 |
01-05-2004, 08:55 PM
|
#1 (permalink)
| | Registered User
Join Date: Oct 2001 Location: UIUC
Posts: 1,144
| » 
Cache
Isn't there a point where having too much cache becomes a disadvantage?
edit: Whoops, I meant on a processor.
|
| |
01-05-2004, 08:56 PM
|
#2 (permalink)
| | Registered User
Join Date: Oct 2001
Posts: 6,533
|
yup..look at all those lottery winners whose lives have been ruined.
(lol, I knew what you meant, I was just being silly)
actually you may be thinking of "too much memory" perhaps....I have heard statements like "thats too much memory, cuz the cpu has to refresh it all" and then of course "thats too much memory because the cpu cant cache that much anyway"
dont know that I ever heard that too much cache was bad....the more cache you have the more stuff can be stored there and also now they have intelligent pre-fetch etc..so that it prefetches what it think might be the next lines of code needed etc...and since cache is soooo fast..you are better off if you can read the info from cache as opposed to from Ram or hard drive.
(yes, that was a very un-technical description)
__________________
"Even a fool is thought to be wise if he is silent"
Last edited by John Prophet; 01-05-2004 at 09:01 PM.
|
| |
01-05-2004, 09:12 PM
|
#3 (permalink)
| | Registered User
Join Date: Oct 2001 Location: Chicagoland IL
Posts: 1,539
|
Re: Cache
Quote: Originally posted by Chiguy Isn't there a point where having too much cache becomes a disadvantage?
edit: Whoops, I meant on a processor. | Don't think that point has been reached yet.... else, why would the "Big Two" be trying to increase cache size on their highest-end CPUs. Plus, CPUs' cache is so fast compared to system speed, that even with the suggested potential "lag" (because of the increased time the larger CPU cache needs to be refreshed, etc.) hit is still way ahead of the system. The CPU cache is running at gHz speeds.... everything else is running at mHz speeds.... an order of magnitude difference.
__________________
A man becomes rich not by having what he wants, but by wanting what he haves.
|
| |
01-05-2004, 11:24 PM
|
#4 (permalink)
| | Registered User
Join Date: Oct 2001 Location: UIUC
Posts: 1,144
|
The reason I ask is that I remember seeing something a while ago where it said that having a large cache is a disadvantage when it mispredicts a branch and needs to flush everything and reload.
Then again, with the pretty high accuracy for branch prediction, this isn't too much of a problem. Or is it?
|
| |
01-05-2004, 11:36 PM
|
#5 (permalink)
| | Registered User
Join Date: Feb 2002 Location: midvale, utah
Posts: 1,296
|
Well this shouldn't be a problem as it should be able to instantly flush the cache anyhow, however i wouldn't believe all processors would be effected, unless your friend is correct and the cache is effected.
However having more cache can be a great bennefit, the more you have, the less it matters having faster ram in most cases.
So no I wouldn't assume it's a problem if it hasn't been fixed yet.
|
| |
01-06-2004, 12:09 AM
|
#6 (permalink)
| | Guest
Join Date: Jun 2003
Posts: 1,203
|
Its never a dissadvantage. But CPU's with tons of it can render your wallet empty.
|
| |
01-06-2004, 03:08 AM
|
#7 (permalink)
| | Registered User
Join Date: Apr 2003 Location: PA. USA
Posts: 1,372
| Quote: Originally posted by Chiguy The reason I ask is that I remember seeing something a while ago where it said that having a large cache is a disadvantage when it mispredicts a branch and needs to flush everything and reload.
Then again, with the pretty high accuracy for branch prediction, this isn't too much of a problem. Or is it? | Your talking Pentium 4 ONLY here. Little mixed up too.
"when it mispredicts a branch and needs to flush everything and reload."
P4 has 20 stages. XP 10. When XP fumbles data no biggy not so may stages are rerun and time wasited effecting the Work per cycle ratio. So Intel had to do somthing. The williamette was horrible (I know I had one LOL). The reason for probs mainly. Large stages = high mhz BUT also means and data corrupted needing to be reset wasited all thos stages it had already ran killing the Work per Cycle MAJORLY. The p4 branch prediciton unit is supposed to be able to cut down the 20 stages to 11-15 or so by skipping stages in pipe that perticular data need not go to. Kinda like a elevator or a backdoor to each stage without having to single file it (1 bit at a time in order) through standard stage route. When prediction fails it sends to wrong place. Meaning data MUST be resent to cpu and again go through all the tedious stages. This data comes off HDD. So a LARGER L2 helps by storing the data close to cpu for quickest refetching in these situations. So if anyhting its the opposite of what you said. It helps (large l2) not hurts in this case. Sole reason for P4 having such large L2 right there. Look how dinky P4's L1 is1! Why? Cause its direct on the die cahce. Its in the cpu and fast and $. It runs at exact same speed as cpu. THe L2 runs half speed of cpu and is on substrait making it cheaper both to adhere to cpu and to buy. L1 chip bad you got a bad cpu-nothing you can do as its embeded.
__________________
In the mind of a mad man!
|
| |
01-06-2004, 06:38 AM
|
#8 (permalink)
| | Registered User
Join Date: Oct 2001 Location: Chicagoland IL
Posts: 1,539
|
Along the lines....
Along the lines of this topic area, see the article "Dreaming of a Cacheless Society" by Tom Halfhill, on page 14 of the Feb 2004 issue of MaximumPC.
__________________
A man becomes rich not by having what he wants, but by wanting what he haves.
|
| | |
Currently Active Users Viewing This Thread: 1 (0 members and 1 guests) | | | | Thread Tools | | | | Display Modes | Linear Mode |
Posting Rules
| You may not post new threads You may not post replies You may not post attachments You may not edit your posts HTML code is Off | | | | Most Active Discussions  | | | | | Recent Discussions  | | | | | |