CPU/Mobo dual channel question

N

NoobOverclocker

Novice
hey i have 2 ddr2 800 1 gb ram. My htt only runs at 1000mhz /2000mt/s. Since 800*2=1600mhz in dual channel, does this mean my ram frequency is faster than my htt so there will be no different if i ran them both at 500mhz? (500*2=1000mhz) Because if it is, ill go for the 500mhz because i can have lower timings. Or does 1000mhz htt run at 2000mhz effectively and can support 1600mhz of ram frequency?

thnx in advance
 
dual channel gives you more bandwidth, running 2* ddr800 means they run in parrallel and not at double the speed.

there is no need for you to bump down the speed or anything if you add a second stick
 
hmm but doesnt the data output from the ram is too much for the htt to handle? 2 one gigs of ram running at 800mhz has an output of 12.8 gigs. I read somewhere that the htt running at 1000mhz can only support about 9 gigs/s. Is that true? Is 1 gb of ddr 800 ram faster than 2 512mbs ddr400?
 
You've got it a bit mixed up, let me try to explain.

1. Conventional FSB architecture used on Intel platforms:



2. A64 platforms utilising HTT and an on-die mem controller:



As you can clearly make out in (1) , the Memory controller is part of the Northbridge. As such the memory bandwidth available to the CPU shares the FSB and is restricted by the max bandwidth of the FSB.

In (2) , the Mem controller is part of the CPU and directly accesses the memory. The HTT is not at all a part of the mem data channel and does not restrict it. The HTT is merely the interface between the CPU <-> Northbridge <-> Southbridge . (In multi-processor A64 setups too, additional HT links are used for procs to communicate with each other)

HyperTransport bandwidth is 8GB/sec.

DDR2 800Mhz dual-channel bandwidth should be around 12.8GB/sec.

One DDR2 stick @ 800Mhz is *theoretically* equal to two DDR1 sticks @ 400Mhz running in dual channel in terms of memory bandwidth alone but at the cost of additional latencies. How this affects CPU performance is largely dependant on the architecture - a particular CPU architecture may benefit from better latencies than more memory bandwidth or vice versa.
 
wow thnx that clears stuff up alot. but there are some questions i would like to be answered. itll be great if u or someone can answer them.

1. Because my htt can only support 8gigs of data/sec, does that mean if i use a graphics card with over 8 gigs of memory transfer rate, say the 7900gs( 42.2 gigs/s), will my graphics card only run at 8 gigs/sec?
2. Whats the use of a wider htt? in my mobo theres configurations of 4/4 8/8 16/16.
3. so is running 2 1gb ddr2 800 ram same as a 2 gig 1600mhz ram?
4. is the data transfer rate of the htt apply to the entire southbridge or to each individual device?

thnx
 
1. Nope, the bandwidth you're talking about is between the Graphics card's GPU and its RAM. (A simple analogy you can think of is the graphics card being a motherboard by itself, with the GPU acting as a CPU and the VRAM acting as regular RAM.)

A graphics card communicates with the CPU via the PCI-Express/AGP bus and transfers only a modest amount of data to and from the CPU. A PCI-E 16x channel (which most Gfx cards use) offers a max bandwidth of 4GB/sec (8GB/sec bidirectional) and that at the moment doesnt seem to be choking any of the graphics cards.

2. A wider HTT obviously offers more bandwidth ;) . 4/4 8/8 16/16 refers to how wide the HTT bus is in bits. However i recall in earlier times where people would go as low as 400Mhz from the default 1Ghz on the HTT and see a very negligible performance loss.

3. Theoretically yes. Its like how a dual-core CPU with 2 cores running at 1.8 Ghz can be compared to a single-core CPU running at 3.6 Ghz.

Of course in reality, efficiencies are lower than a two-fold increase and largely depends on how good the memory controller is at utilising dual-channels.

*A point to note, while DDR2 @ 800Mhz would be equal to 2 x DDR1 sticks @ 400Mhz , the DDR2 stick achieves such high speeds at the cost of additional latencies. If you had a single DDR1 stick capable of 800Mhz speeds, it would be faster than the DDR2 stick - of course since thats not possible, the DDR2 stick outclasses the DDR1 sticks by its higher clocks and offsets the latency disadvantage.

4. Yep, the bandwidth of the HTT applies to the entire southbridge (when communicating with the CPU - though i have no idea if devices on the southbridge would communicate between themselves :ashamed: ).
 
wow im learning more than i ever had, but a few things still confuse me:

1. how can a graphics card like the 7900gs process 42.2 gigs/s between the gpu and graphics ram when the link that supplies it with bandwitdh is only 8gigs/sec?
2. is the frequency of ram unlimited in an amd architecture motherboard? and intel is limited to bus frequency?
3. where is the harddrive located in the diagram? is it a part of the northbridge?or something else? is it limited by the fsb bus?
4. i have a asus m2n-mx se socket am2 motherboard. unfortunately it only allows me to underclock it in the bios. is there any way to mod the bios to get it to allow overclocking? ive tried clockgen and other software but they all didnt work.

thnx again man
 
1. The PCI-E link between the system and the graphics card is different/separate from the data channel between the GPU and the VRAM.

Also the GPU just receives very basic data from the CPU which is why the PCI-E bandwidth is enough.. the GPU performs additional work on this data - floating point operations, the rendering of images, etc and these are much more memory intensive processes, hence the higher bandwidth between GPU and VRAM (someone with in depth knowledge of graphics cards can explain this better :ashamed: ).

2. In an AMD platform the frequency and bandwidth is theoretically unlimited. In an Intel platform, you can run memory 'asynchronously' to the FSB so as to run the memory at higher frequencies than the FSB.. however the bandwidth would still be limited by the FSB. Nothing operates at 100% efficiency, hence the idea behind running the memory async at higher frequency is so that the the memory bandwidth would approach the max possible theoretical bandwidth.

3. Hard drives are hooked up to the Southbridge -> In the diagram, the block labelled "Via DriveStation" refers to the hard drives. It would be limited by the link between the Northbridge <-> Southbridge, but even a RAID setup wouldnt really saturate this link.

4. I'm not sure about the 'SE' model, but the Asus M2N-MX has decent overclocking options. Have you updated to the latest BIOS revision? If you have a BIOS editor you could possibly tamper and unlock any hidden options.. but not sure how safe it is. An easier option would be just getting a good OC'ing capable board.
 
ahh ok i understand the graphics thing now :). Man u sure know heaps.
Damn my mobo cant let me change fsb and voltage or multiplier. Ive tried amibios and other bios editors but all the one si tried had problems opening it. I tried opening with hex editor and that worked but i dont know what to do.

My mobo: asus m2n-mx se
bios: 308 latest version for se edition
cpu: amd x2 4000 at 2.1 gigahertz
current fsb: 200
multiplier: 10.5

the bios only let me underclock by lowering multiplier. no other options :(

can u have a look at my bios?
ive attached it
thnx eddy
 

Attachments

  • Asus m2n-mx bios.zip
    435 KB · Views: 99
Back
Top