About The Architecture
To date, most of the talk with regard to the Athlon 64's on-die memory controller focused on its high-speed, low-latency operation, and its inherent performance benefits. But the K8 architecture on which the Athlon 64 is based was also designed with multi-core in mind from the very beginning. And the A64 actually has an on-die Northbridge, it's not just a memory controller. This design gives AMD's new Athlon 64 X2 processors a marked advantage over Intel's offerings.
AMD Athlon 64 X2 Architectural Overview
Intel's current dual-core processors, and their next-gen 65nm dual-Presler core processors, require a traditional Northbridge with an integrated memory controller. This means that any core-to-core communications have to be done over the front side bus, which obviously comes at a performance penalty. AMD's Athlon 64 X2 processors, however, don't suffer from this limitation. Where the Pentium-D will have to send some data off and back onto the CPU, the Athlon 64 X2 does not. Each of the X2's execution cores will instead send a request to the on-die System Request Interface and when resources are available, the request will be sent to the appropriate core. This is fundamentally similar to Intel's approach, but AMD's processors handle the entire process without having any data required to leave the confines of the CPU.
There is one shortcoming that both Intel and AMD are contending with, however. Both architectures must share the total amount of memory bandwidth available with each processor core. This situation exists to maintain compatibility with current platforms, but it will likely be remedied in the future when AMD and Intel move to new sockets, and use faster memory technologies.
When compared to other processors in the Athlon 64 line-up, the X2 4800+ closely resembles other .09 micron Athlon 64 processors...
As you can see in the above chart, the Athlon 64 X2 4800+ shares many of the same features of the Athlon 64 4000+ / FX-53, but there are a few notable differences. Their clock speeds are the same, but because the X2 4800+ has two cores, there is double the total amount of cache on-die. Each of the X2 4800+'s cores have 1MB of L2 cache, for a total of 2MB, and each core has 128K of L1 cache, for a grand total of 256K. And because the X2s are basically two processors-in-one, their die size is quite large at 199mm2. They're actually only a little larger than the older Athlon 64's built using AMD's .13 micron manufacturing process. The new Athlon 64 X2 4800+ however, is built on AMD's revised .09 micron manufacturing process, so the die isn't astronomically large. And like the recently released Rev. E (Venice) Athlon 64s, the X2 4800+ has more manageable power requirements and TDP. In fact, the Athlon 64 X2 4800+ requires only 1.35v-1.4v and has a max thermal power only 6 watts higher than an FX-55. Athlon 64 X2 processors will also incorporate the recent changes make to the Venice core, which include:
|·_Mismatched DIMM support (ability to configure and use different sized DIMMs on the same channel)
·_Improved memory mapping (more efficient use of memory space)
·_Improved memory loading (fully populating the memory with double-bank DIMMs no longer requires the memory controller to run at DDR333)
·_SSE3 Instruction Support
We should also note, that the above chart doesn't reflect all of the Athlon 64 X2 processors being announced today. AMD is actually announcing four dual-core desktop processors with various configurations:
|Frequency / Cache Sizes: 4800+ 2.4GHz w/ 1MB L2 cache-per-core
Frequency / Cache Sizes: 4600+ 2.4GHz w/ 512KB L2 cache-per-core
Frequency / Cache Sizes: 4400+ 2.2GHz w/ 1MB L2 cache-per-core
Frequency / Cache Sizes: 4200+ 2.2GHz w/ 512KB L2 cache-per-core
The Athlon 64 X2 line of processors will initially include the 4800+ we're looking at here, and a 4600+ model which is also clocked at 2.4GHz, but with half of the amount of L2 cache available per core. The Athlon 64 X2 4400+ and 4200+ share a similar relationship, but they are clocked 200MHz lower at 2.2GHz.