YiffyGriffy

So you went ahead and bought the Athlon 64 because either your friends told you to, or the guy at the shop did, but why? What's so special about it anyway?

The AMD Athlon 64 represents a new technological leap in home computing, the use of an extended x86 code to allow use for 64-bit registers over the previous and aging 32-bit code from the Windows 95 days. But the code it can recognize isn't the only thing different. While the architecture is much the same as the K7 (Athlon and Athlon XP), several things are quite different. For example, the on die memory controler, the extended cache, and Hyper-Transport I/O link. I'll try my best to explain these and some other features of this processor here.

The Front Side Bus
This seems to be an area of confusion for most people, and I guess this can be attributed to most retailers listing the FSB as being "integrated into chip" or sometimes 800MHz (Socket 754) or 1000MHz (Socket 939) The first is true, the actual Front Side Bus does lie within the processor core, but what does it do? Firstly, the Front Side Bus does two things, it carries the memory data to and from the processor and the memory controller, and it carries the data from the AGP and PCI buses to the processor. In Athlon 64s Hyper-Transport in essence replaces the FSB in one of these functions, that of linking the processor to the AGP and PCI buses, but I'll get into that in a bit. The remaing action of the FSB, linking the processor core to the memory bus, still exists in the Athlon 64 core, and because the memory controller is also inside the Athlon 64 core, so is the Front Side Bus. The Front Side bus operates at whatever the core frequency is and is capable of moving 128 bits of data per clock cycle.

Hyper-Transport Link
Hyper-Transport also seems to confuse people a little bit because they aren't sure of what it actually does. It basically is a high speed two-way link from the processor core to the motherboard chipset. Also, it multiple CPU configurations, it serves as a high speed connection for linking processors together. The Hyper-Transport bus for all platforms (Socket 754, 939, and 940) is two-way, meaning it can both send a recieve data once per clock cyles, both data paths are 16-bits wide for an effective data transmision of 32 bits per clock cyles. Socket 754 Hyper-Transport operates at 800Mhz clock rate, but because it's bidirectional, it is said to have an effective bus frequency of 1600MHz. Socket 939 and 940 Hyper-Transport buses operate 1000MHz real clock rate or 2000MHz effecitve bus speed. Hyper-Transport also plays one very critical role, that of determining the CPU frequency. On the motherboard, there is a small chip called a PLL, or a frequency generator. This frequency generator creates a base frequency of 200MHz for the chipset, and inside the chipset there is an internal multipler, much like that of the CPU, to determine to full Hyper-Transport frequency. On socket 754, that multiplier is x4, and on Socket 939/940, it is x5. It also determines the CPU frequency by going through the CPU's internal multiplier. The base Hyper-Transport frequency is still 200MHz for the CPU, so for an Athlon 64 3000 (Socket 754) for example, it would be 200MHz HTT (Hyper-Transport) x 10 = 2000MHz CPU frequency.

On-Die Memory Controller
What does it mean? Exactly that, the memory controller is located on the CPU die (aka core) instead of inside the Northbridge chipset controller. What does it mean to you? Faster memory operations with significantly less latency. However, contrary to popular belief, the memory bus does not run at CPU core frequency, that is an Athlon 64 3000 (Socket 754) does not have a 2GHz memory bus. This is where the FSB comes back into play. The FSB is the link between the memory controller and the processor's on-die cache, and it does operate at the CPU frequency. The memory frequency is determined by a divider of the core cpu frequency, and the default is the same as the cpu multiplier. Seen below is a chart of CPU frequencies and the memory bus frequencies that would result at the use of a different divider.
Also, since the controller is inside the core, the physical length that the data has to travel to reach the core cache is much, much smaller, so memory operations can be done a 1/3 the latency over external controller operations, improving efficiency and speed at low bus speeds.

New and Extended Caching Structure
Athlon 64s and the K8 Architecture use what is known as an exclusive caching structure. On-die cache has levels, Level 1 (L1) and Level 2 (L2) cache is found on most desktop processor, with some higher end models and server systems can use a Level 3 (L3) cache. Cache is small spaces of memory directly on the CPU core where it stores data it uses often, much like how RAM is used to store information off the hard drive when its used often. However, on-die cache is much faster than RAM, but significantly smaller with the average processor supporting only 1280-640KB of cache. How Caching works is when the processor makes a call from data, it first checks its L1 cache, then L2, and if possible, L3, and if isn't there, it looks on the RAM, when it finds the the data, it is written to the cache and then used by the processor. It is written on the cache because there's a high chance that it will need the same data again, and storing it on the cache will allow the processor much faster access to it rather than having to read from the memory. But the problem with that is the cache will then quickly become full. In an Exclusive cache structure, when data is pulled from the memory, it is written to the Level 1 cache, then used by the processor. When the Level 1 Cache becomes full an LRU (least recently used) algorithm chooses data to be moved from the L1 cache to the L2 cache, freeing space for the new request to be written on the L1. If the procesor then makes a call for the data on the L2 cache, it then goes through the same procedure of freeing space on the L1 by copying part of it to the L2 and then copies the new request from the L2 onto the L1 Cache.

The AMD64 x86-64 Instruction Extension
A lot of people believe that the reason an Athlon 64 is faster than its competetion is because it's a 64-bit processor, and this is not true. Yes, it is a 64-bit processor, but as there are no mainstreame 64-bit applications or operating systems, it is still running the same 32-bit code as an Intel Pentium 4 or Athlon XP. Its speed is attributued to its efficneicy and high rate of operations done per second, as well as the low latency memory operations of the on-die memory controller. The AMD64 technology is an extension of the x86 IA-32 code. In a nut shell, all the AMD64 instruction set allows the processor to do is identify and execute 64-bit code and 32-bit code at the same time. The benefits of 64-bit coding is more adressable system memory and more registers to use in coding. More registers allows coders to have more options for a given execution and not have to go back and write additiona lines of code to "free-up" other registers. This makes the code faster and more efficient.

Overclocking
Because the processors are quite different than a "normal" processor, overclocking can be a bit more difficult.

• Normaly, one overclocks a processor by increasing the FSB, but you don't have that option on an Athlon 64. Instead, you increase the Hyper-Transport base frequency. As described above, it is the Hyper-Transport base frequency multiplied by the CPU's internal multiplier that determines the CPU frequency.
• The Hyper-Transport bus itself is also vunreable to instability much like the FSB would be in a normal type of system, but since with Athlon 64s you often have the option of lower the HTT multiplier, doing so can eleviate you from that instability. It is recommended that you should keep you Hyper-Transport bus close to around 1000MHz (2000MHz effective rate) to maximize performance and stability.

Other than that, overclocking is much the same.

Differences in Cores
There exists 5 different cores based on the K8 Architecture. They are ClawHamer, SledgeHammer, Paris, NewCastle, and Winchester. Both ClawHammer and SledgeHammer have 1MB of L2 Cache, whereas NewCastle and Winchester only have 512KB of L2 Cache. ClawHammer, SledgeHammer, and NewCastle are all built on 130nm process and use 1.5 volts for the core voltage, Winchester is built on 90nm process and uses 1.4 volts for the core voltage. Winchester is slightly faster than the other cores (.5%) at the same clock speed in ALU performance. ClawHammer was the first Athlon 64 to be introduced, it is only availabe on Socket 754 and can only be bought in a DTR or Mobile chip or the Desktop 3700+ Model, as the majority of desktop models was replaced by NewCastle. NewCastle can be bought on either Socket 754 or 939, and was initially designed to be a lower cost alternative to ClawHammer. Winchester can only be bought on Socket 939 and is AMD's first 90nm chip, and is inteded to eventually replace NewCastle. SledgeHammer is available on Socket 940 (Opteron) or Socket 939 (FX series and Model 4000 and up), it mainly is used on the server market under the Opteron line, but also for the ultra-high end Socket 939 FX series. SledgeHammer and ClawHammer are the best performing cores because of their 1MB of L2 cache. Paris is identicle to NewCastle except it has half the cache (only 256KB L2) and has the AMD64 instruction disable, though it may lay dormant inside the chip. Paris so far is only used on the Sempron 3100, but eventually all Semprons will be moved to Socket 754 and use the Paris core.

__________________
Went almost two years without any major system overhauls, so I'm treating myself

muffin

nice

finally got the energy to do it

__________________
Nikon D90 + D40x / 18-55mm DX / 18-105 DX VR / 55-200mm DX VR / 50mm AF f1.8 / 70-210mm AF f4-5.6

YiffyGriffy

lol yeah I was sitting here thinking about it.

__________________
Went almost two years without any major system overhauls, so I'm treating myself

muffin

? I'm tired, or does that sound funny? Not technically, just .... uhhhh need sleep

Sorry, picking faults already

__________________
Nikon D90 + D40x / 18-55mm DX / 18-105 DX VR / 55-200mm DX VR / 50mm AF f1.8 / 70-210mm AF f4-5.6

YiffyGriffy

Please keep doing so, need to find all errors and change them so I don't confuse people even more

__________________
Went almost two years without any major system overhauls, so I'm treating myself

archer_456

Very nice Yiffy, this will come in handy when I decide to breakdown and get a (mumbling under his voice..) Athlon 64.

__________________
PSU Sleeving Guide
Rounding IDE cable Guide

Evil XP2400

you're missing a Y there.

See, the guide is so good archer is thinking of switching to the good side .

Goodjob man!

__________________

Zeus

nice write up

__________________

pazsion

[*]Normaly, one overclocks a processor by increasing the FSB, but you don't have that option on an Athlon 64. Instead, you increase the Hyper-Transport base frequency. As described above, it is the Hyper-Transport base frequency multiplied by the CPU's internal multiplier that determines the CPU frequency.[*]The Hyper-Transport bus itself is also vunreable to instability much like the FSB would be in a normal type of system, but since with Athlon 64s you often have the option of lower the HTT multiplier, doing so can eleviate you from that instability. It is recommended that you should keep you Hyper-Transport bus close to around 1000MHz (2000MHz effective rate) to maximize performance and stability. [/list]Other than that, overclocking is much the same.[/QUOTE]

Yes, this is what a would adjust, but how do i adjust it. do i go in a flip switches, wire up some pins? or is the only way to d/l software or manually decode the cpu?

YiffyGriffy

It's still adjusted in BIOS like older platforms. Some motherboards will call it HTT frequency, others still call it FSB or external clock. If your board supports overclocking options, it'll be the number that's 200 is basically the best I can say since practically every BIOS has a different name for it.

__________________
Went almost two years without any major system overhauls, so I'm treating myself