L2 Cache...need this cleared up

[Bona Fide]

See, this has really been bugging me for a while now...what the hell is the point of an L2 cache? People have always been saying here, that the difference between 512Kb and 1Mb is about 3-5% in most applications. If that's the case, I should just go with a 3500+ Venice, right? As opposed to the 3700+ SD that I was considering ordering...

 

[Jeff7181]

What's the point of an L2 cache? To provide the processor with some high speed local memory.

*EDIT* As for your Venice vs. San Diego question, it's up to you. Do you want the extra 3-5% performance? Is it worth the extra cost?

There's another difference too. I've been hearing that the Venice doesn't use strained silicon, and the San Diego does... which means San Diego's should have higher overclocking potential.

 

[Fox5]

I think both Venice and San Diego use strained silicon...

Anyhow, 3%-5% isn't that bad. 5% of 2.4ghz is 120mhz, typically cache has the effect of adding the performance equivilant of 100mhz to 200mhz everytime it's doubled.

Imagine if that were done to an Athlon XP.
Start with Barton, say at 2.2ghz is equivlant to maybe an athlon 64 at 1.8.
Now quadruple the cache to 2MB, and you have an athlon xp that can match an athlon 64 in mhz to mhz performance.(hmm...but an athlon 64 with 512KB cache or 1MB? dual or single channel?)

 

[Furen]

It's not 3-5% every time you double the cache, It's around 0-10% with an average at around 3-5%. Those numbers just apply to going from 512k-1mb on AMD64 90nm cpus. I, personally, dont think it's worth it unless you are mostly a gamer, since some games actually gain more than most other apps. It does seem, however, that San Diegos are better overclockers (overall) than Venices.

 

[Jeff7181]

Originally posted by: Fox5
I think both Venice and San Diego use strained silicon...

Anyhow, 3%-5% isn't that bad. 5% of 2.4ghz is 120mhz, typically cache has the effect of adding the performance equivilant of 100mhz to 200mhz everytime it's doubled.

Imagine if that were done to an Athlon XP.
Start with Barton, say at 2.2ghz is equivlant to maybe an athlon 64 at 1.8.
Now quadruple the cache to 2MB, and you have an athlon xp that can match an athlon 64 in mhz to mhz performance.(hmm...but an athlon 64 with 512KB cache or 1MB? dual or single channel?)

Yeah you're right about Strained Silicon. I was thinking of Dual Stress Liners... San Diego uses that, Venice doesn't.

 

[fstime]

Originally posted by: Jeff7181
What's the point of an L2 cache? To provide the processor with some high speed local memory.

*EDIT* As for your Venice vs. San Diego question, it's up to you. Do you want the extra 3-5% performance? Is it worth the extra cost?

There's another difference too. I've been hearing that the Venice doesn't use strained silicon, and the San Diego does... which means San Diego's should have higher overclocking potential.

If thats true, i'd take the higher quality silcon over the other stuff anyday, same stuff the FX series uses.

You get the cream of the crop silcon with the SD.

 

[l Xes l]

L2 cache is worth the cost in my opinion...

 

[NokiaDude]

Just ignore the idiots who say "oh it only gives you 2-5% more performance!".

 

[Hacp]

Anandtech's benchmarks don't lie. In games it does give you only like 3-5% improvement. Amd processors don't need all that extra cache since they add up l1 and l2 cache, and also since they have shorter pipes, and finally, because they have hypertransport so if data needs to be used outside of the cache, it can get to the processor very fast.

 

[Acanthus]

Originally posted by: NokiaDude
Just ignore the idiots who say "oh it only gives you 2-5% more performance!".

Show me otherwise.

The only case where it DOESNT make a small difference is cpus that are crippled, like celerons, semprons, celeron-D etc.

 

[Hacp]

Also, in an article in ANANDTEch covering the Celeron D, the writer said that the benefits of increased cache followed a logarithmic curve, and not a straight line. That means as you add more cache, the performance improvements decreases.