If 99% of data is processed in either L1 or L2 cache, is there any real speed increase with lots of main memoryRAM?

[Ghost]

I've browsed 2 different pc upgrade/repair manuals recently. Both have mentioned that most of the data manipulation occurs between the CPU and either the L1 or L2 cache, and a very small percentage actually hits the main memory RAM.

The one book stated that L1 cache handles about 90 percent of the activity and the L2 cache handles about 9 percent of the remaining. This leaves about 1 percent of the data to be processed with RAM.

Therefore, is having 128meg+ or 256meg+ or 512meg+ of RAM really beneficial?

 

[Vinny N]

Well, there obviously is a need for memory, otherwise there wouldn't be such a thing as memory requirements for running applications/games.

100% of that data has to be pulled from the hard drive, then to main memory. Without main memory as a bridge/buffer, it would seem to take "forever"(at least to the CPU) to retrieve any data for l1/l2 cache.

As for how seriously beneficial, it depends on what your doing, the most commonly cited uses are for manipulation of large data, be it sound, still images, or motion video.

It'd be nice to a have a processor with gigs of cache

 

[Hector13]

99% That sounds unbelievably high to me.

I am by no means an expert on cache memory, but I can't see how 99% is possible.
How does the data get into cache in the first place?
I can maybe believe that data is transfered from RAM to cache first whenever it is needed and then the CPU manipulates the data in cache. But that does not mean that RAM memory was not used.
This would make sense as cache is supposed to hold whatever data has been recently used right?
So I am assuming that when data is needed, the cpu looks in Cache then RAM then hard drive.
I doubt that 99% of the time the data that the cpu needs is in the cache already. The books probably mean that 99% of the time the data is retrieved from whereever it is and copied into cache to be operated on.

Of course, like I said, I am NOT very familiar with how cache works.

 

[Soccerman]

yeah first the DATA has to get to the Cache, so if the CPU is chewing on a 2 meg photo, the cache has the data needed immediately, and the RAM feeds the cache with chunks of the photo that will be needed later. of course the CPU can work through all it's data faster then RAM can feed it, so cache is used alot to help keep the CPU going at all times.

 

[Viper GTS]

L1 & L2 handle applications, while main memory takes the rest of the data. The actual executable code that the CPU works with is in it's cache, but it uses the RAM as a temporary staging area between fixed storage & it's cache. Even the slowest RAM will kick ass on the fastest hard drive, so yes, there is definitely a need for RAM. I know I noticed the difference going from 64 to 128, & then from 128 to 192.

Viper GTS

 

[Scifione]

Memory is important but just not quit as important as cache. Remember cache is a faster memory that can pass data to the CPU faster than main memory. You can have a Win98 computer with 32MB of RAM and it will run games and Office. There is a difference in performance when you have 32, 64, 128, and 256 of RAM. Some time 512MB will make an improvement (ex. Adobe Photoshop Pro). This performance increase starts of good but starts to level off with greater amounts of RAM. MHz makes a difference. 66Mhz , 100Mhz, 133Mhz are examples of what you will see today, but you will have to have a PC/motherboard that will run the memory at that speed. Latency is another factor. Cas2 is Better than Cas3.

There is great volumes of articles on the net about memory, but I don't think one of them address all the factors of memory and performance. I did not list everything above, just a big start.

Signature of the Day:
There's just one kind of man you can trust. That's a deadman and a beloved patriot like me.
From: Move: "Gunfight at Red Sands" Song: "A beloved patriot Like Me"

 

[ltk007]

I wonder if there will ever be 128 meg caches.

 

[Mday]

there will be ltk007

but we won't be able to afford the first ones.

right now SRAM is very expensive. when the pentiums had the cache on the mobo, it was like $50 or something for 512KB. So imagine $100 per MB, which means...

and i am thinking hard to manufacturer properly or something, otherwise we'd have many, and intel and amd would not have had units with large onchip L2 whose L2 fail.

 

[Scifione]

I have noticed that the cost of motherboards has increaced since on chip L2 cache. They may be harder to design and manufacture, but buyer demand has to play a role in the prices. Look! Just Look at the prices ! $100 for some and $140 for another and $170 for another. Same form factor and a great range of prices. Well I am getting into another Toppic so time for a new tread.

 

[Scifione]

Oh ya Ghost, that is not very tactful posting on a BBS (Cyberspace) and mentioning something about hardcopy. Hardcopy, YUCK! Just kidding about the hardcopy. You can also find good info on the web.

Scifione's Signature of the Day:
There's just one kind of man you can trust. That's a deadman and a beloved patriot like me.
From: Move: "Gunfight at Red Sands" Song: "A beloved patriot Like Me"

 

[OS]

99% of data is not processed by L1, L2 cache. That number is just flat out wrong.

 

[DaddyG]

I believe that 90% is closer to the actual figure. BTW L1 and L2 are not filled with applications. L1 on both AMD and Intel cpus is split into Instruction and Data in equal parts. L2 has both instructions and data. By fetching a block of data from main memory to cache the cache controller attempts to keep the processor fed. Programs tend to process data in a sequencial manner therefore the cache will normally have the data after the first miss. Instructions process sequencially except when a Branch occurs, Branch prediction helps ensure that the next instruction is available.

 

[PowerJoe]

For a sense of scale:

L1 cache - 1 cycle latenct
L2 cache - 1-4 cycles
PC100 RAM - 30-60 cycles
Fastest HD (assuming 5 nsec seek) - 4,000,000 cycles

-PJ