Imagine memory like a builder with tools. Hard drive is the shop that sells tools. Lots of tools to choose from but a lot slower as he'd have to drive to get them. RAM is like his van. Smaller in size, but a lot faster for him to access and do his work. L3 cache is like his tool belt. Right on him, very small range, but incredibly quick to access, barely slows him down at all. L1 & L2 cache are like the tools in his hand. Even more limited, generally only one tool at a time, but incredibly quick to use as he doesn't need to even grab them, they are stored right there ready for immediate use.
Registers are also a type of memory and they are even faster! But very small at the same time. (512 byte per core on Intel) As the processor can only process data that are stored in registers (needs to be loaded from RAM to cache and from cache to register), organizing cache hierarchy is one of the most important things apart from the actual processing.
To put it in perspective, more CPU cache means CPU can bite more information from RAM, means more data will be done in same amount of time. Manufactures usually scale CPU caches with a CPU power. For example, cracking ZIP password with brute force attack is usually based on true CPU power, his cores and caches, so... T9400 are Intel mobile CPU for laptops, it got 2.66GHz speed, got 2 cores, L2 cache of 6MB as 24-ways, using 35W. Phenom II X6 1100T are AMD desktop CPU, it got 3.3-3.7GHz, got 6 cores, L3 cache of 6MB as 48 way set, using 125W. So, AMD got like 2 times more GHz, 2 times more transistors, 3 times more cores, additional cache level thats supposed to organize data better and make it faster, and its freaking desktop CPU, but, they both cost same amount of money, both got 6MB of cache, both are 45nm, etc., and tests shows they both have same power, around 27 000 000 passwords per second. So keep in mind that CPU cache is one additionally factor in processing, bigger cache are like bigger mouth, the more you can bite, the more you can process and eat in same amount of time.
@@zjp735 Oh, its been 5 years, but thank you. 👍 Also noticed i didnt say that "bite" are referring to cache size bite from the RAM, and "eat" to CPU cores/threads/speed. Means the biggest "cache bite" of a eg. 96MB would be totally useless (energy useless, price useless, speed useless, parts useless) if CPU would processing it with single tooth (like 1 core, single thread, 1Mz), because it would take long before its swallowed. Thats why i think no one should pay attention to cache, specially if its smart shared cache, and it can be used as some kind of a CPU horsepower representation, like how big mouth he got, how much he can bite and process, if he got a big mouth, theres reason why he can eat more, and its set up by engineers who created it. For example 3MB in modern technology can do everything for home/office and even gaming, 6MB can probably do more calculations, bigger bite, less time accessing RAM (RAM is like 20GB/sec speed, while cache are like 20TB/sec, so reading RAM could be considered as "slow down"), if its old, with no new technology, its probably bad for next gen stuff, and if it has 96MB+, thats some kind of a Xeon range that are only needed by servers and ECC RAM or something like that. Also pay attention to "shared", AMD can put big 24MB cache, while at the same price range Intel can have small 6MB, good for ADS, differences are that shared or smart cache actually using single big chip for cache (cheaper) thats divided according to CPU cores (eg. 24MB, 4 cores, means 6MB per core), something like having 4 partitions on single USB drive, it can be bottleneck, its a single drive. While standard cache, eg. 6MB (not smart, not shared) are more expensive then big shared/smart, so if eg. Intel says 4 cores, 6MB cache, it means each core got his own 6MB cache (more expensive, more smaller, less heat per individual cache), means 4x6MB, also 24MB in total for 4 cores, but 4 times the speed (4 drives got more throughput then 1 drive alone), more expensive, and of course "24MB shared" sound better and cheaper then "6MB", or "48MB shared" vs "12MB", thats kind of a ADS to pull buyers who believe in free stuff, "good but cheap", even they know theres no such thing as "bad but expensive" to ask for. But i would not buy CPU watching only cache on its own, as i said, CPU with 6MB are good, but the one with 12MB probably have reasons (like overall performance speed) why its built with that amount of cache, its probably his performance "limit", where having more cache would be useless, and having less would be bottleneck for a given cores.
@@RadiusNightly thanks dude I learn a lot, I been using computer for decades but I never check CPU cache when buying I just check CPU like i5 i7 or i9 the higher the number means better also higher the clock speed the higher computing power.
Nice explanation. You've mixed add a couple concepts there I would like to clear out. First of all, the problem with RAM is not bandwidth but latency. A RAM module can transfer data pretty quickly, but it does spend a fair amount of time since the processor asks for a block of memory until it begins to transfer. That's why cache is essential, it's basically a very low latency memory. Second, most desktop modern processors (and even ARM smartphone processors) have prefetch mechanisms. This is the part you describe as the processor guessing "what will the software need next" (which is an accurate description actually), but you do imply that the algorithms used by prefetch are the culprits of all the cache misses, which is not true. Not everything loaded into the cache is loaded by the prefetch hardware. It's more accurate to think of the cache as more of a "Recently used memory" kind of deal, with prefetch being a nice addition to it (turning into a "Recently used memory + memory we are probably going to use"). Third, by no means the cache (or prefetch) uses "complex algorithms". The whole point of a cache is to be fast, and if it spent too much time thinking what block to get next it would slow down the processor and by the point it had decided what block to load, the processor would already need it. This is why algorithms used in prefetch are very simple (and they work well). Some older processors just prefetched the next block when data was accessed. There are some other more complicated optimizations, of course, but the whole point of them is speed, and simplicity is the key of speed. Sorry for the long comment guys, and nice work Luke!
Short answer: Because the point of it is that RAM is slower than cache memory. Long answer: RAM is slower than cache because- 1 - It's built differently. Cache memory is built purely with transistors which is fast, but expensive. RAM is built with very very small condensers, that take time to "write" and "read", but are cheaper to make, so the technology itself is slower. 2 - You need to re-write it every time you read it, and several times a second. In order to read the bits, you need to discharge the condensers in the memory, which effectively destroys the data each time you read it, so you have to re-write it. If it wasn't enough, they will discharge alone. About 20 times a second. 3 - It's big. RAM's logic is much more complex, as it needs to activate certain memory rows in certain chips only, and connect them to the output, which takes time. The time this process takes is called "Latency", and sometimes it's greater than the time it takes for the data itself to travel to the processor. 4 - It's far away. Yes, in modern processors even the time it takes for the electrons to travel within the chip is significant. The time it would take for the data to travel from the RAM itself to the processor if it had no latency would be greater still than reading cache memory. So the point of cache is to consult RAM as little as possible, because it being slower would mean the processor having to go slower.
I've wondered for quite some time why my Core2Quad had 12MB of L2 cache, while processors like Sandy Bridge touted their cache being L3? Why was there a sudden exodus from CPUs being heavy on L2 cache?
i've always wondered if it would be possible to make an L4 cache, have it be basically as fast as L3 but outside of the processor, in a small LGA or rPGA socket near the processor, with say 600 pins with the pin density of lga2011, but with this dedicated cache socket you could have an al-gore-rhythm that caches, say 1GB of system ram with a 95+% hit rate
Overall a good video. It could have been cool to point out how big of a portion of the total die the cache actually takes up. To me it is cool information that is still accessible to the general public. I understand that you may not necessarily want to talk about paging and how the cpu decides what extra data to grab after a cache miss and so on.
Will you explain this? I'm taking an online course in computers and its so abstract. CPU itself I'm having trouble understanding it's function... It powers computers on and off and transfers information and memory with electricity? Save me.
Cache also helps rendering a lot of youtube videos while multitasking like downloading at the same time.If cache isnt available it will make the game to stop loading to render the youtube video or the opposite.
It's not a question of speed but also a question of latency. Cache memory always has really low latency, hence the incredibly high price of cache memory.
If the CPU has more cache memory inside of it, it could play video games much better, a triple core processor with 20gb of Cache you could save the patch of the game, so the patch of the game can be saved in the cache and it could run 4K without losing frame rates, which is very good idea.
That's the whole compromise, you can access data faster, but have less capacity or less safty like in GDDR and HBM. Also the Cache isn't the fastest memory in Computers, it's the Procesor register...
Actually: In order to make any sort of calculation your CPU has to hold a certain set of numbers inside of it called a "register". Since the signals all mutually depend on each other the processor has a build in memory of it's own. Your CPU wouldn't even work without it since it needs to be able to hold the instructions to process the information it's receiving. It's nearly impossible to use a CPU bus for calculations without using these built in memory caches.
I can't believe you talked about cache without covering set associativity, which is very important in cache design. A 4MB all-way set associative cache will be MUCH MUCH better than an 8MB 2-way associative cache.
How many trillion floating point operations your gpu can processes in one second. Fundamentally 'graphics' are just a bunch of numbers and graphics rendering all comes down to about processing and calculating numbers. The faster you can do it the faster you can render graphics.
My XEONs have 50MB level 3 cache in total, though they are low at frequency, they still perform good in games and they are a lot better when doing rendering and map baking.
One important factor is that if a programmer wants or needs to they can force the computer by some programming trickery to keep things in cache... this is mostly important in scientific or data intensive tasks
Hit rate isn't everything. A really big cache suffers the same problem as system memory. That is, the hit TIME is very slow, so even with a high hit rate, overall it is slow to access. But anyway, large caches aren't needed, as even small caches with a few MBs these days can achieve hit rates of 90%+ due to temporal/spacial locality, prefetching, associativity, and a number of other tricks.
L3 is not the fastest. The ideal would be a huge L1 Cache, as it's the fastest Cache (Memory in registers is even faster and can match CPU speed). But as mentioned, L1 is huge (size wise), it costs a lot (money), and it consumes a lot of energy (Hence it generates a lot of heat). in order to relieve some of that, L2 was implemented. Then L3 and L4. So DRAM feeds L3, L3 to L2, and L2 to L1. The lower the level the faster, but with downsides. The higher the level, the higher memory amount. Intel implemented L4 in some GPU's.
logically u can bt will u pay $1 million for ur CPU? bcoz cache memory is very expensive as it requires flip flop circuits which require larger area and has small memory density as compared to capacitors on dram bt on the positive side cache memory is incredibly fast like L1 cache memory operates at the speed of CPU which means it operates around ~50GB/s bt on downside they are expensive
A simpler (and more accurate) explanation is that your CPU wouldn't be able to work continuously at its full speed if it was always dependent on the system RAM, because the system RAM has latency, which is a time delay between the point where information is requested, and when it becomes available to the CPU. The CPU cache has no latency, and it also runs at CPU speed--so it allows the CPU to work continuously at its full speed, instead of spending most of its time doing nothing while waiting out the latency period of the system RAM. So, the CPU cache's job is to 'hide' the main memory's latency.
It's a shitty Windows 10 PC with missing features to make up for inadequate hardware. Games have the more computationally expensive graphics settings disabled entirely (AA, most anisotropic filtering), and still choke on games a 1060 has no problem with on a real PC.
the SRAM is around 32KB of never changing instruction while the the other part of L1 is 32KB per core for very fast cache of data just used stored to be again for the next operation done as L1 is so small it can go almost as fast if not as fast as the cpu physical core L2 is bigger so it takes longer to search and is used for when the data in L1 is finsihed for the first round of loops around L3 is used as a storage place for what the memory controller thinks the cpu core will need next so fetches it from system RAM and stores it inside L3 to be used
After seeing this, I finally understand the point of optimizing programming code for cache hits, wish I had seen this sooner -_- (oh, and a music group with the name "Ca$h-E", would give "Ca$h-Es Greatest Hits!"... yeah... -_-)
[1:12] is so hilarious and so informative. As a hard working employee that can turn around to bite you. Because you finish your tasks very quickly, only to have boss man catching you sitting doing nothing. If you do labour intensive work they don't care that you did your task. They care that you work all of the hours they pay you for.
When is anyone ever going to do testing to find the REAL benefit of more cache? Usually the explanation of cashe is that "cache is fast memory and makes your processor faster hurr durr", yet nobody has a clue about what the real world difference would be between 6MB of cache and 12MB of cache if the processors were otherwise identical.
the analysis of that would require probably insider knowledge of how Intel's prefetching algorithms and architecture works as well. unlike DRAM or general storage, there is no standard protocol for communicate between the cores and it's cache. the level of optimization could go down to the specific configuration of the cpu (i.e. a different core/clock config will have different optimizations for its cache).
Does cache matter for serious editing and rendering, for example 30+ sampled instruments in Reaper or 10+ tracks in Vegas? Will 8MB be enough or should I go for something like 15-20? Thanks :)
Imagine memory like a builder with tools.
Hard drive is the shop that sells tools. Lots of tools to choose from but a lot slower as he'd have to drive to get them.
RAM is like his van. Smaller in size, but a lot faster for him to access and do his work.
L3 cache is like his tool belt. Right on him, very small range, but incredibly quick to access, barely slows him down at all.
L1 & L2 cache are like the tools in his hand. Even more limited, generally only one tool at a time, but incredibly quick to use as he doesn't need to even grab them, they are stored right there ready for immediate use.
brilliant dude!
Adolph hitler approves
Mr. Waffentrager Literally didn’t even notice xD
This is better than this 4 min video. Thx
HDD is the farthest shop
Ssd is nearer
And m.2 nvme SSD is the nearest shop
Woow, Linus hit puberty hard! He is huge!
that's not linus............... AND HES A GROWN MAN what do you mean puberty XD
lel
that's what she said
Solitude *more manly, you mean* ;)
I don't get it
"Cash with an E"
So you're talking about Cesh?
No, he means Czech /s
Or Cashe
Ke$ha
or case
@@mparagamesit is actually spelt Cache
"Cache with an 'E'"
That is... not the only difference...
cache with an e and cash with an s, does that mean eash and sash?
Cash with an e and cash with an s are mutually exclusive and only refer to a single word each, not a problem
@Arch Linux yeah he should've said with an "h".
Cashe
this actually explains so much because I have never had a clue what the CPUs Cache does and how significant it is.
don't scroll down to the comments its has nothing related to the video subject
thank you ... thank me later ;)
wth you're actually a genius
Thank (hope I'm not too late)
Where can i download more cache ?
You gotta have high quality pixels first
Can I use high quality rips instead?
I'm almost afraid you are serious lol
The process for doing that is very...MESI... :)
You can't download cache! You only can download RAM.
stupid kid
Registers are also a type of memory and they are even faster! But very small at the same time. (512 byte per core on Intel) As the processor can only process data that are stored in registers (needs to be loaded from RAM to cache and from cache to register), organizing cache hierarchy is one of the most important things apart from the actual processing.
Our ICT teacher showed us this today and i got a bit too excited it was a techquickie video
I'm using a hard drive for my cache it's really fast.
I'm using a floppy disk, Works great
CD-RW is best cache.
I use relays and punch cards.
I just upgraded to punch tape. My old cache was a rope with knots in it.
***** I use electrons...
I learned about caching in my CS classes and this was a great simplistic overview of what the cache is/does!
I didn't think this would be that accurate. But you covered all the basic points. You guys are really doing your research. I appreciate that.
SRAM in Polish is literally "I'M SHITTING"
in my language im shitting is "serem"
No tak
I don't have alot of cache but do you take debit?
Do you know if he accepts ebt?
dad joke
@cheese stick man Dad jokes best jokes
you mean "a lot"
@@ubernerd_ -.- yes I mean “a lot”
More cash, more cashe.
+Christopher Tan Well, honestly no. Accidental typo... but let go with that.
+Christopher Tan Let me guess, spelled "spelled" wrong on purpose
+Snoop “MLG” Doge It isn't wrong though... He spelt "spelt" correctly. Spelled/spelt are both correct :)
+Christopher Tan ahahah yeah :)
lol.....................,......,.....................................,...............
To put it in perspective, more CPU cache means CPU can bite more information from RAM, means more data will be done in same amount of time. Manufactures usually scale CPU caches with a CPU power.
For example, cracking ZIP password with brute force attack is usually based on true CPU power, his cores and caches, so...
T9400 are Intel mobile CPU for laptops, it got 2.66GHz speed, got 2 cores, L2 cache of 6MB as 24-ways, using 35W.
Phenom II X6 1100T are AMD desktop CPU, it got 3.3-3.7GHz, got 6 cores, L3 cache of 6MB as 48 way set, using 125W.
So, AMD got like 2 times more GHz, 2 times more transistors, 3 times more cores, additional cache level thats supposed to organize data better and make it faster, and its freaking desktop CPU, but, they both cost same amount of money, both got 6MB of cache, both are 45nm, etc., and tests shows they both have same power, around 27 000 000 passwords per second.
So keep in mind that CPU cache is one additionally factor in processing, bigger cache are like bigger mouth, the more you can bite, the more you can process and eat in same amount of time.
Dude your explanation is much better 👍
@@zjp735 Oh, its been 5 years, but thank you. 👍
Also noticed i didnt say that "bite" are referring to cache size bite from the RAM, and "eat" to CPU cores/threads/speed. Means the biggest "cache bite" of a eg. 96MB would be totally useless (energy useless, price useless, speed useless, parts useless) if CPU would processing it with single tooth (like 1 core, single thread, 1Mz), because it would take long before its swallowed.
Thats why i think no one should pay attention to cache, specially if its smart shared cache, and it can be used as some kind of a CPU horsepower representation, like how big mouth he got, how much he can bite and process, if he got a big mouth, theres reason why he can eat more, and its set up by engineers who created it.
For example 3MB in modern technology can do everything for home/office and even gaming, 6MB can probably do more calculations, bigger bite, less time accessing RAM (RAM is like 20GB/sec speed, while cache are like 20TB/sec, so reading RAM could be considered as "slow down"), if its old, with no new technology, its probably bad for next gen stuff, and if it has 96MB+, thats some kind of a Xeon range that are only needed by servers and ECC RAM or something like that.
Also pay attention to "shared", AMD can put big 24MB cache, while at the same price range Intel can have small 6MB, good for ADS, differences are that shared or smart cache actually using single big chip for cache (cheaper) thats divided according to CPU cores (eg. 24MB, 4 cores, means 6MB per core), something like having 4 partitions on single USB drive, it can be bottleneck, its a single drive. While standard cache, eg. 6MB (not smart, not shared) are more expensive then big shared/smart, so if eg. Intel says 4 cores, 6MB cache, it means each core got his own 6MB cache (more expensive, more smaller, less heat per individual cache), means 4x6MB, also 24MB in total for 4 cores, but 4 times the speed (4 drives got more throughput then 1 drive alone), more expensive, and of course "24MB shared" sound better and cheaper then "6MB", or "48MB shared" vs "12MB", thats kind of a ADS to pull buyers who believe in free stuff, "good but cheap", even they know theres no such thing as "bad but expensive" to ask for.
But i would not buy CPU watching only cache on its own, as i said, CPU with 6MB are good, but the one with 12MB probably have reasons (like overall performance speed) why its built with that amount of cache, its probably his performance "limit", where having more cache would be useless, and having less would be bottleneck for a given cores.
@@RadiusNightly thanks dude I learn a lot, I been using computer for decades but I never check CPU cache when buying I just check CPU like i5 i7 or i9 the higher the number means better also higher the clock speed the higher computing power.
Informative, well paced and with interested editing. This is an example of a perfect tech explanation video! Thanks
My lecturer showed this during our lecture and that was the only time I learned something during her lecturer. Thanks Techquickie!
Played @ 1.25 speed after 5 seconds..
"ah there we go, linus' speed"
😂
Thats better.
HDD
what is the large difference between M.2 SSD and SSD in speed?
@@miltonrazon7838
M.2 SSD is vague, it can be the slower SATA 3 standard or the much faster PCIe specification.
U gotta add register memory (fastest)
Most computers now have more cache than a 1.44MB floppy. My laptop with a Core i5 has 3MB L3 cache.
still doesnt make it any better
mine has 30mb >.
i have 24mb of smart cache
My 2011 dual core i2 has 6mb L2 cache
i2 processors are da best
omg, 2 minutes in and this is already one of the most informative, accurate, and concise videos I've ever found on this topic.
"Even more better"
Even betterer
Even superior betterment
Nice explanation. You've mixed add a couple concepts there I would like to clear out. First of all, the problem with RAM is not bandwidth but latency. A RAM module can transfer data pretty quickly, but it does spend a fair amount of time since the processor asks for a block of memory until it begins to transfer. That's why cache is essential, it's basically a very low latency memory.
Second, most desktop modern processors (and even ARM smartphone processors) have prefetch mechanisms. This is the part you describe as the processor guessing "what will the software need next" (which is an accurate description actually), but you do imply that the algorithms used by prefetch are the culprits of all the cache misses, which is not true. Not everything loaded into the cache is loaded by the prefetch hardware. It's more accurate to think of the cache as more of a "Recently used memory" kind of deal, with prefetch being a nice addition to it (turning into a "Recently used memory + memory we are probably going to use").
Third, by no means the cache (or prefetch) uses "complex algorithms". The whole point of a cache is to be fast, and if it spent too much time thinking what block to get next it would slow down the processor and by the point it had decided what block to load, the processor would already need it. This is why algorithms used in prefetch are very simple (and they work well). Some older processors just prefetched the next block when data was accessed. There are some other more complicated optimizations, of course, but the whole point of them is speed, and simplicity is the key of speed.
Sorry for the long comment guys, and nice work Luke!
Short answer: Because the point of it is that RAM is slower than cache memory.
Long answer: RAM is slower than cache because-
1 - It's built differently. Cache memory is built purely with transistors which is fast, but expensive. RAM is built with very very small condensers, that take time to "write" and "read", but are cheaper to make, so the technology itself is slower.
2 - You need to re-write it every time you read it, and several times a second. In order to read the bits, you need to discharge the condensers in the memory, which effectively destroys the data each time you read it, so you have to re-write it. If it wasn't enough, they will discharge alone. About 20 times a second.
3 - It's big. RAM's logic is much more complex, as it needs to activate certain memory rows in certain chips only, and connect them to the output, which takes time. The time this process takes is called "Latency", and sometimes it's greater than the time it takes for the data itself to travel to the processor.
4 - It's far away. Yes, in modern processors even the time it takes for the electrons to travel within the chip is significant. The time it would take for the data to travel from the RAM itself to the processor if it had no latency would be greater still than reading cache memory.
So the point of cache is to consult RAM as little as possible, because it being slower would mean the processor having to go slower.
***** answered you but forgot to mention you ;)
1:40 "I can assure you, MY CACHE IS TREMENDOUS! Believe me, everyone says so" - D.J.T.
puns were bad as always.....10/10 would watch again
Very informative video as always :D
Computerphile did a really good job explaining this, but you guys made it way cooler!
I've wondered for quite some time why my Core2Quad had 12MB of L2 cache, while processors like Sandy Bridge touted their cache being L3?
Why was there a sudden exodus from CPUs being heavy on L2 cache?
techquickie has the best comment section!
Cache is pronounced as Cash? I thought it was pronounced cash-aye.
Raidenx147 Have you say, I haven't head your version before. I've only heard "cash" and "cay-sh"
That is the french pronunciation like 'forte'.
ikr, I need to learn a different language, english is too messed up
Caché
@@WikiForce Same
Love how they put a picture of oldschool DIMMs instead of DDRs when they mentioned RAM. Hadn't seen those in a long time.
i've always wondered if it would be possible to make an L4 cache, have it be basically as fast as L3 but outside of the processor, in a small LGA or rPGA socket near the processor, with say 600 pins with the pin density of lga2011, but with this dedicated cache socket you could have an al-gore-rhythm that caches, say 1GB of system ram with a 95+% hit rate
Very fast and vivid explanation, nice that someone like faster speed in technology, perhaps it increases a better money-stream, THANKS.
U WOT M8?
U WOT M8” is a shorthand for “You what, mate?
Overall a good video. It could have been cool to point out how big of a portion of the total die the cache actually takes up. To me it is cool information that is still accessible to the general public. I understand that you may not necessarily want to talk about paging and how the cpu decides what extra data to grab after a cache miss and so on.
Will you explain this? I'm taking an online course in computers and its so abstract. CPU itself I'm having trouble understanding it's function... It powers computers on and off and transfers information and memory with electricity? Save me.
Would you like to say how you dont need hyperthreading now?
Do not confuse the terms 'memory' and 'storage'. Do not refer to a HD as a 'memory'.
I'm out of cache. How do I download more? Will my 5kbit internet be fast enough?
i downloaded some cache from 4chan with my 2bit connection you're set
Sure, but if you find a high quality site, you're gonna want to download faster internet, shouldn't take long though!
Make a vid about ram timings
Why does Luke look like the Hulk?
try to say that 3 times reaallly fast ;)
1. Why does Luke look like the Hulk?
2. Why does Luke look like the Hulk?
3.Why does Luke look like the Hulk?
Anything else you want me to do?
+Victor Parker Gfys
+You Nes done.
Anything else you want me to do?
+Victor Parker That's just sad, Bro. It was just a joke...
Cache also helps rendering a lot of youtube videos while multitasking like downloading at the same time.If cache isnt available it will make the game to stop loading to render the youtube video or the opposite.
What about creating a level 4 cache with HBM and abandon DDR4? Wouldnt that be a nice boost?
Yep
Chuckiele genius
It's not a question of speed but also a question of latency. Cache memory always has really low latency, hence the incredibly high price of cache memory.
The editing job on this video is marvelous though.
I hear cache and think CS GO
...
Why
@@WonderfulBoness It's a map name in CSGO.
@@chrisbrummond4832 cool
@@mami1330 letter from Charles de almond butter what did I do to your stronger
If the CPU has more cache memory inside of it, it could play video games much better, a triple core processor with 20gb of Cache you could save the patch of the game, so the patch of the game can be saved in the cache and it could run 4K without losing frame rates, which is very good idea.
Luke is getting quite the belly...
He's got the dad bod going
Hahahaha nice
Me too
Hmm, thought he looked skinnier. What the heck are we talkin' about we are men! Leave the vanity to the women brah!
I thought he was looking slimmer.
Haha I remember they showed us this video in 9th grade (UK)
And the teacher said after “alright that’s a cpu cache explained by a cool Canadian fella”
We need cache RAM disks (Cache-disks?)!
Imagine the possibilities: Slowing down your CPU for having an ultra fast floppy disk XD
Not big enough to store anything we have nowadays.
+ki2ne - Well, it could be. It would just be prohibitively expensive...
+Robert Faber That too.
I had the exact same thought. Clearly this storage would be super fast, but it's only a few MB. You won't be able to store much data.
That's the whole compromise, you can access data faster, but have less capacity or less safty like in GDDR and HBM.
Also the Cache isn't the fastest memory in Computers, it's the Procesor register...
Wow, baby Luke. So young, so full of hope.
casually helping me pass my GCSE whoop whoop
Actually: In order to make any sort of calculation your CPU has to hold a certain set of numbers inside of it called a "register".
Since the signals all mutually depend on each other the processor has a build in memory of it's own. Your CPU wouldn't even work without it since it needs to be able to hold the instructions to process the information it's receiving.
It's nearly impossible to use a CPU bus for calculations without using these built in memory caches.
I use an abacus as cache
Can I please have a job making your hilarious illustrations? They get me every time.
I was expecting Linus lol
I can't believe you talked about cache without covering set associativity, which is very important in cache design. A 4MB all-way set associative cache will be MUCH MUCH better than an 8MB 2-way associative cache.
Teraflops explained as fast as possible?
Search in google bro
How many trillion floating point operations your gpu can processes in one second. Fundamentally 'graphics' are just a bunch of numbers and graphics rendering all comes down to about processing and calculating numbers. The faster you can do it the faster you can render graphics.
My XEONs have 50MB level 3 cache in total, though they are low at frequency, they still perform good in games and they are a lot better when doing rendering and map baking.
Ok I always pronounced it as CPU catch instead of cash...
Good video Luke. Very good description of Cache not Cash.
Cash + e != Cache
LOL really? I have never noticed there was a difference /s
Starmain This guy Reddits.
One important factor is that if a programmer wants or needs to they can force the computer by some programming trickery to keep things in cache... this is mostly important in scientific or data intensive tasks
*POP QUIZ* why do general processors have only 3MB L3 cache... why not just chuck 1GB in caches? Then you get Cache hit rate of like 95%?
Hit rate isn't everything. A really big cache suffers the same problem as system memory. That is, the hit TIME is very slow, so even with a high hit rate, overall it is slow to access. But anyway, large caches aren't needed, as even small caches with a few MBs these days can achieve hit rates of 90%+ due to temporal/spacial locality, prefetching, associativity, and a number of other tricks.
It's expensive and takes up more space per gigabyte. The tiered system was brought in when cpu clock overtook ram clock back in the 80s or something
Zap is right, but also the faster SRAM is also much more expensive.
L3 is not the fastest. The ideal would be a huge L1 Cache, as it's the fastest Cache (Memory in registers is even faster and can match CPU speed). But as mentioned, L1 is huge (size wise), it costs a lot (money), and it consumes a lot of energy (Hence it generates a lot of heat).
in order to relieve some of that, L2 was implemented. Then L3 and L4.
So DRAM feeds L3, L3 to L2, and L2 to L1. The lower the level the faster, but with downsides. The higher the level, the higher memory amount.
Intel implemented L4 in some GPU's.
logically u can bt will u pay $1 million for ur CPU?
bcoz cache memory is very expensive as it requires flip flop circuits which require larger area and has small memory density as compared to capacitors on dram
bt on the positive side cache memory is incredibly fast like
L1 cache memory operates at the speed of CPU which means it operates around ~50GB/s bt on downside they are expensive
A simpler (and more accurate) explanation is that your CPU wouldn't be able to work continuously at its full speed if it was always dependent on the system RAM, because the system RAM has latency, which is a time delay between the point where information is requested, and when it becomes available to the CPU.
The CPU cache has no latency, and it also runs at CPU speed--so it allows the CPU to work continuously at its full speed, instead of spending most of its time doing nothing while waiting out the latency period of the system RAM.
So, the CPU cache's job is to 'hide' the main memory's latency.
Ram name suggestions:
shRam
bRam
tRam
mRam
gRam
fRam
fRam = Fuck Ram
@@Vyrkhan LA QUARANTAINE ALLEMANDE
U missed vram
@@Vyrkhan volume 51
@@sigmaswan2969 ksbbz
CPU cache? More like “Cool information that’s fresh!” Thanks for sharing.
AHOY from the x3D age.
96 megabytes of L3 cache is literally insane man
Thank you for the great video
so does v cache cpu had more latency than non v cache cpu? or it does not matter
Thanks! I'm learning so much from your channel :)
i did 771-775 mod and my Xeon x5450 have 12 mb or l2 cache xD
HookerWithAPenis blsbnlagudu
Thank you, this is helping me in my studies.
Who's Watching this in 2021?
Cache misses can also be avoided by good programming
2019
AMD Ryzen 3000 have up to 64mb cache
Whole lot than Intel offer
But still worse than intel in gaming.
@@jamesisaac7684
Aah, nice joke
core2 q9550 has 12mb cache. Still a beast 4ghz (clocked) processor even my today's standards.
Claim your “here before a million views” ticket here
While watching this video I got an ad starring Linus about pulseway. Not realizing it was an ad I kept watching. Only to realize it was an ad
Explain the new xbox one *S*... I bet u can't
you*
Navindu Amarakoon 😂
It's a shitty Windows 10 PC with missing features to make up for inadequate hardware. Games have the more computationally expensive graphics settings disabled entirely (AA, most anisotropic filtering), and still choke on games a 1060 has no problem with on a real PC.
oof, u right
Dudes, the new Ryzen 3900x! 64 MB of L3!!!!! Dwarf Fortress will never lag again.
what is like to see in a future video is a video on OSVR and how it stacks up against steam VR and occulus.
damn techquickie almost at 1M subs!
Thank you for taking time to explain this although I must say I got nothing
There was an ad before this video and the endorser is linus
Hey bro, I really like that Pocket animation.
I was wondering about this earlier, I get home from school and bam answers right in my sub box
2x 55MB L3 should be enough?
да.
3:42 Speaking of cash.... Come on, Luke, it's such an obvious segue!
the SRAM is around 32KB of never changing instruction while the the other part of L1 is 32KB per core for very fast cache of data just used stored to be again for the next operation done
as L1 is so small it can go almost as fast if not as fast as the cpu physical core
L2 is bigger so it takes longer to search and is used for when the data in L1 is finsihed for the first round of loops around
L3 is used as a storage place for what the memory controller thinks the cpu core will need next so fetches it from system RAM and stores it inside L3 to be used
After seeing this, I finally understand the point of optimizing programming code for cache hits, wish I had seen this sooner -_-
(oh, and a music group with the name "Ca$h-E", would give "Ca$h-Es Greatest Hits!"... yeah... -_-)
Just wandering, why is techquickie still on 1080p while LTT is way above that?
Barely knowing much about computer science, are registers and stack data stored in CPU cache?
[1:12] is so hilarious and so informative. As a hard working employee that can turn around to bite you. Because you finish your tasks very quickly, only to have boss man catching you sitting doing nothing. If you do labour intensive work they don't care that you did your task. They care that you work all of the hours they pay you for.
Man Luke was thin two years back.
Oh, the memories.
that's all interesting but one question, can we store arms there?
When is anyone ever going to do testing to find the REAL benefit of more cache?
Usually the explanation of cashe is that "cache is fast memory and makes your processor faster hurr durr", yet nobody has a clue about what the real world difference would be between 6MB of cache and 12MB of cache if the processors were otherwise identical.
the analysis of that would require probably insider knowledge of how Intel's prefetching algorithms and architecture works as well. unlike DRAM or general storage, there is no standard protocol for communicate between the cores and it's cache. the level of optimization could go down to the specific configuration of the cpu (i.e. a different core/clock config will have different optimizations for its cache).
Does cache matter for serious editing and rendering, for example 30+ sampled instruments in Reaper or 10+ tracks in Vegas? Will 8MB be enough or should I go for something like 15-20? Thanks :)
Man, I wish I had the Cach for a new CPU.
Ok sir
typically you would want as much cache as possible if you're a 3D modeling animator, blender uses a ton of processing power the bigger the project
Luke: "...Or your hard drives, or SSDs"
Hybrid storage: *"Am I a joke to you?"*