OPEN DISCUSSION

Explain why cache memory uses DRAM instead of SRAM as SRAM is several times faster than DRAM?.
I would like all of you to participate in the discussion and submit your views.
Thanks!
*~*Susi*~*

 

Hi Susi,

Let's see... SRAM imposes no load on the system for refreshes, is significantly faster... So, then we use DRAM for cache because it's cheaper!. It slows the system and performs poorly, but the cost is less.

Always look for the botton line.

 

Hello,

Just in case that you want to go a bit more in deep, this link has very straight forward explanation about memories.

http://www.corsairmicro.com/memory_basics/...3707/index.html

Kind regards.

 

Hi,

Thanks for the link, Useless, but I'm on a slow rural modem, and can't spend the 8 or so hours necessary to do a slide show. I volunteer at a school, so maybe I can get to it in a few weeks on a shared T1 line.

I'll still maintain that the bottom line is a major determinant in selecting what goes into a conputer. On the other hand, the time and effort put into branch prediction and other strategies that attempt to maximize cache performance makes you wonder.

And we never get to see the other side of the coin - the system running without cache memory. Modern memory is so much faster, as is the frontside bus. Could processors get cheaper and do well with no attempt made at caching blocks of memory? Would better compilers do as much?

 

Hi beenthere,

About the link, you are more than welcome.

Now about the economical reasons for choosing components, well, I totally agree with you.

But about a processor without cache, I am afraid that I cannot agree with you, the reason for it, is, memory latency.
Please correct me if I am wrong, as I am not an expert, but if the processor has to go outside to get every single piece of information to the external memory, will not that decrease the performance of the computer?

Kind Regards.

 

Hi useless,

Yes, reducing fetch latency was the reason cache memory came into use. It's a sort of averaging mechanism, as it has to be flushed and refilled from system ram periodically.

In the bad old days, with the system bus clocking at 33 MHz, latency was a real problem. Modern systems run 800 MHZ buses, and ram clocks out fast enough to keep up. So, my question is: is cache memory really that critical for fast system operation? A modern desktop has more computing capability than some of the early Cray supercomputers.

Intel and AMD are pretty much dedicated to a particular way of doing computing. There are alternative methodologies, which are now very much in the shadows. A so-called Harvard architecture processor has separate data and program busses. Would that enable a cacheless and therefore less complex processor to operate with similar performance. The point is that there are more possible solutions that what is big on the market.

 

DRAM is significantly denser and uses significantly less power. It is not as fast as SRAM, and requires refreshing. These is a price for everything......

 

Cache remains a very important component to achieve effective overall processor performance. As pointed out in a previous append, memories and attachment busses are faster, but so are the "on chip" processor cores. It's the ratio that matters, processor cycle time to memory subsystem time. Cache will be around for a long time.