Sign in with
Sign up | Sign in
Your question

I Dont Get It 1:1 Ratio? Whats that?

Tags:
  • CPUs
  • Memory
  • Product
Last response: in CPUs
Share
October 28, 2006 2:08:54 PM

Hi

I have been reading topics about memory and cpu fsb and most people talk about some 1:1 ratio. People say it makes faster your computer and you get more perfomance? How do you do this? I have a Core 2 Duo system to be built in no longer than 3 weeks and my memory will be Corsair TWIN2X1024 1GB PC6400 DDR800, also my mobo will be a NVIDIA nForce 680i. Just in case you need my system info.

Can somebody explain me what is all this 1:1 ratio about and what will happen if I dont do it in my system? and how to do it?

More about : dont ratio whats

October 28, 2006 2:43:28 PM

In your case, 1:1 ratio will be running your memory at 266MHz or DDR2 533MHz when at stock. Then you increase the RAM speed locked with the fSB to overclock if going that route...
October 28, 2006 2:46:41 PM

No, Im not going to overclock. But what does that mean? I will have to downclock my 2 sticks of RAM from 800mhz to 533mhz?
Related resources
October 28, 2006 3:08:23 PM

Hmmm... I second your interest in this, because people talk about this all the time but I've yet to read anything informative about it. Noone demonstrated that recently; as far as I can remember, that was important in the DDR days of 533/800Mhz FSB-ed P4s... but right now, I've yet to see any real evidence.

What I do know is that:

In theory, running a 1:1 ratio means that your memory's cycles match your FSB's cycles, meaning that it is easier for information to pass because commands don't need to wait to get along. Because of this match, all bandwidth is matched, also meaning that there is no bottleneck in either direction in terms of bandwidth.

So, again, in theory, anything higher than Dual-Channel DDR2-533 saturates the C2D bus, and the "ideal match" is only DDR2-533.

Now, back to reality, I've seen many benchmarks where DDR2-667 or DDR2-800 or even DDR2-1067 demonstrated tangible performance increases. This indicates that, although the bandwidth numbers for DC DDR2 memory are much more than C2Ds bus, there can still be a benefit from non-1:1 memory setups.

Consider, for instance, that DC DDR2-1067 has, in one of its channels, enough bandwidth to feed the C2D bus, yet there are still benefits from running it instead of DC DDR2-533, if you look at the benchmarks. Actually, DDR2-1067 is so fast that it uses a 2:1 ratio... Which makes me wonder why there isn't any widespread support for DDR2-1067. Current mobos support 5:4 ratio (DDR2-667), 3:2 ratio (DDR2-800), but no 2:1 ratio, which would seem more natural to me...

So this is why I second your interest. Do any of you guys know more about this?... Why is 1:1 supposed to be so good?
October 28, 2006 3:20:51 PM

According to wikipedia

Quote:
DDR2 Memory Modules
Unlike the previous Pentium 4 and Pentium D design, the Core 2 technology sees a greater benefit from memory running synchronously with the Front Side Bus (FSB). This means that for the Conroe CPUs with FSB of 1066 MT/s, the ideal memory speed is PC2-4200. In some configurations, using PC2-5300 can actually decrease performance. Only when going to PC2-6400 is there a significant performance increase. While expensive DDR2 memory models with tighter timings do improve performance, the difference in real world games and applications is negligible.[


So PC2-5300 sucks with Conroes. Pc2-4200 is the best but I dont like to stick with PC2-4200 because thats kinda old tech. It says only when using PC2-6400 there is a significant perfomance increase. So using DDR800 is still ok right? you get better perfomance compared to pc-4200. thats what I understand.
October 28, 2006 3:22:46 PM

From the Wikipedia, I think this would fairly simple explanation the ratio:

Quote:
CPU

The frequency at which a processor (CPU) operates is determined by applying a clock multiplier to the front side bus (FSB) speed. For example, a processor running at 550 MHz might be using a 100 MHz FSB. This means there is an internal clock multiplier setting (also called bus/core ratio) of 5.5; the CPU is set to run at 5.5 times frequency of the front side bus: 100 MHz x 5.5 = 550 MHz. By varying either the FSB or the multiplier, different CPU speeds can be achieved.

Setting a FSB speed is related directly to the speed grade of memory that a system must use. The memory bus connects the northbridge and RAM, just as the frontside bus connects the CPU and northbridge. Often, these two buses must operate at the same frequency. Pushing the front-side bus to 170 MHz means pushing the memory to 170 MHz in most cases.

In newer systems, it is possible to see memory ratios of "4:5" and the like. The memory will run 5/4 times as fast as the FSB in this situation, meaning a 133 MHz bus can run with the memory at 166 MHz. This is often referred to as an 'asynchronous' system. It is important to realize that, due to differences in CPU and system architecture, overall system performance can vary in unexpected ways with different FSB-to-memory ratios.

In complex image, audio, video, gaming, and scientific applications where the data set is large, FSB speed becomes a major performance issue. A slow FSB will cause the CPU to spend significant amounts of time waiting for data to arrive from system memory.
October 28, 2006 3:27:41 PM

Since this is kinda confusing

I just want to know if my system will run OK and will not have any perfomance decrease because of the use of DDR800.

And if there is any, how could I fix it?
October 28, 2006 3:28:36 PM

Hmmm that's what I understood from that too! DDR2-800 will result in a performance increase, it's OK for you do to so.

Though I must say that running 2:1 with DDR2-1067 isn't, mathematically, an asynchronous operation! The cycles match perfectly.

That's why I don't understand why there isn't more widespread support for DDR2-1067. It would be an excelent pairing, because the FSB cycles would [/i]always match 2 memory cycles![/i] It's not like 5:4 or 3:2 pairing, in which FSB cycles don't always match the memory cycles...

And if you can actually get Dominator DDR2-1067 @ 4-4-4 latency you'd have a latency of 2 FSB cycles as perceived by the CPU!... :twisted:
October 28, 2006 3:40:09 PM

You will have no problems running 800MHz, however, I would recommend sticking with 533MHz. The performance increase with 667 for instace is actually worse in some situations, but 800 is a bit better than 533 but honestly, if you dont plan on OCing, dont get the 800, you will see such a small difference in performance that its better IMO to get the 533 which should be cheaper. The only reason you will ever need anything over 800 is for some heavy duty OCing, stick with 533 and save the money.
October 28, 2006 3:41:13 PM

ok first off, im a newb but ive been trying to comprehend this info as well. and this is my take.

all the experts think geting a cd2 and not overclocking is like geting a ferrari and never going over 25mph, and from the results it seems the cd2's are over clocking very easily.

now from what get really the lower speed memory is ideal keeping with the 1-1, however, if you want to overclock thats where the better memory shines and brings you the best peformance.

i personally have ddr2 pc2-8000, which is 1000mhz, im going to change my multiplyer down to 8, crank the clock speed to 400 (which gives you a 1600mhz fsb) and turn my memory down to 400(aka x2 800mhz) and use better cas.

this will give me a 3.2 proc, 1-1 ratio, nice cas, fast fsb.

look in overclocking check for the sticky at the top wusys guide.

again im barely at the understaning of this concept so definatly read wusys info hes the real genius on this stuff.
October 28, 2006 3:45:22 PM

Morning to all.
I found this after some stumbling around looking for the last box of Fruit Loops:
Quote:
Common to both Intel and AMD systems is that you must set three things: FSB, latencies and ratio. Ratio? Right, ratio. RAM can work in either synchronous or asynchronous mode. In synchronous mode, the RAM and CPU operate in a 1:1 ratio, i.e., with a 200 mHz FSB, the RAM operates at 400 mHz (DDR) and the CPU at 800 mHz (QDR). That introduces some inherent efficiencies, since the data cycles are in tandem: 1 cycle for the motherboard = 2 cycles for the RAM = 4 cycles for the CPU. On the other hand, for reasons discussed later in this Guide, you may want to operate the RAM asynchronously, or out-of tandem, at a lower percentage than the 100% or 1:1 FSB ratio.

For example, you may want or need to operate the RAM at 80% of the 1:1 speed, at a 5:4 ratio or at a DDR rate of 320 mHz assuming a 200 mHz FSB. Here the RAM speed is 2 x 200 x .8 = 320 mHz. The ratio is set in the motherboard's BIOS, and sometimes it is expressed as an overt ratio, other times it may be expressed as a frequency relative to a 200 mHz FSB. Thus, 2:1 or 50% of the synchronous rate may be expressed as a "2:1", "1:2" or "200", depending on your motherboard's BIOS settings nomenclature. It is very important to know and understand what each setting in your motherboard's BIOS really means, and how each should be interpreted. Remember the old saying, "If all else fails, read the instructions!"? It is even more important to heed this advice if you intend to overclock safely.


Full article here:
http://www.abxzone.com/abx_reviews/al2/article_p2.html

Also look at this:
http://www.andrew.cmu.edu/user/fma/P4tweakRAM3.htm

And this:
Quote:
CPU

The frequency at which a processor (CPU) operates is determined by applying a clock multiplier to the front side bus (FSB) speed. For example, a processor running at 550 MHz might be using a 100 MHz FSB. This means there is an internal clock multiplier setting (also called bus/core ratio) of 5.5; the CPU is set to run at 5.5 times frequency of the front side bus: 100 MHz x 5.5 = 550 MHz. By varying either the FSB or the multiplier, different CPU speeds can be achieved.

Memory

Setting a FSB speed is related directly to the speed grade of memory that a system must use. The memory bus connects the northbridge and RAM, just as the frontside bus connects the CPU and northbridge. Often, these two buses must operate at the same frequency. Pushing the front-side bus to 170 MHz means pushing the memory to 170 MHz in most cases.

In newer systems, it is possible to see memory ratios of "4:5" and the like. The memory will run 5/4 times as fast as the FSB in this situation, meaning a 133 MHz bus can run with the memory at 166 MHz. This is often referred to as an 'asynchronous' system. It is important to realize that, due to differences in CPU and system architecture, overall system performance can vary in unexpected ways with different FSB-to-memory ratios.

In complex image, audio, video, gaming, and scientific applications where the data set is large, FSB speed becomes a major performance issue. A slow FSB will cause the CPU to spend significant amounts of time waiting for data to arrive from system memory.


And finally this...
http://arstechnica.com/guides/askars/ask-ars-20040710.ars

If it helps, you're welcome. If it doesn't.... meh. Now where is the milk?....
October 28, 2006 3:52:47 PM

Ive never heard of that, I always thought that 1:1 was the way to go.

What benefits could one see from 5:4, I mean, your basically underclocking you memory, is it to improve performance or stability :? ?
October 28, 2006 4:01:02 PM

I'm not sure, but I'll go out on a limb and say yes.
October 28, 2006 4:02:23 PM

Quote:
Hmmm that's what I understood from that too! DDR2-800 will result in a performance increase, it's OK for you do to so.

Though I must say that running 2:1 with DDR2-1067 isn't, mathematically, an asynchronous operation! The cycles match perfectly.

That's why I don't understand why there isn't more widespread support for DDR2-1067. It would be an excelent pairing, because the FSB cycles would always match 2 memory cycles! It's not like 5:4 or 3:2 pairing, in which FSB cycles don't always match the memory cycles...

And if you can actually get Dominator DDR2-1067 @ 4-4-4 latency you'd have a latency of 2 FSB cycles as perceived by the CPU!... :twisted:
[/i][/i]


This topic should be a sticky - need to research if it's already.
Plus - depending if AMD/Intel - there's a different view.

The 2:1 ratio is an interesting point. Jack - Wusy - any remarks?
October 28, 2006 4:02:37 PM

Not sure, but I think its different for Intel / AMD platforms.

The ratio would help AMD system for more stablity, in OC'ing.

For Intel, perhaps lil more performance (difference in secs for benchmarks) in OC'ing.

I found this review, that help show the peak performance of 533/667/800/1066 speeds. It doesn't explain much of the ratios, but is more based upon CAS timings, which low timing tend to help C2D performance:

Tweaking Core 2 For More Performance
October 28, 2006 4:29:51 PM

Good idea.
If there is such a sticky, I hope it will include RAM timings. Such as what does 4-4-4-16 mean? Is 3-3-3-16 better than 5-5-5-8 ? Is 4-4-4-16 800 Mhz better than 2-2-2-3 at 533 Mhz? How to get as fast as possible without losing stability ? I least I know the difference between DDR and DDR2.
October 28, 2006 5:14:56 PM

Hi,
In all this talk of memory speed, no one has really stressed the important issue of memory timings. Speed in MHz and latency timings are inversely proportional!
So, say you buy PC2-6400 which is rated at 400MHz (800DDR).
This memory, let's use my memory for example, will have 5-5-5-15 timings stored in SPD - the memory will do faster timings but this is JEDEC standard SPD @ 1.8V.
Now, at 333MHz (666DDR) the SPD timings change to 4-5-5-13.
At 266MHz (533DDR) the timings change to 3-4-4-10.
So basically, PC2-6400 memory can be PC2-4200 with very fast timings! At the slower, synchronous 1:1 (FSB:memory) speed, the memory bandwidth is entirely adequate but the timings are far superior.
Sure, you could just get PC2-4200 memory but it will be unlikely to offer 3-4-4-10 timings @ 1.8V!
Therefore, 'faster' memory will give more benefit to you at a speed that is less that its maximum rating (since that speed is all the CPU requires) and the big advantage of the pricier memory then becomes much tighter latency timings.
My PC2-6400 runs at 266MHz (533DDR) using 3-3-3-8 @ 2.0V - this is very quick!
At risk of stating the obvious, latency is a time delay measured in clock cycles. 3-3-3-8 is good; 5-5-5-15 is not so good.
Excessive memory bandwidth in Core2 (beyond the bandwidth of the CPU) seems to be beneficial only to memory bandwidth benchmarks...
Whereas 3-3-3-8 timings, with complete stability, are guaranteed to help out anyone's rig :^)
Hope this helps. Does it make sense? Remember inversely proportional, it is a recurring theme in science.
Just like going up, up into the mountains - the temperature drops.
The altitude and temperature are therefore said to be inversely propotional,
Regards
October 28, 2006 5:17:18 PM

Quote:
Hi,
In all this talk of memory speed, no one has really stressed the important issue of memory timings. Speed in MHz and latency timings are inversely proportional!
So, say you buy PC2-6400 which is rated at 400MHz (800DDR).
This memory, let's use my memory for example, will have 5-5-5-15 timings stored in SPD - the memory will do faster timings but this is JEDEC standard SPD @ 1.8V.
Now, at 333MHz (666DDR) the SPD timings change to 4-5-5-13.
At 266MHz (533DDR) the timings change to 3-4-4-10.
So basically, PC2-6400 memory can be PC2-4200 with very fast timings! At the slower, synchronous 1:1 (FSB:memory) speed, the memory bandwidth is entirely adequate but the timings are far superior.
Sure, you could get PC2-4200 memory but it will be unlikely to offer 3-4-4-10 timings @ 1.8V!
Therefore, 'faster' memory will give more benefit to you at a speed that is less that its maximum rating (since that speed is all the CPU requires) and the big advantage of the pricier memory then becomes much tighter latency timings.
My PC2-6400 runs at 266MHz (533DDR) using 3-3-3-8 @ 2.0V - this is very quick!
At risk of stating the obvious, latency is a time delay measured in clock cycles. 3-3-3-8 is good; 5-5-5-15 is not so good.
Excessive memory bandwidth in Core2 (beyond the bandwidth of the CPU) seems to be beneficial only to memory bandwidth benchmarks...
Whereas 3-3-3-8 timings, with complete stability, are guaranteed to help out anyone's rig :^)
Hope this helps. Does it make sense? Remember, inversely proportional (it is a recurring theme in science).
Just like going up, up into the mountains - the temperature drops.
The altitude and temperature are therefore said to be inversely propotional,
Regards


http://www.xbitlabs.com/articles/memory/display/core2du...
Agree with what you say except that some memory cannot do low latency even with low frequency.
October 28, 2006 5:43:52 PM

Yes, good points Jack. 8.4GB/s CPU bandwidth is matched perfectly by 2 x PC2-4200 in dual-channel (128-bit) mode.
Things are viewed differently in AMD-land since Socket AM2 is now used.
The basic clock is 200MHz and everything is derived, by multiplier or divider, from this reference.
When you O/C your AM2 and begin to increase the 200MHz (lots of folks running ~250MHz or whatever) everything increases: PCI, AGP/PCI-E, HT, and therefore memory.
The actual AM2 memory speed is derived from the CPU speed with a divider applied, since the memory controller is on the CPU. This is different than Intel where the memory speed is derived from the FSB with a multiplier applied.
However, bandwidth is bandwidth and latency is still latency...
Modern AM2 rigs have PCI locked at 33MHz and PCI-E at 100MHz - or they would NOT be so friendly to overclock!
Anyway, 2:1 in AM2-land means 200MHz (HyperTransport) with 400MHz memory (800DDR).
This is different than Intel 2:1 which for Core2 would be DDR2-533 @ 1066MHz.
Again, I urge folks to worry less about speed in MHz (bandwidth) but pay more attention to latency.
Everyone knows AMD AM2 has more memory bandwidth anyway, by design...
L8R
October 28, 2006 6:11:08 PM

Jack,

How's your build going with the x6800? What's in it? What are u clocking at?

I'm building mine in a few months. Waiting for the dust to settle on
graphics front with R600 and 8800series. And the 680i chipset, and
DFI's new MB with ATI's RD600 chipset.
October 28, 2006 6:22:17 PM

Quote:
At risk of stating the obvious, latency is a time delay measured in clock cycles. 3-3-3-8 is good; 5-5-5-15 is not so good.


Now let me state something not as obvious: if it's measured in clock cycles, then the actual latency will depend on the cycle rate - read, frequency - as well.

DDR2-800 @ 4-4-4-12 has the same response time as DDR2-533 @ 2.7-2.7-2.7-8 if such a thing existed.

Corsair's Dominator DDR2-1111 @ 4-4-4-12 has the same response time as DDR2-533 @ 1.9-1.9-1.9-5.8 - which kind of looks much better than DDR2-533 @ 3-3-3-8, doesn't it?

Now what people fail to realize is that there seems to be a performance advantage in some applications even if the CPU's FSB bottlenecks the actual bandwidth number available for the system memory. This is probably a memory controller inefficiency being countered by ultra high bandwidth... So while in theory DC DDR2-533 is already enough, in practice you can get more from DC DDR2-800 or DDR2-1067!!

As an example, I've seen a Pentium D 805 running at 20x200 (4Ghz) and with DDR2-800 running a memory-intensive benchmark 80% faster than with DDR2-400 and 20% faster than with DDR2-600, but both DDR2-600 and DDR2-800 actually saturate the bus!...

Here's evidence that speeds beyond DDR2-533 show performance increases, even if bandwidth isn't matched:

From Anandtech: "memory scales nicely through the various speed options."

From Anandtech (2): C2D can show performance increases, though not as linearly as A64s. See for yourself.

From Xbitlabs: Here, there is an observed bandwidth increase of up to 17% when going upwards in DDR2 frequencies. This is obviously not optimal, but it still is a performance increase. Also, the higher-frequency modules have the lowest overall latencies, as I explained above.

So why do people keep saying that there's no point in upping to DDR2-800? Even if you stay at stock speeds, there seems to be some advantage... Granted, not that much, but it's not just one or two percent!!
October 28, 2006 6:33:17 PM

Quote:
At risk of stating the obvious, latency is a time delay measured in clock cycles. 3-3-3-8 is good; 5-5-5-15 is not so good.


Now let me state something not as obvious: if it's measured in clock cycles, then the actual latency will depend on the cycle rate - read, frequency - as well.

DDR2-800 @ 4-4-4-12 has the same response time as DDR2-533 @ 2.7-2.7-2.7-8 if such a thing existed.

Corsair's Dominator DDR2-1111 @ 4-4-4-12 has the same response time as DDR2-533 @ 1.9-1.9-1.9-5.8 - which kind of looks much better than DDR2-533 @ 3-3-3-8, doesn't it?

Now what people fail to realize is that there seems to be a performance advantage in some applications even if the CPU's FSB bottlenecks the actual bandwidth number available for the system memory. This is probably a memory controller inefficiency being countered by ultra high bandwidth... So while in theory DC DDR2-533 is already enough, in practice you can get more from DC DDR2-800 or DDR2-1067!!

As an example, I've seen a Pentium D 805 running at 20x200 (4Ghz) and with DDR2-800 running a memory-intensive benchmark 80% faster than with DDR2-400 and 20% faster than with DDR2-600, but both DDR2-600 and DDR2-800 actually saturate the bus!...

Here's evidence that speeds beyond DDR2-533 show performance increases, even if bandwidth isn't matched:

From Anandtech: "memory scales nicely through the various speed options."

From Anandtech (2): C2D can show performance increases, though not as linearly as A64s. See for yourself.

From Xbitlabs: Here, there is an observed bandwidth increase of up to 17% when going upwards in DDR2 frequencies. This is obviously not optimal, but it still is a performance increase. Also, the higher-frequency modules have the lowest overall latencies, as I explained above.

So why do people keep saying that there's no point in upping to DDR2-800? Even if you stay at stock speeds, there seems to be some advantage... Granted, not that much, but it's not just one or two percent!!

I think it is just not cost-effective for DDR2-800.
October 28, 2006 6:46:34 PM

qcmadness: The latest technology with that last 10-15% performance increase is never cost-effective. If it were so, noone in their right mind would buy an "Extreme" Processor, an overkill-motherboard, enormous cooling solutions and over-the-top memory - it just isn't cost effective. Yet people buy these things anyway. Silly people. :D 

JumpingJack: Exactly, that's what I meant. Plus, from the memory prices nowadays, you'll pay a 30%-40% price premium for CL4 DDR2-800, but you'll get better latency and better real-world bandwidth. If you need bandwidth, then it's justified. Heck, if you get the E6700 instead of the E6600 then you'd get only a 11% increase in clock for 50% more money! And that's not even mentioning the C2X @ 2.93Ghz, which gives you 9% more frequency for nearly twice the price!!

If you think about it, paying a 30% premium for DDR2-800 CL4 over DDR2-533 and getting better latencies and better bandwidth doesn't sound so dismissable for a high-end system. Sure, DDR2-533Mhz is probably enough for joe sixpack, but not for a high-end enthusiast system...
October 28, 2006 8:15:12 PM

Quote:
Hi,
In all this talk of memory speed, no one has really stressed the important issue of memory timings. Speed in MHz and latency timings are inversely proportional!
So, say you buy PC2-6400 which is rated at 400MHz (800DDR).
This memory, let's use my memory for example, will have 5-5-5-15 timings stored in SPD - the memory will do faster timings but this is JEDEC standard SPD @ 1.8V.
Now, at 333MHz (666DDR) the SPD timings change to 4-5-5-13.
At 266MHz (533DDR) the timings change to 3-4-4-10.
So basically, PC2-6400 memory can be PC2-4200 with very fast timings! At the slower, synchronous 1:1 (FSB:memory) speed, the memory bandwidth is entirely adequate but the timings are far superior.
Sure, you could just get PC2-4200 memory but it will be unlikely to offer 3-4-4-10 timings @ 1.8V!
Therefore, 'faster' memory will give more benefit to you at a speed that is less that its maximum rating (since that speed is all the CPU requires) and the big advantage of the pricier memory then becomes much tighter latency timings.
My PC2-6400 runs at 266MHz (533DDR) using 3-3-3-8 @ 2.0V - this is very quick!
At risk of stating the obvious, latency is a time delay measured in clock cycles. 3-3-3-8 is good; 5-5-5-15 is not so good.
Excessive memory bandwidth in Core2 (beyond the bandwidth of the CPU) seems to be beneficial only to memory bandwidth benchmarks...
Whereas 3-3-3-8 timings, with complete stability, are guaranteed to help out anyone's rig :^)
Hope this helps. Does it make sense? Remember inversely proportional, it is a recurring theme in science.
Just like going up, up into the mountains - the temperature drops.
The altitude and temperature are therefore said to be inversely propotional,
Regards


I want you to be my teacher. I really understood everything you said. You really know how to explain things. So I still have some questions left... do you think I can do that with my Corsair2X1024 1Gb DDR800 PC2-6400? Lower the speed from 400mhz to 266mhz so it can be synchronus with my Core 2 Duo FSB and at the same time low the latencies from 4-4-4-12 to maybe 3-3-3-8?

PD: When I finish building my rig, I will run 3dmark (general test) with everything on default. Then I will write down the score and restart the pc, lower the memory speed from 400mhz to 266mhz and the timings if possible to 3-3-3-8. Then Ill run 3d mark (general test) again and check the score. If the modified test (with lower latencies and lower speed) is WAY higher than the default, then I will just leave it modified (Only if it is way higher). But if the default configuration (with normal timings and normal speed) is higher with a reasonable score then Ill leave it in default.

My question is... which one will perform better and by how much approx? Also, will I need to INCREASE voltage in the ram while modifying the timings and lowering the speed? Because I dont really know how to do that. Also, will I need to change anything in the FSB? or modify anything in the CPU/FSB section?
October 29, 2006 12:21:54 AM

Think of it in terms of gear ratios.You have a 1 to 1 gear ratio,then your front gear is the same size as the rear gear bothe running at the same speed.when you get a 2 to 1 gear ratio,it means the front gear is half the size of the rear one making it turn twice the speed of the rear one,thereby giving you a 2 to 1 gear ratio.Now with cpu to memory you have the cpu running a 200 mhz.fsb and your memory is running at 200 mhz.,thereby giving you a 1 to 1 ratio.Got it so far?Good.Now a 2 to 1 gear ratio is when your cpu runs at 200 mhz. and your memory runs at 166 mhz. thereby giving you a 2 to 1 ratio.Doing this enables you to clock higher by giving your memory more head room so to speak.Of corse there is a bit more to overclocking,but that is the most basic example I can give for ratio's.Goodluck.

Dahak

AMD X2-4400+@2.4 S-939
EVGA NF4 SLI MB
2X7800GT IN SLI
2X1GIG DDR IN DC MODE
WD300GIG HD
EXTREME 19IN.MONITOR 1280X1024
ACE 520WATT PSU
COOLERMASTER MINI R120
October 29, 2006 12:25:24 AM

Hmmm you're just overdoing it right now. You're worrying too much.

Firstly, you're considering lowering your DDR2 speed to 2/3rds of its normal speeds - which'll probably reduce real-world bandwidth - and you're trying to make up for it by setting 3-3-3 latencies, which are worse than 4-4-4 @ DDR2-800 speeds (see my previous posts), all because of what is, in this context, a myth.

You will, in all likelyhood, not see any performance advantage. Think about it: higher latencies, lower bandwidth... only for 1:1 operation, which won't give you the boost you're expecting.

Secondly, you shouldn't be worrying about what exactly each configuration will give you in terms of performance right now. You can always try all this later on: you're working too much on the hypothetical side of things here. In theory, we can say a lot of crap, but it always comes down to what you can measure in the end...

So I guess what I'm saying is "take it easy"... :D 
October 29, 2006 12:32:55 AM

Well yeah but Im just asking because I dont want after finishing building my rig, have a great pc but with a small bottleneck between the FSB and memory bandwidth.

Ok so I guess now I understand more about this 1:1 and I will try it myself when I get my pc. I will try not to mess with voltage because I really get scare when thinking of blowing my pc :p 
October 29, 2006 12:39:44 AM

Im a gamer, so in what zones MIGHT I see bottlenecks?
October 29, 2006 1:19:50 AM

Will there be any disavantage with a 7900GS KO w/ HDCP?
October 29, 2006 1:39:42 AM

Im going to use 1280x1024

I prefer to stay with NVIDIA, I think the 7900GS KO is a good card.
October 29, 2006 2:23:25 AM

Interesting results for your benchmarks, but I think there's more to this memory issue than these results seem to indicate.

There are bandwidth-sensitive programs out there, but they're probably not games. So, if you're a gamer, you shouldn't really worry too much about your memory, right, but if you're a power user and if you're also concerned about performance elsewhere, there's plenty of code that utilizes system memory in much more stressful ways than games...

Highly interesting results nonetheless: we should try to avoid overestimating the effect memory has on the C2D platform!!
a c 87 à CPUs
a b } Memory
October 29, 2006 3:38:47 AM

That Wiki article is wrong, I should go change it...

Quote:
The frequency at which a processor (CPU) operates is determined by applying a clock multiplier to the front side bus (FSB) speed.


Not true. I also thought this way, until someone corrected me. The speed of the CPU is the multiplier times the "base frequency". The FSB is the "base frequency" times the number of bits that can be transferred per clock. HUH? Let me give some real examples.

http://www.newegg.com/Product/Product.asp?Item=N82E1681...
Here is an x6800 C2D Extreme. 2.93GHz clock, runs on a 1066FSB. If the Wiki article was correct, you would get a multiplier of 2.75. Ram would be double the FSB, and you'd need 2133GHz ram. (Holy Sh!t) Anyone who runs one of these knows you need to set the "fsb" to 266MHz. This is because Intel "quadpumps" their FSB. (two bits on the front and back of the wave.) This math makes a lot more sense. 1066MHz / 4 = 266MHz. 2.93GHz / 266MHz = 12.5 Multiplier, which is one I've heard of. The ram would be double the 266, which means you'd need DDR2-533 ram, which is again something you can find.

Many people mistakenly call this base frequency the FSB. As I mentioned at the start, this is wrong. The base frequency determines the FSB, CPU clock speed, and ram speed. CPUs haven't used a non pumped base frequency in quite some time. The p4 was the first to use "pumping". The base frequency fell from 133MHz on the P3, to 100MHz on the P4 willy core. The FSB actually rose however from 133MHz to 400Mhz effective

As for ram, at a 1:1 ratio, it runs double of what the base frequency is. If you overclock the x6800s base frequency from 266 to 300, you now have a CPU running at 3.75GHz. (300 * 12.5) If you have the ram set at 1:1, you now need DDR2-600 ram. (300MHz * 2) This might be more then DDR2-533 can handle, but DDR2-667 should be ok.
October 29, 2006 12:21:50 PM

In overclocking the FSB - what are the limiting factors - I'm waiting to
see what the 680i board will do.

In setting a high FSB - How much is the MB? How much is it the CPU?
October 29, 2006 12:58:41 PM

Quote:
That Wiki article is wrong, I should go change it...

The frequency at which a processor (CPU) operates is determined by applying a clock multiplier to the front side bus (FSB) speed.


Not true. I also thought this way, until someone corrected me. The speed of the CPU is the multiplier times the "base frequency". The FSB is the "base frequency" times the number of bits that can be transferred per clock.

I think Wikipedia and you are correct actually. Perhaps it is not fully describing everything (todays technology), but it still applies to the older technology before the P4 and XP Athlons were out.

Did you know the FSB, is also known as the system bus?

What I'm contrasting there, is there are so many different discriptions that mean the same thing, so there can be confusion if one doesn't totally understand their meanings.

Calling it the base frequency doesn't really establish a particular bus. I mean just as you explain the memory speed has a base frequency that is doubled, or ran by dual channels.

Perhaps they should re-define it. Like the system bus would be better known as the base frequency of that bus, and the FSB would be better known as the total or quad frequency bus (who knows, they may increase it more then the quad pumped bus later on).

LOL, ah hell it's all confusing... memory timings, or CAS, quad pump, dual core, MHz, Ghz, MT/s, GT/s, the doohicky, and the whachamacallit.
October 29, 2006 1:27:18 PM

Well, actually, the FSB never runs at 1067Mhz from a purely technical point of view, and neither does DDR2-800 work at 800Mhz.

DDR2, for instance, has a 4-bit prefetch, which means that it transfers 4 bits of information every cycle. DDR2-800 makes use of a 200Mhz clock generator and transfers 4 bits every cycle, effectively transferring information at a rate of 800 million transfers a second (hence MT/s), though in terms of definition of the word "frequency", you could say that data gets transferred with a frequency of 800Mhz, 800 times a second. However, there is no clock generator ticking away at 800Mhz, nor is there any clock generator ticking at 1067Mhz for the Conroe FSB.

If you look at it in a conservative way, 1067Mhz FSB is actually a marketing perspective. From an engineering point of view, there simply aren't that many full cycles a second.

The "base clock" is the actual full cycle speed for the motherboard, not the other numbers. The confusion arises because 1) the people who don't know this obviously get confused and 2) this is so natural to the people who know this that they don't bother clearing up all the misconceptions 3) the words "frequency" and "mhz" can refer to either full cycles (from engineering point of view) or bit transfer rate (the number that marketing wants you to look at), and neither is factually wrong. The problem is that, while none of these uses is a lie, that doesn't make this any less confusing. Get it?

So that's why there is a lot of confusion around... :?
October 29, 2006 1:38:43 PM

Quote:
MHz, Ghz, MT/s, GT/s


These were all created because people wanted to clear things out. This has been discussed to death in this forum before (mainly because of P4's at-that-time-newly-introduced FSB). In my opinion though, MT/s and GT/s is a sign of more confusion, not less. My previous post explains that the use of Mhz is justified in every case here, but this was sometimes used to measure frequencies of different events (full cycles or bit transfers) without any discretion, hence the confusion. Current FSBs have 1066Mhz bit transfer speeds and 266Mhz full cycle speeds.

The initial mess was created when both Intel and AMD started stating bit transfer frequencies instead of full cycle frequency. To me, it seems they should just have marketed 4x100Mhz, 4x133Mhz and 4x200Mhz... But hey, that's just me... :?
October 29, 2006 1:45:19 PM

Well, I think the blame for misconception or confusion for people other than engineers, is prolly more of the marketing.

What sounds better, and would sell better?

Base Freq of the system bus 266mhz listing on the specs? Or the whopping FSB 1066mhz on the front label?

:lol: 
a c 87 à CPUs
a b } Memory
October 29, 2006 4:17:38 PM

Quote:
In overclocking the FSB - what are the limiting factors - I'm waiting to
see what the 680i board will do.


I would assume the same limiting factors apply to overclocking here. The first issue is heat, the second is quality of the part. (both in the design of the part, and the specific one you have.) I have no idea how well the 680i boards will overclock, hopefully better then the 570/590 boards.

Quote:

In setting a high FSB - How much is the MB? How much is it the CPU?


GRRR, did you ever read my post? Depending on how high you want to go, much of it is the MB. As you pointed out, the 570/590 chipset doesn't overclock as well as the 965/975 chipset. I assume the reviewers use the same CPU, only the motherboard was changed.
a c 87 à CPUs
a b } Memory
October 29, 2006 4:28:25 PM

Quote:


I think Wikipedia and you are correct actually. Perhaps it is not fully describing everything (todays technology), but it still applies to the older technology before the P4 and XP Athlons were out.


Alright, you got me there. The wiki isn't wrong, just outdated.

Quote:
Did you know the FSB, is also known as the system bus?


That would be what FSB stands for, the "Front System Bus". In case anyone was wondering, there used to be a back system bus that was used to connect external L caches. (L2 or L3)

Quote:
What I'm contrasting there, is there are so many different discriptions that mean the same thing, so there can be confusion if one doesn't totally understand their meanings.


But thats what I was trying to point out. The FSB and the "base frequency" are different beasts. The FSB is the "way out" for the CPU. Its the path (bus) that Intel and older AMD CPUs use to communicate with the NB. This allows communication with the system RAM, and all other devices. The base frequency is not a bus at all. No communication travels on the base frequency. It its simply a timer, that sets the speed of all devices in the computer. It might be called something else, base frequency is simply what I learned it as.
a c 87 à CPUs
a b } Memory
October 29, 2006 4:33:16 PM

Quote:
Well, actually, the FSB never runs at 1067Mhz from a purely technical point of view, and neither does DDR2-800 work at 800Mhz.

DDR2, for instance, has a 4-bit prefetch, which means that it transfers 4 bits of information every cycle. DDR2-800 makes use of a 200Mhz clock generator and transfers 4 bits every cycle, effectively transferring information at a rate of 800 million transfers a second


Thats why in my first post I said effective. (I even put that word in italics) I also mentioned Intels quadpumping. AMDs Socket A used Double pumping. (200MHz base clock with two bits per tick.) 400MHz effective FSB.
October 29, 2006 4:43:38 PM

Quote:
Did you know the FSB, is also known as the system bus?


That would be what FSB stands for, the "Front System Bus". In case anyone was wondering, there used to be a back system bus that was used to connect external L caches. (L2 or L3)



Ummm... FSB is Front Side Bus, not Front System Bus.
a c 87 à CPUs
a b } Memory
October 29, 2006 4:54:13 PM

What does SQL stand for? Standard query language or Structured query language? Does the fact that the first word is different mean your talking about something else? Whether its System or Side, I don't think that really matters. (Its not like using one or the other means something else.) We are both talking about the bus that the CPU uses to communicate with the NB. I would be far more concerned if this was like PCIx. Where some people think that means PCI express.
October 29, 2006 5:34:48 PM

Didn't mean to upset you.

FSB is Front Side Bus, technically.

SQL is Structured Query Language, technically.

Again, there's so many ways to discribe certain things in particular.

I mean, some people think CPU stands for something else then Central Processing Unit... :lol: 

Misunderstanding do happen, so don't take it too personal.
October 29, 2006 5:50:16 PM

Your DDR2-800 won't run 1:1 with your CPU as a pair of DDR2-533 modules would. However, because the DDR2-800 is faster, it is able to sort through the inconsistencies and still provide increased performance while running outside of 1:1.
a c 87 à CPUs
a b } Memory
October 29, 2006 6:13:01 PM

Quote:
However, because the DDR2-800 is faster...


How is DDR2-800 faster then DDR2-533 when being run at 1:1? Its no longer DDR2-800, but DDR2-533 (at stock speeds). The only advantage I can see is you might be able to lower the timings if you use the faster rated ram.
October 29, 2006 6:19:40 PM

Thanks a lot. More reasons to calm myself :) 
October 29, 2006 8:36:17 PM

Quote:
How is DDR2-800 faster then DDR2-533 when being run at 1:1? Its no longer DDR2-800, but DDR2-533 (at stock speeds). The only advantage I can see is you might be able to lower the timings if you use the faster rated ram.


I'm saying that when the DDR2-800 runs at its rated speed (800MHz), it is obviously faster than the 533MHz RAM. Because of this advantage in speed, it's fault in not being able to run 1:1 is superseded by the faster operational speeds of the modules. Information won't match up when things are not running synchronously, so it will take more signals sent by the processor in order for it to be interpreted correctly by the RAM. But, because the RAM operates faster, the bottleneck isn't as apparent in real word situations. It sorts through the inconsistencies at a fast enough pace where it is surely faster (even in 3:2 ratio) than 533MHz (in 1:1).
October 29, 2006 9:50:48 PM

Quote:
In overclocking the FSB - what are the limiting factors - I'm waiting to
see what the 680i board will do.


Quote:

In setting a high FSB - How much is the MB? How much is it the CPU?


GRRR, did you ever read my post? Depending on how high you want to go, much of it is the MB. As you pointed out, the 570/590 chipset doesn't overclock as well as the 965/975 chipset. I assume the reviewers use the same CPU, only the motherboard was changed.

I did read the post. It's just that many were disappointed that the x6800
didn't come out as the 1333MHz version. So, if it's mostly a MB issue,
it's no big deal. I'm doing the Quad when it comes out, so it might be nice
to up the FSB.
October 30, 2006 1:00:04 AM

Quote:
Thats why in my first post I said effective. (I even put that word in italics) I also mentioned Intels quadpumping. AMDs Socket A used Double pumping. (200MHz base clock with two bits per tick.) 400MHz effective FSB.


I wasn't really correcting you, I was justing adding a little more info... :) 
!