 2 with 2.5 ghz. And much much better than 5 with 1ghz. Even if you could parallelize 95% of your program (which you can't).
Putting more cores on the die is not the answer, just look at 3DFx. We pretty much reached the limits with todays technology. We can only shrink production size so much before the electrons start jumping across the circuits. So the next step will be either biological or quantum.
Coldie - Note quite, its a tad bit (lot) more complicated than that. That has to do with the properties of silicon, that we are nearly at the limit of. Conenion - it is not only possible to parallelize 95% of a programs compute work but it is done at nearly 100% (theoretical max of ~99.99999998% on Titan) on supercomputers. For most most desktop applications what you said does hold true, because OpenGL and most other graphics librarys must operate on a single thread...
I typed in Clockwork Computer and got you talking about electronic computer internal clocks. Were are my cogs and gears very literal Clockwork Computer with mechanical battery?
Then we have graphics memory technology, which often runs at a far higher effective clock than a CPU. On a modern GPU such as the GTX 1070, for example, the 8GB of internal GDDR5 memory uses a quadrupler to essentially push an 8GHz effective clock. Some models will let you overclock this to over 9GHz relatively easily.
Its not true. The transfer rate of a GDDR5 memory is not the same as the frequency. For DDR memory it transfer two times for every clock. For marketing reasons they there for use the frequency of the data rate in steed of the clock. GDDR4 and GDDR5 transfers one level for ever flank, it also transfers two bits for every level, making the bit transfer rate 4 times that of the clock frequency. So if it says 4Ghz, its really 1 Ghz, not 4. This is really marketing bullshit
I think you're getting too conceptual about a simple problem. The circuitry inside the electronics can only change states so quickly, before it may malfunction. The clock can never have and interval beyond or at this point. It may just glitch, in some cases a higher clock can even work, but generate hazardous heat. Great to see a utube channel without monetisation, can watch all these without interruption.
May also ad that a modern CPU don´t run from a central clock any more. Most CPU run on at least one clock per core, and quite a few have a clock for every pipe. The most modern CPU have also switches to inhibit the clock for every step of every pipe making it possible to suspend operations of every step of every pipeline independently. Making the whole processor take quite a bit less power. Also, the clocks is generally also out of synk for different part of the CPU: The reason for this is that for every clock puls the cpu initially use a bit more current, then less and less to the end of the clock pulse. This makes the voltage drop making the CPU demand more voltage. By having the clocks asynchronous this problem can be mitigated.
Coxswain's don't command the timing of a boat. They shout commands of when to start and stop and they steer the boat. It is up to the rowers themselves to keep in time. However the coxswain will shout of someone is out of time. Even if it is a fraction of a second.
Okay....what about virtual cores? These I do not understand at all and are well above my pay grade. How can something that is not there do what it does?
The name of the time setter is coxswain. The etymology of the word gives a literal meaning of "boat servant" since it comes from cock, a cockboat or other small vessel kept aboard a ship, and swain, an Old English term derived from the Old Norse sveinn meaning boy or servant, according to Wikipedia.
5:02 i know this is an old video, and not worth correcting, but this is actually false. Transistor speed does not match clock speeds. Clock speeds are actually determined by the propagation delay of a cycle ( the longest stage in a pipelined design). It's actually why increasing pipeline length results in higher clock speeds.
Actually, its a Milliard in the UK. Billion means Million Million. American numerics go up by thousands, Billion being thousand million, trillion being million million, and so on. In the UK it used to be Million, Milliard (thousand million), Billion (Million Million), Billiard (Thousand Million Million), Trillion (Million Million Million), Trilliard (Thousand Million Million Million.) Essentially just addressing the number of zero's. So Thousand Million isn't incorrect and saying otherwise makes you a pedantic asshole.
I assume that was a joke. ;p if not, the definition of a billion has varied by country, but for cpu's, it's always 1000 million. (also, 1000 million for anything that uses the metric system)
+mipmipmipmipmip completely agree. The term in British English has been hijacked since 2008 to mean a thousand million. I have no problem with that, guess it means all English speakers are referring to the same value now :)
MIPS matter a lot more than the clock speed of a chip. There's a reason why GPUs and supercomputers aren't rated around clock speeds, but around MIPS/BIPS and FLOPS. Because ultimately, the clock speed doesn't matter if you can get more done per-tick.
![Why Intel Stopped Using Processor Numbers [Byte Size] | Nostalgia Nerd](/img/1.gif)
man I remember having to calculate cycles, back when I use to do assembly programming on my amiga and MSX. As fun as this was, I'm glad we don't have to worry about things like that anymore :)
I was looking at some of my old 6510 assembler code earlier this evening, it has comments added to each instruction to describe the cycles used, including things like overlapping cycles (6502 and its derivates have a simple pipeline) and special conditions (crossing page boundaries requiring an extra cycle, 'stolen' cycles due to video or dma etc). The code is intended to run in cycle exact lock step with the painting of the screen (actually writing directly to the dac to implement a kind of software video controller). Like you, I'm glad to not have to do that anymore, but it was fun to do, and fascinating to read back (how on earth did I have all the info for that before the internet age.. much of it gleamed from diagrams and lots of experimentation)
One of the things people miss: The signal is electricity. All the actions are powered by this. The digital signal is really high voltage-low voltage-high voltage...etc. The electronics are designed to do something when the voltage goes up, being powered by the voltage jump. They then drop into their new state when the voltage drops. They then do the next thing when the voltage jumps up again. The clock is more than just something to synchronize work, it powers it. The timing is critical, since the pulse creates a cascade of work, which all has to complete before the next pulse. That is why you have a limit on how fast any particular hardware works.
Cheers lad, you tackled a very complicated subject like a champ.
You're most welcome bonny chap
Nostalgia Nerd Whats the theoretical highest clock rate that could ever be achieved as known physics allow them today and what is the highest possible clock rate based on the certain theoretical physical properties that may or may not exist.....kind of like science fiction...like is it theoretically possible something might be faster than light could be used instead of electricity. Could you have electrons travel shorter distances in or outside the cpu using space bending things like tiny wormholes and such....also how small could transistors get using the same logic? What about cpi?.....how high can that get?...sorry for all the questions I just really want to know.
@@jet100a Those would be questions for a physicist.
That's a lot of info crammed into one video but put together really well. No one can say they didn't learn something.
Really concise and easy to follow explanation on topic rarely explained. Very well done sir!
Very good! I particularly enjoyed you almost losing it during, "stroke! stroke! stroke!"
I thought you were gonna show us a computer that was actually made with clockworks or something but... This is just as good, and very informative!
Wellllll, I did consider that title may be misconstrued as that. But I thought, what the hell. It still works. Also, thank you muchly :)
That'd be something.
Maybe you could do a future episode on Charles Babbage's machine, one of the first mechanical computers in the 19th Century.
linguapress.com/advanced/babbage.htm
en.wikipedia.org/wiki/Charles_Babbage#Analytical_Engine
@@jawuku3885 An episode on unit record equipment (one of the predecessors to computers) would also be cool, if you could manage to get your hands on preserved working copies.
love your documentary like videos. your a great story teller
Thanks very much. I really appreciate that :)
You stopped yourself from laughing when saying "stroke, stroke, stroke". I know what you were thinking about
This makes me remeber my teenage years and my overclock mania at the time. Great vid
Excellent explanation and video!
Glad you enjoyed it :)
12 years ago CPU's hit 3GHz. Today, we've barely gone over 4GHz. Lots of cores and efficiency in today's chips but I think we've hit the clock limit. We'll be stuck with 4-5Ghz chips for the next decade, at least, and only in high-end chips. Above 5GHz and the amount of heat generated is not worth the minimal performance gains compared to multi-cores and modern instruction sets.
If you look at overclocking, the record is something like 8.7 ghz.
Ironically, the highest speeds are accomplished with cpu's specifically from 2006 and 2009.
The best anyon's done with a more recent cpu is about 7.5 ghz.
That sounds impressive, but who knows what the cooling looked like. (liquid nitrogen most likely - which, while effective, boils off in a matter of minutes, meaning you have to manually keep adding more, or risk destroying your computer. - great for showing off, not so impressive for actually trying to use the system)
Aside from that, power use is a concern. There were cpu's that you could reasonably overclock by 50-60% or more of their designated speed and it'd be quite usable, with a good cooler, but power usage could double or triple to get that extra performance out of the chip...
Yeah. I recently upgraded from a generation 1 i7 to generation 5 i5. Even with less threads the new instructions and features give outstanding results in heavy tasks like video encoding or emulation compared to my old chip. I think in last decade still these upgrades also came with boosts to clocks, though also boosted clocks were back then significant, but now we have moved to other areas to get significant performance boosts and the clock rate is less significant really.
There is still room for better and better optimization of multiple cores too, as it is not yet all encompassing especially in gaming.
diablomix
> Above 5GHz and the amount of heat generated is not worth the minimal
> performance gains compared to multi-cores and modern instruction sets.
Because of Amdahl's Law single core performance is king. If possible, a 5 ghz cpu is
much better than 2 with 2.5 ghz. And much much better than 5 with 1ghz.
Even if you could parallelize 95% of your program (which you can't).
Putting more cores on the die is not the answer, just look at 3DFx. We pretty much reached the limits with todays technology. We can only shrink production size so much before the electrons start jumping across the circuits. So the next step will be either biological or quantum.
Coldie - Note quite, its a tad bit (lot) more complicated than that. That has to do with the properties of silicon, that we are nearly at the limit of.
Conenion - it is not only possible to parallelize 95% of a programs compute work but it is done at nearly 100% (theoretical max of ~99.99999998% on Titan) on supercomputers. For most most desktop applications what you said does hold true, because OpenGL and most other graphics librarys must operate on a single thread...
Would have liked to see an explanation of how the clock multiplier circuits work -- especially the non-integer varieties.
I typed in Clockwork Computer and got you talking about electronic computer internal clocks. Were are my cogs and gears very literal Clockwork Computer with mechanical battery?
this was actually a pretty good summary
Please, don't sound too surprised ;)
As always a pleasure to watch :)
A Bad Influence, a Games Master, and a heady Byte-Size that I actually learned a few things from?
Banner week, old man.
Glad to be of service.
Loved the vid man. No idea how you get the time.
Not sure, but I'm lacking sleep.
Lots of new information here. Thanks.
The ultimate shutdown to someone goes "time is an illusion!" "Yeah, then how do computers work?"
Haha
Excellent explanation and great video
Thanks very much :)
Then we have graphics memory technology, which often runs at a far higher effective clock than a CPU. On a modern GPU such as the GTX 1070, for example, the 8GB of internal GDDR5 memory uses a quadrupler to essentially push an 8GHz effective clock. Some models will let you overclock this to over 9GHz relatively easily.
Its not true. The transfer rate of a GDDR5 memory is not the same as the frequency.
For DDR memory it transfer two times for every clock. For marketing reasons they there for use the frequency of the data rate in steed of the clock.
GDDR4 and GDDR5 transfers one level for ever flank, it also transfers two bits for every level, making the bit transfer rate 4 times that of the clock frequency. So if it says 4Ghz, its really 1 Ghz, not 4. This is really marketing bullshit
I used to work for Cow Spotted Computers in the 90s. I forgot how horrible those Intel ads were. Thanks?
I think you're getting too conceptual about a simple problem. The circuitry inside the electronics can only change states so quickly, before it may malfunction. The clock can never have and interval beyond or at this point. It may just glitch, in some cases a higher clock can even work, but generate hazardous heat. Great to see a utube channel without monetisation, can watch all these without interruption.
I noticed that you showed the Zuse Z3 when you mentioned the ENIAC computer.
You are correct Sir. But they're both pretty similar looking.
Pretty interesting. But why are you showing a Zuse Z3 while talking about the ENIAC?
André Reichelt E&O
Surprisingly the DX4 does actually not quadruple the clock speed. It rather tipples it. Why it was called the DX4 i don´t know.
May also ad that a modern CPU don´t run from a central clock any more. Most CPU run on at least one clock per core, and quite a few have a clock for every pipe. The most modern CPU have also switches to inhibit the clock for every step of every pipe making it possible to suspend operations of every step of every pipeline independently. Making the whole processor take quite a bit less power.
Also, the clocks is generally also out of synk for different part of the CPU: The reason for this is that for every clock puls the cpu initially use a bit more current, then less and less to the end of the clock pulse. This makes the voltage drop making the CPU demand more voltage. By having the clocks asynchronous this problem can be mitigated.
the background music is really loud I can barley hear you
Adjust your equalization up at around the 400hz frequency. That should do the trick ;)
It's funny because games like Crysis were expecting things like 8 ghz processors to become the norm :D
Awesome vid! thnks :D
Nice
So Clock Doubling X Dancing in a Colorful Shiny Space Suit = Task Speed Benchmarking?
Coxswain's don't command the timing of a boat. They shout commands of when to start and stop and they steer the boat. It is up to the rowers themselves to keep in time. However the coxswain will shout of someone is out of time. Even if it is a fraction of a second.
Freakin amazing! :)
That's not the ENIAC at 6:30. It's a Zuse Z3. Or the Zuse Z3, to be a little more precise.
Okay....what about virtual cores? These I do not understand at all and are well above my pay grade. How can something that is not there do what it does?
Wait, WHAT? The guy on the boat... actually shouts "stroke"... and his name is cockson?
...glorious
The name of the time setter is coxswain. The etymology of the word gives a literal meaning of "boat servant" since it comes from cock, a cockboat or other small vessel kept aboard a ship, and swain, an Old English term derived from the Old Norse sveinn meaning boy or servant, according to Wikipedia.
aMAZEing!
i just stroked my Coxswain to the beat of the great music in this video.
5:02 i know this is an old video, and not worth correcting, but this is actually false.
Transistor speed does not match clock speeds. Clock speeds are actually determined by the propagation delay of a cycle ( the longest stage in a pipelined design).
It's actually why increasing pipeline length results in higher clock speeds.
I can't remember what I said, as you're right, it's pretty old. I'll double check.
"one thousand million" is A FUCKING BILLION
Chill out Emps
Whatever happened to the good old British billion where it was a million millions? :p
Actually, its a Milliard in the UK. Billion means Million Million. American numerics go up by thousands, Billion being thousand million, trillion being million million, and so on. In the UK it used to be Million, Milliard (thousand million), Billion (Million Million), Billiard (Thousand Million Million), Trillion (Million Million Million), Trilliard (Thousand Million Million Million.) Essentially just addressing the number of zero's. So Thousand Million isn't incorrect and saying otherwise makes you a pedantic asshole.
what is and where did you get the video at 5m34sec?
5:34
6:05
Why not just say "100 billion?"
Only really since the 2008 crash. Before a billion was a million million
I assume that was a joke. ;p
if not, the definition of a billion has varied by country, but for cpu's, it's always 1000 million. (also, 1000 million for anything that uses the metric system)
nope
+mipmipmipmipmip completely agree. The term in British English has been hijacked since 2008 to mean a thousand million. I have no problem with that, guess it means all English speakers are referring to the same value now :)
Correct. Hence the reason I went with 1,000 million. :)
The Brit's clumsy naming for large numbers is frustrating.
Has anyone ever built a steampunk computer? Itd be an interesting thing to attempt anyway
Next up photonic integrated circuit
So that’s why it’s called overclocking
a thousand million...do you mean a billion?
i wish i could overclock my cpu to 4,77 ghz
why?
its a Nerdy thing, like the original cpu ran at 4.77 mhz , i wish i could run my cpu at 4,77 ghz. You see 4,77 mhz .. 4,77 ghz :P
Put your BCLK to 99.375 MHz and your multiplier to 48. (101.49 and 47 also works)
or FSB to 530 and multi to 9. (or 9.5x502)
I feel cleverer now
ClevererER*
0:39 It will if you fuck with vCore! :P
Ghz is obs-elite in 2017, sos.
So many more views then pinball dreams ?
2:25 Not really. The term "Hertz" was only made official in 1960, and only really came into prominence in the subsequent decades.
"mean progress" lol
Downplaying MIPS? Seriously? Way to take credibility and chuck it out the window.
I didn't realise you and MiPs were so close. I do apologise.
MIPS matter a lot more than the clock speed of a chip. There's a reason why GPUs and supercomputers aren't rated around clock speeds, but around MIPS/BIPS and FLOPS. Because ultimately, the clock speed doesn't matter if you can get more done per-tick.
+Sable Drakon (Resistance L8) Correct. Essentially what I said. It measures task speed.
Which is a lot more important these days, since CPUs have been trapped at the 3.5-4GHz threshold for nearly a decade.
+Sable Drakon (Resistance L8) Indeed. But this video is about clock speed, not the finicky details of how your preferred tasks take to complete.
My watch is mechanical..........just saying...........
I thought this was going to be about Antikythera mechanism
6969 views
...