DDR3 for £99 **This week only** Results inside

james.miller said:
helmut the core frequency is the frequency of the memory chips, which as i explained run at 1/2 the IO speed (the speed that the memory module actually runs at).


look at the ddr2-400 specs that i posted above. you'll see that it tallies up with the chart you posted

a ddr800 module runs at 400mhz. 400mhz is 1:1 on a quad pumped '1600mhz' intel cpu.
Click to expand...

your mixing symmetric dual channel mode with true 1:1 ratio.

basically the fsb is quad pumped, so cpu can access ram at an effective (fsb x 4) and ram speed is derived from ddr ram speed x 2.

so at 400fsb cpu has effective 1600mhz link speed to the ram, now, if the ram ran at 2ghz, the cpu could not make use of the extra bandwidth since its link speed to the ram is not as fast as the speed of the ram. and vice versa, if the fsb runs faster than the total ram speed then the ram is the limiting factor.

best way to work things out is to use the final speeds, fsb 1333 should be mated to 1333mhz ram for no theoretical bottlenecks.
 
Sorry not making any comments on the actual thread itself.

I know its only typing but the way you people talk to each other on here is absolutly pathetic, how about acting like mature adults & treating people with respect, if people are wrong then tell them in the right manner.

Wrong Reality|Bites, Yellowbeard is a Corsair Rep/Tech as I have to do an rma which I corresponded via his actual Corsair e-mail address, he does know what he is talking about.

I'm surprised the the dons dont say anything about the way people act on these forums.
 
Last edited:
Cyber-Mav said:
your mixing symmetric dual channel mode with true 1:1 ratio.

basically the fsb is quad pumped, so cpu can access ram at an effective (fsb x 4) and ram speed is derived from ddr ram speed x 2.

so at 400fsb cpu has effective 1600mhz link speed to the ram, now, if the ram ran at 2ghz, the cpu could not make use of the extra bandwidth since its link speed to the ram is not as fast as the speed of the ram. and vice versa, if the fsb runs faster than the total ram speed then the ram is the limiting factor.

best way to work things out is to use the final speeds, fsb 1333 should be mated to 1333mhz ram for no theoretical bottlenecks.
Click to expand...

Indeed. This is the way to keep things absolutely bottleneck free and everything, on a bit for bit level, equal. But that actually ignores dual channel, the big 'buzz phrase' since nForce 2 and the first Quad pumped Intel chipsets (cannot remember name I was an AMD man then). When you run a board and memory 1:1 the actual 'real frequencies' are/should be being used to work our the speed, so for 1600MHz Quad, that's 400MHz actual, and DDR2 that's 800MHz. Double team the sucker and you're filling every bit of the FSB channel with memory data as you double the output with it accessing both sticks at the same time. So really you either go by 1 frequency and negate the benefit of Dual channel, or go by the other, and negate the need for faster, more expensive RAM.

It should equate to the same bandwidth throughput overall, I believe. Of course that ignores the imperfect world we live in and the fact that latency will have an effect, and nothing works 100% efficiently, but overall, they are 2 situations with an almost identical output.
 
helmutcheese said:
A dual channel Mobo since NF2 days is 128bit.
Click to expand...

so a single channel link from cpu to ram would be 64bit, and dual channel would be 128bit from cpu to ram, so its the same end result. that means that to get close to 100% bandwidth usage ram would have to run at the same speed as the cpus fsb (which is quad pumped). so 1333mhz fsb + 1333mhz ram = happy days?
 
The FSB doesn't actually run at 1333Mhz though, That's the rated equivelent speed compared to a single pumped(channel) fsb. Same as DDR800 only runs at 400Mhz due to being double data rate.

It's only a 128bit interface between the memory and the CPU's memory controller, Whether(real word never typed it before) it be in the NB or intergrated into the CPU.
 
Wow lots of gubbins here lads - keep up the mutterings :)

So erm, ;), I have an AMD AM2 6000+ CPU with 4GB PC6400 800MHz Ram - is this ram really worth it - could I get equal performance using "inferiorer" PC 5300 Ram?
 
Reality|Bites said:
The FSB doesn't actually run at 1333Mhz though, That's the rated equivelent speed compared to a single pumped(channel) fsb. Same as DDR800 only runs at 400Mhz due to being double data rate.

It's only a 128bit interface between the memory and the CPU's memory controller, Whether(real word never typed it before) it be in the NB or intergrated into the CPU.
Click to expand...

seeing as how data is actually transfered during the rise and fall of the clock it does not make sense then how the memory bandwidth is higher than the theoretical limit imposed by the bus speed.

e.g: lets use this for an example:
q6600%203.2ghz%20corsair1600.png


above shows a 1600mhz bus mated with 1600mhz memory.

now in your eyes your seeing a 400fsb mated with 800mhz ram.

now lets do some maths:
bandwidth = (128bit memory interface * 400mhz clock speed) divided by 8 bits in a byte

so for the above calculation we get (128*400) /8 = 6400mb per second.

hold on a second, if the theoretical maximum bandwidth provided by 128bit bus width running at 400mhz is 6.4gig per second, then how the hell do you get the score above which is considerably higher than the theoretical maximum?

now step back and take the time to think why they have named 800mhz ram pc2-6400. any why do they name 1333 ddr ram pc3-10666? its to do with the max bandwidth they can theoretically provide.

the bus is quad pumped so it has a much higher theoretical bandwidth available than what your saying pal ;)
 
Last edited:
In the above picture:

Yes the CPU has an actual FSB of 400Mhz

The memory used is DDR3 1600 which runs at an actual 800Mhz

now lets do some maths:
bandwidth = (128bit memory interface * 800mhz clock speed) divided by 8 bits in a byte

so for the above calculation we get (128*800) /8 = 12800MB per second.

It's not a true 1600Mhz bus though is it ;)

What your saying is like taking a quad core CPU clocked at 2Ghz and saying it runs at 8Ghz.

You have been suckered in with the marketing havn't you :)
 
somethings wrong there. yse the pc-6400 rating or whatever it happens to be is the bandwidth figure, but it's also the bandwidth figure for a single stick of ram. maximum theoretical for two sticks of pc6400 is 12800mb/sec, of course.... again, from the wiki:

Note: DDR2-xxx (or DDR-xxx) denotes data transfer rate, and describes raw DDR chips, whereas PC2-xxxx (or PC-xxxx) denotes theoretical bandwidth (though it is often rounded up or down), and is used to describe assembled DIMMs. Bandwidth is calculated by taking transfers per second and multiplying by eight. This is because DDR2 memory modules transfer data on a bus that is 64 data bits wide, and since a byte comprises 8 bits, this equates to 8 bytes of data per transfer.
Click to expand...

so the 64bit trasfer rate of a single stick of ddr2-800 is 6.4gb/sec. in other words, 12.8gb/sec in dual channel.

edit: and why is sandra telling me my ip35-pro is a ddr3 bearlake with 2x2gb of ram. its not its 4x1gb ddr2 lol
 
Last edited:
james.miller said:
somethings wrong there. yse the pc-6400 rating or whatever it happens to be is the bandwidth figure, but it's also the bandwidth figure for a single stick of ram. maximum theoretical for two sticks of pc6400 is 12800mb/sec, of course.... again, from the wiki:

so the 64bit trasfer rate of a single stick of ddr2-800 is 6.4gb/sec. in other words, 12.8gb/sec in dual channel.
Click to expand...

your your right the bandwidth figures are based on single stick, or 64bit configuration, not the dual channel mode configuration.
as i said its working at theoretical levels, and the peaks bandwidth will not be attained unless all ram timings run at something like 1-1-1-1.

as for Reality|Bites, if you say the cpu in the above picture i have put up is running on a fsb of 400mhz, then why is the bandwidth showing over 10,000mb/s? since a 400mhz 128bit interface will not have more than 6400mb/s throughput.
going by what your saying the screenshot i posted above defies the laws of physics/electronics/common sense.
 
Because of the way that test works.

It bypasses the CPU and only tests the memory bandwidth.

Results are listed as

Memory bandwidth

CPU L1 bandwidth

CPU L2 bandwidth

The test is run between the memory controller and the Memory. not memory to CPU

Data travels CPU - NB - Memory (intel) and vice versa

For AMD it's just CPU - Memory.

I've had enough of you now really, You are just on the wind up.
 
Last edited:
Reality|Bites said:
Because of the way that test works.

It bypasses the CPU and only tests the memory bandwidth.

Results are listed as

Memory bandwidth

CPU L1 bandwidth

CPU L2 bandwidth

The test is run between the memory controller and the Memory. not memory to CPU

Data travels CPU - NB - Memory (intel) and vice versa

For AMD it's just CPU - Memory.

I've had enough of you now really, You are just on the wind up.
Click to expand...

oh realy? take a step back from the keyboard and think for a minute, where does the L1 and L2 cache reside?


thats right, on the cpu. :cool:
 
lol, what no comeback?

you could have replied saying the L1 and L2 cache resides on the northbridge, then we could carry on discussing this topic some more. :)
 
You must log in or register to reply here.
Back
Top Bottom