CPU L2 Cache - Does it make that big of a difference?

[AOforever1]

Well... I've been on this topic with tons of facts all over the place; real experience from friends and so forth, however, I am still yet to find a conclusion.

I brought a AMD X2 4600+ (Manchester 512 kb cache) over the 4400+ Windsor which has a 1 mb cache.

I managed to finalize this decision with facts from Tomshardware CPU charts. But... I again have found myself having doubts.

I need a 100th opinion (not a 2nd any more :P)

Am I losing out on anything besides multitasking and certain games? I am not running servers.
(Note: I do not want to overclock)

[ImClueless]

Well since you already made the decision. Honestly, it does not make that significant a difference except for certain applications. Then you already knew that. This is just to make you feel better.

[AOforever1]

Originally Posted by ImClueless

Well since you already made the decision. Honestly, it does not make that significant a difference except for certain applications. Then you already knew that. This is just to make you feel better.

It DOES make me feel better.
With that avatar of yours that is =P

Thanks, I have some reassurance now, unless someone else can say otherwise.

[Jinto]

I'ld like to point out this link:

http://forums.animesuki.com/showthread.php?t=34732

(just for reference, so you have something to read)

Personally I think 512kb for a dual core is rather low. And since some cache is used as Instruction cache (commonly too much instruction cache for too little data cache for whatever reason) and only a part of it is actually data cache, the cache side effects may become visible in highly cache optimized algorithms/programs.

Especially for dual core applications, the cache seems to be bottom boundary of what is usefull. How much cache has influence on the speed, depends on the used software too, but consider that many programs nowadays are compiled in a way to use medium sized caches (like the 1MBytes one - which translates to 512KBytes per core).

Still the difference I'ld say could be something in the 15% range. You must decide yourself, if it justifies the higher energy consumption of the 1MBytes cache architecture. Besides the 4600+ is faster in the cores so that compensates a little bit I think.

[AOforever1]

Originally Posted by Jinto Lin

I'ld like to point out this link:

http://forums.animesuki.com/showthread.php?t=34732

(just for reference, so you have something to read)

Personally I think 512kb for a dual core is rather low. And since some cache is used as Instruction cache (commonly too much instruction cache for too little data cache for whatever reason) and only a part of it is actually data cache, the cache side effects may become visible in highly cache optimized algorithms/programs.

Especially for dual core applications, the cache seems to be bottom boundary of what is usefull. How much cache has influence on the speed, depends on the used software too, but consider that many programs nowadays are compiled in a way to use medium sized caches (like the 1MBytes one - which translates to 512KBytes per core).

Still the difference I'ld say could be something in the 15% range. You must decide yourself, if it justifies the higher energy consumption of the 1MBytes cache architecture. Besides the 4600+ is faster in the cores so that compensates a little bit I think.

That's one of my fears as well.
Will the 4600 last me until my next upgrade when Apps are optimized for 2-4 mb caches?

[tritoch]

Am I losing out on anything besides multitasking and certain games? I am not running servers.
(Note: I do not want to overclock)

Who DOESNT multitask? =p

[killmoms]

Fuck guys, come on, the Conroe's that had 2MB of cache showed noticeable performance losses vs the 4MB versions, and that's on normal, unmodified games from today. I'd say the more cache the better.

Then again, you fucked up by buying AMD in the first place. Core 2 is wiping the ground with AMD in terms of raw performance, and price/performance ratio.

[ImClueless]

Leave the poor guy alone if you go here you will see that the 4mb cache conroe only shows minor performance gains over its 2 mb counter part. When you factor in the price its simply not worth it. Same goes for the decision to go with AMD rather than Intel. Its all about the price for performance and not the absolute performance. (unless you are filthy rich...then simply stfu please you braggart).

[Kurz]

Jinto Lin he was saying 512kb per core.

Killmoms never ever compare two completely different Architures. You can compare performance though.

Just because you have RX-7 (Racing car) doesnt mean a very light car with a mediocre engine cant catch up to you.

AMD still makes good processors, of course I'll be gaming/Overclocking so Core 2 Duo is right for me. However Motherboards for Core 2 Duo's are EXPENSIVE (Well the overclocking ones).

Cache will only be usefull in certain situations.
I believe encoding, compling and other tasks.

Then again we are talking about 1~15% performance increase in those situations.

[Jinto]

Originally Posted by Kurz

Jinto Lin he was saying 512kb per core.

In that case I think the cache is okay. 512KByte per Core are more than enough for most of the stuff one is running.

[killmoms]

Originally Posted by ImClueless

Its all about the price for performance and not the absolute performance. (unless you are filthy rich...then simply stfu please you braggart).

Did you notice the part where I pointed out that Conroe beats the Athlon 64 in terms of performance per dollar as well? Clearly you did.

Let me reiterate: AMD is no longer the best performance for your dollar, anywhere in the line, from mobile to server.

And also, I'm not filthy rich—I just have a job, and know how to save. So, fuck you sir.

[tritoch]

But didn't AMD produced those dual cores as well? I've read some benchmarks (gaming settings) that AMD can still outclass intel.

[Jinto]

Originally Posted by tritoch

But didn't AMD produced those dual cores as well? I've read some benchmarks (gaming settings) that AMD can still outclass intel.

I guess he was referring to the twice as wide arithmetic units (FPUs/ALUs) in the Core (2) Duos. Its either twice as precise or twice as fast compared to the AMD if used in SSE mode.
Thats the most important advantage. The next thing worthy to mention would be the better fitting architecture (at least I think the 2MB cache models are quite energy friendly when compared against any modern dual core AMD).

[Poseidal]

But didn't AMD produced those dual cores as well? I've read some benchmarks (gaming settings) that AMD can still outclass intel.

That was only true when the Pentium was Intel's flagship desktop processor, which is no longer the case.

Core 2 Duo outperforms the Athlon 64 family (including all their dual core chips) at everything currently, including gaming.

On the subject of Cache: It makes a difference if you do a lot of encoding; for gaming and general use the difference is much much less. For the most part, the lowest end chip, the E6300 is the best value purchase you can buy as far as processors are concerned; in relation to power/price.

[AOforever1]

Yeah....
I don't know how I overlooked this when I was thinking about it with my brother... but ... okay I screwed up (disreguarding that thought).

I have my 4600+ now guys... should of gotten a Conroe. How did I forget about that?

But anyways, what I have now is not all that bad because I never wanted to spend 200+ more bucks on DDR2 RAM when I have 2 gigs worth already. Also I have an ASrock MB which is a hybrid mobo; AGP, PCIe, 939, AM2.

I am currently quite satisfied with the performance I am getting. System response times are a lot better, and I am able perform more than one intensive task at once.

Spoiler:

Full Size Pic:
http://www.box.net/public/7zia72jtsu

Now I am wondering if I need a fresh install of windows to fully utilize the new processor, such as SSE3.

[Jinto]

Streaming SIMD (single instruction mutli data) Extension 3 is a processor feature and almost always used with special machine code (so it is platform independent) for example you just need to use inline assembler in C/C++ and a special compiler to use it in a program. SSE 3 is able to calculate a wider range of arithmetic operations as for example SSE 2 (or MMX) can. It works basically like this; instead of processing a single data value in an instruction, it uses data vectors of certain length (the arithmetic operation will use almost the same time to process the data vectors compared with the single value - that explaination is done on the keep it simple stupid principle, in that sense it is not complete but easy to understand)