- Joined
- Dec 28, 2005
- Location
- In your mirror
Physical differences between DDR and DDR2:
EDIT: Bench between DDR2 frequency and timings:
http://www.xbitlabs.com/articles/memory/display/ddr2-oc1ghz.html
Sorry about the previous colors
DDR and DDR2 modules have the same size, but DDR modules have 184 contacts, while DDR2 modules have 240. Thus there is no way to install a DDR2 module on a DDR socket and vice-versa unless you have a hammer and pound it in. All DDR2 chips use BGA (Ball Grid Array) packaging, while DDR chips almost always use TSOP (Thin Small-Outline Package) packaging. There are DDR chips with BGA packaging on the market, but they are not so common. Currently BGA parts are the same as TSOP but fabrication is more expensive because of tighter tolerances and inspection.
Electrical differences between DDR and DDR2:
DDR2 uses the exact same sending technique during read cycles, two bits per clock cycle, however, the internal memory cell array and buffers have been changed. DDR2 allows a much broader range of data to be sent which is achieved by effectively a dual core system within the stick. Two cores are added to the cell array allowing both to operate independently; each core has the two bit system, which doubles this to four bits total. The data buffer in order to allow both the cores to
operate at their peak runs at twice the speed of the cell array, allowing each array to dump its memory into the buffer for the trip along this bus. While the bus remains at the same speed, this new architecture works something like the Intel Quad Data Rate bus in order to give 4 times the speed of the internal clock. DDR at 100MHz clock has a speed of 200MHz and DDR2 at 100MHz clock is effectively 400MHz. This is how DDR2 gets it speed increase, as it allows the clocks to be lowered while increasing the overall data throughput. With this in mind, 200MHz DDR put into a DDR2 fashion would give us 800MHz, well beyond that of the DDR capabilities. Naturally this helps for bursing and reading consecutive locations but doesn't help when reading single bytes or reads from random locations.
Buss termination:
DDR motherboards have termination resistors to balance line impedence. The resistor values are chosen in compromise because the motheroard manufacterer must choose a value to accomidate all memory types and impedance characteristics. DDR2 memory uses internal termination located inside the memory chips, a technique called ODT, On-Die Termination. This allows the memory manufacturer to set the termination value exactly as required. In general, as the memory is overclocked, signal integrety is reduced and DDR2's internal resistors keep signals cleaaner allowing DDR2 memory to achieve higher and more stable overclocks.
Power Consumption:
DDR2 memories have a lower power consumption compared to DDR memories and are fed with 2.5 V while DDR2 memories are fed with 1.8 V, Smaller die sizes and reduced power yield a much higher overclocking potential. Intel's Presler and Conroe will get their high overclocking ability for this same reason (IMHO)
DDR2 Off Chip Driver Calibration:
To further improve signal integrety, DDR2 also contains an off-chip driver (OCD) calibration to reduce the mismatch between pull-up and pull-down output driver characteristics, which can improve the system timing margin.Without OCD calibration, the DRAM has a nominal output driver strength of 18 ohms +30% and a pull-up and pulldown mismatch of up to 4 ohms. Using OCD calibration, a system can reduce the pull-up and pull-down mismatch and target the output driver at 18 ohms to optimize the signal integrity. OCD calibration is performed by the LOAD MODE command using the EMR, which enables the DRAM to be in OCD drive mode with the outputs driving HIGH or LOW, or the DRAM can be in OCD adjust mode. I'm not sure which motherboards, if any, support sending the LOAD MODE command during boot.
DDR2 Posted CAS and Additive Latency:
I'm still working on understanding this and not sure if the BIOS allows changing this parameter or the ram manufacturer has it "hard coded" in. Initial understanding is that this is a "delay", it is used to prevent collisions between the ram's command and data phases, eliminating the need for resynchronization, causing gaps between data bursts. This parameter only applies to read cycles. I'll update this post once I understand it more and it if has an impact on performance.
[/COLOR]
Don't let higher timing requirements scare you:
DDR2 memories work with higher timing numbers than DDR memories. Higher timing numbers do not necessarily mean longer latencies because DDR2 can run at a higher frequency. In other words, they delay more clock cycles to deliver a requested data, but the clocks are faster....basically 6 in one, half a dozen in the other.
When comparing memories with different speeds and between DDR and DDR2, it may be easier to convert the manufacturer's timings into "real time" nanoseconds. DDR400 CL3 memory, this “3” means that the memory delays three clock cycles and since this memory runs at 200 MHz, each clock tick measures 5 ns (T= 1/f).
Thus its latency if of 5ns X 3 = 15ns. It's interesting to note that a 5 volt pulse can travel about 15" in a nanosecond, thru a copper trace at room temperature.
DDR2-533 CL3 memory, this “3” also means that the memory delays three clock cycles to start delivering the request data, but since this memory runs at 266 MHz and not 200Mhz, each clock tick measures 3.75 ns, so its latency is of 11.25 ns, making this memory faster to data delivery than our DDR400 CL3 memory. So a DDR2-533 CL4 memory has the same latency as a DDR400 CL3. The table below list the actual time for each clock tick for a rated frequency. So, just get the number below depending on the memory type you want to compare and multiply it by the latency value in order to know the latency duration in nanoseconds, allowing you to compare latencies of memories with different speed ratings and to know which memory is faster.
Memory Clock Tick Duration (each one)
DDR266 = 7.5 ns
DDR333 = 6 ns
DDR400 = 5 ns
DDR2533 = 3.75 ns
DDR2667 = 3 ns
DDR2800 = 2.5 ns [/COLOR]
Consider forward dividers:
Some may think that running ram at 800MHz or 1000MHz isn't possible with todays motherboards. Traditionally mobos allowed scaling down the memory speed to allow higher fsb number to be used with slower ram, but possibilities exist to run a multiplier rather than a divider, causing the ram to run faster than the fsb. Although not ideal efficiency, substancial performance can be gained by running DDR with high latency numbers (NOT HIGH LATENCY) at 1000MHz and using
a 1:2 multiplier. End result will be the fsb running at 250 while the ram scream off (internally) at a GIG.
Opinions:
At stock speeds, the differenced between using DDR and DDR2 at 533MHz is minimal. Where DDR2 will excell, is in it's overclocking potential and the promise of future applications making use of it's enhanced speed and bandwidth. As when DDR was first introduced over SDRAM, it will be a while before these advantages are fully realized, but DDR2 will be sure to come into it's own sooner rather than later. Right now, there is no reason to upgrade your existing system based solely
on DDR2. However, I would definitely recommend going with a DDR2 platform for future expandability not to mention it's overclocking potential. Advancements and improvements for DDR will take a low priority as ram manufactures focus on DDR2. As much as enthusiasts loathe the "slow" performance of DDR2 over standard DDR, nothing can stop DDR2 from taking over because as per the JEDEC standard, DDR can't scale any further in terms of non-overclocked frequency.
The fastest chip was 5ns until Samsung released the TCCD that runs at 4ns. Although Samsung was able to advance further, DDR is reaching the light at the end of the tunnel. You can't run it faster than the internal delays. DDR2 is going to take over and has the ability to scale frequencies to outrageously high levels.
EDIT: Bench between DDR2 frequency and timings:
http://www.xbitlabs.com/articles/memory/display/ddr2-oc1ghz.html
Sorry about the previous colors
DDR and DDR2 modules have the same size, but DDR modules have 184 contacts, while DDR2 modules have 240. Thus there is no way to install a DDR2 module on a DDR socket and vice-versa unless you have a hammer and pound it in. All DDR2 chips use BGA (Ball Grid Array) packaging, while DDR chips almost always use TSOP (Thin Small-Outline Package) packaging. There are DDR chips with BGA packaging on the market, but they are not so common. Currently BGA parts are the same as TSOP but fabrication is more expensive because of tighter tolerances and inspection.
Electrical differences between DDR and DDR2:
DDR2 uses the exact same sending technique during read cycles, two bits per clock cycle, however, the internal memory cell array and buffers have been changed. DDR2 allows a much broader range of data to be sent which is achieved by effectively a dual core system within the stick. Two cores are added to the cell array allowing both to operate independently; each core has the two bit system, which doubles this to four bits total. The data buffer in order to allow both the cores to
operate at their peak runs at twice the speed of the cell array, allowing each array to dump its memory into the buffer for the trip along this bus. While the bus remains at the same speed, this new architecture works something like the Intel Quad Data Rate bus in order to give 4 times the speed of the internal clock. DDR at 100MHz clock has a speed of 200MHz and DDR2 at 100MHz clock is effectively 400MHz. This is how DDR2 gets it speed increase, as it allows the clocks to be lowered while increasing the overall data throughput. With this in mind, 200MHz DDR put into a DDR2 fashion would give us 800MHz, well beyond that of the DDR capabilities. Naturally this helps for bursing and reading consecutive locations but doesn't help when reading single bytes or reads from random locations.
Buss termination:
DDR motherboards have termination resistors to balance line impedence. The resistor values are chosen in compromise because the motheroard manufacterer must choose a value to accomidate all memory types and impedance characteristics. DDR2 memory uses internal termination located inside the memory chips, a technique called ODT, On-Die Termination. This allows the memory manufacturer to set the termination value exactly as required. In general, as the memory is overclocked, signal integrety is reduced and DDR2's internal resistors keep signals cleaaner allowing DDR2 memory to achieve higher and more stable overclocks.
Power Consumption:
DDR2 memories have a lower power consumption compared to DDR memories and are fed with 2.5 V while DDR2 memories are fed with 1.8 V, Smaller die sizes and reduced power yield a much higher overclocking potential. Intel's Presler and Conroe will get their high overclocking ability for this same reason (IMHO)
DDR2 Off Chip Driver Calibration:
To further improve signal integrety, DDR2 also contains an off-chip driver (OCD) calibration to reduce the mismatch between pull-up and pull-down output driver characteristics, which can improve the system timing margin.Without OCD calibration, the DRAM has a nominal output driver strength of 18 ohms +30% and a pull-up and pulldown mismatch of up to 4 ohms. Using OCD calibration, a system can reduce the pull-up and pull-down mismatch and target the output driver at 18 ohms to optimize the signal integrity. OCD calibration is performed by the LOAD MODE command using the EMR, which enables the DRAM to be in OCD drive mode with the outputs driving HIGH or LOW, or the DRAM can be in OCD adjust mode. I'm not sure which motherboards, if any, support sending the LOAD MODE command during boot.
DDR2 Posted CAS and Additive Latency:
I'm still working on understanding this and not sure if the BIOS allows changing this parameter or the ram manufacturer has it "hard coded" in. Initial understanding is that this is a "delay", it is used to prevent collisions between the ram's command and data phases, eliminating the need for resynchronization, causing gaps between data bursts. This parameter only applies to read cycles. I'll update this post once I understand it more and it if has an impact on performance.
[/COLOR]
Don't let higher timing requirements scare you:
DDR2 memories work with higher timing numbers than DDR memories. Higher timing numbers do not necessarily mean longer latencies because DDR2 can run at a higher frequency. In other words, they delay more clock cycles to deliver a requested data, but the clocks are faster....basically 6 in one, half a dozen in the other.
When comparing memories with different speeds and between DDR and DDR2, it may be easier to convert the manufacturer's timings into "real time" nanoseconds. DDR400 CL3 memory, this “3” means that the memory delays three clock cycles and since this memory runs at 200 MHz, each clock tick measures 5 ns (T= 1/f).
Thus its latency if of 5ns X 3 = 15ns. It's interesting to note that a 5 volt pulse can travel about 15" in a nanosecond, thru a copper trace at room temperature.
DDR2-533 CL3 memory, this “3” also means that the memory delays three clock cycles to start delivering the request data, but since this memory runs at 266 MHz and not 200Mhz, each clock tick measures 3.75 ns, so its latency is of 11.25 ns, making this memory faster to data delivery than our DDR400 CL3 memory. So a DDR2-533 CL4 memory has the same latency as a DDR400 CL3. The table below list the actual time for each clock tick for a rated frequency. So, just get the number below depending on the memory type you want to compare and multiply it by the latency value in order to know the latency duration in nanoseconds, allowing you to compare latencies of memories with different speed ratings and to know which memory is faster.
Memory Clock Tick Duration (each one)
DDR266 = 7.5 ns
DDR333 = 6 ns
DDR400 = 5 ns
DDR2533 = 3.75 ns
DDR2667 = 3 ns
DDR2800 = 2.5 ns [/COLOR]
Consider forward dividers:
Some may think that running ram at 800MHz or 1000MHz isn't possible with todays motherboards. Traditionally mobos allowed scaling down the memory speed to allow higher fsb number to be used with slower ram, but possibilities exist to run a multiplier rather than a divider, causing the ram to run faster than the fsb. Although not ideal efficiency, substancial performance can be gained by running DDR with high latency numbers (NOT HIGH LATENCY) at 1000MHz and using
a 1:2 multiplier. End result will be the fsb running at 250 while the ram scream off (internally) at a GIG.
Opinions:
At stock speeds, the differenced between using DDR and DDR2 at 533MHz is minimal. Where DDR2 will excell, is in it's overclocking potential and the promise of future applications making use of it's enhanced speed and bandwidth. As when DDR was first introduced over SDRAM, it will be a while before these advantages are fully realized, but DDR2 will be sure to come into it's own sooner rather than later. Right now, there is no reason to upgrade your existing system based solely
on DDR2. However, I would definitely recommend going with a DDR2 platform for future expandability not to mention it's overclocking potential. Advancements and improvements for DDR will take a low priority as ram manufactures focus on DDR2. As much as enthusiasts loathe the "slow" performance of DDR2 over standard DDR, nothing can stop DDR2 from taking over because as per the JEDEC standard, DDR can't scale any further in terms of non-overclocked frequency.
The fastest chip was 5ns until Samsung released the TCCD that runs at 4ns. Although Samsung was able to advance further, DDR is reaching the light at the end of the tunnel. You can't run it faster than the internal delays. DDR2 is going to take over and has the ability to scale frequencies to outrageously high levels.
Last edited: