Discussion Forums > Technology
Intel kicking AMD to the curb?
kureshii:
--- Quote from: kyanwan on July 21, 2009, 06:56:35 PM ---( I don't bother with OCing - I don't have the time, or need. ^_^ )
Those intel CPUs have 2X and 3X the Cache of the AMD competitor. You're paying for that - do you know how expensive that cache is?
By design, the cache preloads code into the CPU - the on-CPU cache executes far faster than RAM, far faster than anything your system has. By design, these CPUs run much faster than the AMD counterpart.
For the amount of cache they have on hand, this only proves to me there is nothing superior about the design of the CPU. The only thing Intel has done, is made a larger line of CPUs with big cache. That's all they're doing.
--- End quote ---
Just to point out something minor, it's the LGA775 generation of Intel chips (excluding the Celerons of course) that have larger cache. To my understanding, this is to offset the slightly higher latency involved in accessing main memory due to the Northbridge-bound memory controller (vs AMD chips which have the memory controller on-die).
[edit] In fact, now that I check, AMD Phenom II X4 955 has double the L1/L2 cache of Intel Core i7 920, but less L3 (6MB on 955 vs 8MB on 920). Q9550 has half the L1 and double the L2 size of the X4 955.
The i7 series actually have less L1+L2 cache than the Core 2 Quad series (about half of it), and IIRC from the Anandtech article/interview this was because Intel had improved the access time for their caches (mostly due to the now-integrated memory controller). The lack of cache storage was offset by adding a shared L3.
--- Quote from: kyanwan on July 21, 2009, 11:50:17 PM ---I've used whatever was available and at a decent price point / performance - course, to last a few years.
--- End quote ---
I do the same thing; buying whatever's best for what I intend to use it for (which isn't always building a personal desktop). Right now the only devices I can safely say I'll always buy Intel for, are laptops.
Tatsujin:
--- Quote from: kureshii on July 22, 2009, 10:07:27 AM ---
--- Quote from: kyanwan on July 21, 2009, 06:56:35 PM ---( I don't bother with OCing - I don't have the time, or need. ^_^ )
Those intel CPUs have 2X and 3X the Cache of the AMD competitor. You're paying for that - do you know how expensive that cache is?
By design, the cache preloads code into the CPU - the on-CPU cache executes far faster than RAM, far faster than anything your system has. By design, these CPUs run much faster than the AMD counterpart.
For the amount of cache they have on hand, this only proves to me there is nothing superior about the design of the CPU. The only thing Intel has done, is made a larger line of CPUs with big cache. That's all they're doing.
--- End quote ---
Just to point out something minor, it's the LGA775 generation of Intel chips (excluding the Celerons of course) that have larger cache. To my understanding, this is to offset the slightly higher latency involved in accessing main memory due to the Northbridge-bound memory controller.
The i7 series actually have less L1+L2 cache than the Core 2 Quad series (about half of it), and IIRC from the Anandtech article/interview this was because Intel had improved the access time for their caches (mostly due to the now-integrated memory controller). The lack of cache storage was offset by adding a shared L3.
--- Quote from: kyanwan on July 21, 2009, 11:50:17 PM ---I've used whatever was available and at a decent price point / performance - course, to last a few years.
--- End quote ---
I do the same thing; buying whatever's best for what I intend to use it for (which isn't always building a personal desktop). Right now the only devices I can safely say I'll always buy Intel for, are laptops.
--- End quote ---
Extremely super newb question, what's L1/2/3 Cashe? I never understood that.
sdedalus83:
High speed local memory for the CPU. Instructions are fetched from RAM and stored in cache until executed.
L1 is for data the cpu will process immediately, L2 for data which has been predicted to be needed in the near future, L3 is generally a high latency extension of L2, but with newer designs, specifically with individual cores having dedicated L2, it stores data which has yet to be assigned a core. It helps make L2 more efficient by minimizing the need for inter core communication.
Lupin:
--- Quote from: Tatsujin on July 22, 2009, 02:18:11 PM ---Extremely super newb question, what's L1/2/3 Cashe? I never understood that.
--- End quote ---
It's a small amount of memory found inside your processor. Data/instructions are fetched from your main memory (RAM) and placed in your cache which are much faster than your RAM (because it's much closer to the processor). There are different levels of cache, L1 being the fastest and L3 the slowest. Intel and AMD have different implementations of the cache. AMD has a big L1 cache (compared to Intel's) and smaller L2/L3 cache (again compared to Intel's) but is exclusive (meaning data in the L1 cache cannot be found in L2/L3 cache. Intel is the exact opposite--very small L1 cache but LARGE L2/L3 cache. Cache maybe also be shared between cores.
--- Quote from: kureshii on July 22, 2009, 10:07:27 AM ---Just to point out something minor, it's the LGA775 generation of Intel chips (excluding the Celerons of course) that have larger cache. To my understanding, this is to offset the slightly higher latency involved in accessing main memory due to the Northbridge-bound memory controller.
--- End quote ---
Very true. It was during the Prescott days when Intel increased the external cache of their processors simply because it can't compete with AMD's on-die controller if it didn't.
kureshii:
--- Quote from: Tatsujin on July 22, 2009, 02:18:11 PM ---Extremely super newb question, what's L1/2/3 Cashe? I never understood that.
--- End quote ---
As explained earlier by Lupin and sdedalus, it's essentially very fast (but very small) memory that sits on the processor die (i.e. packaged with the processor), and is used by the processor to store data and instructions that it works with. Data that it can't find in the cache will have to be pulled from main system memory or your storage devices, which slows things down tremendously.
The main difference between processor cache and your system memory is that your caches consist of SRAM, while system memory consists of DRAM. Without reading those articles, all you need to know is that SRAM requires more components, is more complicated to make and thus is more expensive (so you only have a little of it). DRAM is simpler and cheaper to make, so you have more of it (just compare the circuit diagrams of SRAM and DRAM to get an idea of that).
In addition, DRAM has the disadvantage of being "leaky" (as it's based on capacitors) so it has to be refreshed every so often. Of course, during this refresh no data can be read or written, which further adds to access delay.
If I've successfully piqued your interest, Gustavo Duarte has an excellent writeup of the various things in your computer's processing pipeline that slow it down. In particular, note the latencies involved in accessing cache, memory, hard disks, and the Internet, as well as the respective sizes of the caches. Hope that's enough reading for you ;)
[edit] In case that's not enough, here's a book/writeup discussing it in more detail, and with comparison graphs to boot.
Link: What every programmer should know about memory (PDF version)
Navigation
[0] Message Index
[#] Next page
[*] Previous page
Go to full version