Another tech question AMD 64 Socket 939 vs Socket 754

[Lumax]

Ok.. I'm leaning towards AMD for my upgrade. Intel is nice and all but it seems like you get more bang for your buck going AMD. Anyhow.. I've been pricing Athalon 64 3000s and I'm wondering what the difference is (besides $10) between a socket 754 64 3000 and a Socket 939 64 3000. Any signifigant performance increase going with the 939?

[DocBobo]

benchmark-comparison of socket 754 and 939 (http://www.xbitlabs.com/articles/cpu/display/athlon64-3800.html)
I'll sum it up for you. Potentially the socket 939 has two advantages over 754:
a) ram bandwith is higher
b) it will accept higher clocked processors in the future

Socket 754 is going to be the celeron line of AMD, accepting the Sempron processors (castrated Athlon 64s without the 64bit part and less cache).

The speed differences is part of the model number though.
While a socket 754 Athlon is model number 3400+, a 939 Athlon has a rating of 3500+. Both work internally at 2.2GHz, so no internal speed differences there. The 100+ virtual MHz are the faster bus of the socket 939 platform. Beware the wonders of marketing! :roll:

The main reason to go for socket 939 is that you will be able to upgrade later on. The socket 754 platform has been stopped at 3400+ Athlons, and god knows how fast any Sempron might get.

[Galnor]

One of the main differences is the memory system. The number (754 vs. 939) is the number of pins on the processor. The reason behind the large pin count on the 939 is that it is a dual memory channel controller.

What this means is that you have to put memory into the system in pairs. The system will actually interleave memory requests between the two memory sticks so that it can read and write to both sticks at the same time. This, in essence, doubles your memory bandwidth, as it effectively doubles the width of your memory bus.

The reason this works so well is that computer systems use a very high speed interal memory which is the cache. When the CPU needs inforation, it actually looks at the cache first to see if the data is there. If it isn't, the cache controller will read the data from your main memory. However, to speed things up and make it efficient. it will read more then just the one piece of data being asked for (which is typically 4 or 8 bytes).

What the cache controller will do is read a "cache line" worth of data, which tends to be 32 byte, 64byte, or some other large size. The reasoning behind this is that statistically, the next piece of data you want will be right after the one you just asked for. So, if the system reads ahead, it is very likely that the next data fetch will already be in the highspeed cache memory.

So, lets say the cache controller reads 32 bytes at a time. In a typical system, your memory bus is 64bits (8bytes) wide. So, the memory can only send 8 bytes of data per cycle (the MHz rating of your Front Side Bus which the memory is attached to memory). This means it would need to use 4 cycles to move the data. With dual channel memory, the system can read 8 bytes from each of the memory sticks at the same time, thereby allowing 16 bytes per cycle.

This is why they are going this route.

[Dravvan]

So...that being said, any motherboards sparkle more over the others for use of the 939s?

Thinking of doing the upgrade thing myself, but probably a piece at a time. :-)

[Galnor]

http://www.tomshardware.com/motherboard/20050103/index.html

Latest 939 motherboard review. I haven't read it yet, but they usually do a good job of explaining the differences and what you might want to get.