L2 4MB vs L2 2MB

[pidr1nhu]

I would like to know if putting more memory would compensate the difference of performance between L2 4MB vs L2 2MB?
And if so, what kind of memory and how much?

 

[shabby]

Overclocking would compensate the 3% average difference.

 

[Maximilian]

I really doubt it would since cache is on-chip and memory isnt. Meaning memory will never be as fast as cache, you may see a performance increase with more memory, but 2mb is 2mb, theres nothing thats gonna offset the performance difference between 2mb and 4mb (which isnt too much anyays).

Put 2 gigs of ram in a 2mb cache system and you'll get overall better performance than a system with 1 gig and 4mb cache. But put 2 gigs into that 4mb cache system and you will get slightly better performance than the 2mb cache with 2 gigs of memory.

 

[Shimmishim]

http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=2802&p=4

You can see that the cache doesn't have a HUGE impact on performance in the "real world"

Yes, you'll notice the difference when testing superpi but you're not going to superpi all day long (at least i don't think so).

2 gigs vs 1 gig won't matter in terms of "making up the difference".

This is my opinion.

 

[Duvie]

Originally posted by: Shimmishim
http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=2802&p=4

You can see that the cache doesn't have a HUGE impact on performance in the "real world"

Yes, you'll notice the difference when testing superpi but you're not going to superpi all day long (at least i don't think so).

2 gigs vs 1 gig won't matter in terms of "making up the difference".

This is my opinion.

this is not entirely true. I have found an app that gains 20-25% with the 4mb of cache....

Folding at home core78 with brand new work units 1495, 1497, and 1499

E6400@3.2ghz...

1 instance of 149x = 5min frames
2 instances of 149x = 10min frames

E6600@3.2ghz

1 instances of 149x = 4min frames
2 instances of 149x = 5min frames

QX6700@3.2ghz

1 instance of 149x = 4min frames
2 instances of 149x = 4min frames
4 instances of 149x = 5min frames

All systems test were with at least 1gb of ram at 800ddr2. Each instance takes 100mb of ram. E6400 and E6600 are on identical systems....

Notice how when the 4mb cache chip has to split the shared cache for 2 cores it acts like 1 instance of the 2mb cache chip. Same as the 4 instances on the 8mb (2x 4mb cache cores) cache QX6700@3.2ghz.

this has been proven and duplicated on other folders boxes.

If I build folding boxes now they will only be 4mb cache units...

 

[Shimmishim]

Originally posted by: Duvie

Originally posted by: Shimmishim
http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=2802&p=4

You can see that the cache doesn't have a HUGE impact on performance in the "real world"

Yes, you'll notice the difference when testing superpi but you're not going to superpi all day long (at least i don't think so).

2 gigs vs 1 gig won't matter in terms of "making up the difference".

This is my opinion.

this is not entirely true. I have found an app that gains 20-25% with the 4mb of cache....

Folding at home core78 with brand new work units 1495, 1497, and 1499

E6400@3.2ghz...

1 instance of 149x = 5min frames
2 instances of 149x = 10min frames

E6600@3.2ghz

1 instances of 149x = 4min frames
2 instances of 149x = 5min frames

QX6700@3.2ghz

1 instance of 149x = 4min frames
2 instances of 149x = 4min frames
4 instances of 149x = 5min frames

All systems test were with at least 1gb of ram at 800ddr2. Each instance takes 100mb of ram. E6400 and E6600 are on identical systems....

Notice how when the 4mb cache chip has to split the shared cache for 2 cores it acts like 1 instance of the 2mb cache chip. Same as the 4 instances on the 8mb (2x 4mb cache cores) cache QX6700@3.2ghz.

this has been proven and duplicated on other folders boxes.

If I build folding boxes now they will only be 4mb cache units...

Good observation.

I'm not surprised that 4M L2 is faster. I didn't expect everything to be equal. In general though, if you see the tests AT ran, the order of the results is based on shear clockspeed.

 

[tallman45]

Patience will compensate you more

Intel is changing all E6300 and E6400 procs to 4mb L2 cache and keeping the price the same in 2-3 months

 

[pidr1nhu]

tallman45,
Is this information really true? Where did you get it?

 

[stevty2889]

Originally posted by: pidr1nhu
tallman45,
Is this information really true? Where did you get it?

Read the last paragraph of this page.

 

[tallman45]

Originally posted by: pidr1nhu
tallman45,
Is this information really true? Where did you get it?

Yes indeed it is common knowledge and posted many places including in an article found on this very website, see the above post.

Once released current E6300 and E6400 2mb L2 procs will be obsolete

Quote from that article

"so if you?re contemplating buying an E6300 or E6400 you?d be better off waiting a quarter so you can get the extra cache for free. "

 

[Shimmishim]

Agreed. the 6420 and 6320 will have the 4M L2 but keep the 1066FSB (8x and 7x mult respectively).

 

[pidr1nhu]

about the quadcore xeon's onsocket 775..do you know at least the speculation around its price?

 

[Sniper82]

Originally posted by: tallman45

Originally posted by: pidr1nhu
tallman45,
Is this information really true? Where did you get it?

Yes indeed it is common knowledge and posted many places including in an article found on this very website, see the above post.

Once released current E6300 and E6400 2mb L2 procs will be obsolete

Quote from that article

"so if you?re contemplating buying an E6300 or E6400 you?d be better off waiting a quarter so you can get the extra cache for free. "

is it really worth the wait performance wise? I mean would it make any difference in games down the road that would be noticeable?

 

[justlnluck]

So, if we are running one application at a time, that is made for a single core system, 4mb won't benefit us much. But if we need to run two programs simultaneously, then 4mb would show a larger improvement because each core gets 2mb cache, correct?

 

[Fox5]

Any cpu that benefits significantly from going from 2MB of cache to 4MB of cache (right now) for single threaded performance is either running some serious scientific software, or some seriously unoptimized programs.