jjknoll

Distinguished
Sep 13, 2006
225
0
18,680
Why is it that supposedly ram works best at 1:1 ratio with the FSB?
Do prebuilt systems most always come set as a 1:1 ratio?
If so, are vendors such as Dell only taking your money when they offer high speed memory with their systems? Is it not being utilized?
Currently I am running 800mhz ram with my e6600. I read that my mobo's FSB tops out around 310mhz. Does this mean that on this mobo I will never utilize 800mhz memory? Even if I OC?
If so would it be advisable to buy some 667 with tight timings? How much of a difference will it make going from cas5 to cas3?
Also, I only hvae a single stick of ram right now. Is there a performance increase running dual channel? Or is it rather insignificant?
Thanx for any help/info.
 

MOREPOWERPLEASE123

Distinguished
Jun 14, 2006
19
0
18,510
There is a HUGE performance difference when running memory in dual channel. As far as timings are concerned, it doesn't seem to make as big of a difference on Intel cpu's as what it does on AMD.
 

yakyb

Distinguished
Jun 14, 2006
531
0
18,980
with your ram being underutilised you may find that you can tighten your timings a bit i.e ddr 800 @ 5,5,5,15 may well run fine under 4,4,4,12 @ddr667 speeds it may not do however but that is why the CMOS has a reset option
 

jjknoll

Distinguished
Sep 13, 2006
225
0
18,680
what about the 1 to 1 ratio though, as I am still curious. it seems that people recommend nto doing 8 to 9 ratio or the like. why is that? what makes it bad?
 

Mondoman

Splendid
Current Mainstream Intel CPUs (CMICs) transfer data 4 times per FSB clock cycle. The memory bus feeds into the FSB on CMICs, so you want to make sure the FSB is full, not waiting for data from a slower memory bus. DDR and DDR2 memory transfer data only 2 times per memory bus clock cycle. However, when run in dual channel mode, two streams of memory data are transferred at once, doubling effective throughput. Thus, we want:
membusclock x2 (DDR) x2 (dual-channel) = FSB x4 (CPU quad-pumped) to supply enough throughput for the FSB. Even faster memory bus speeds can reduce latencies, but the gains are incremental.
 

jjknoll

Distinguished
Sep 13, 2006
225
0
18,680
so in review memory speeds that are slightly faster than 1 to 1 ratio with the fsb are not necessarily a bad thing but it is more or less crucial that you run memory in dual channel? Thanx for the help.
 

zjohnr

Distinguished
Aug 19, 2006
577
1
18,980
so in review memory speeds that are slightly faster than 1 to 1 ratio with the fsb are not necessarily a bad thing but it is more or less crucial that you run memory in dual channel? Thanx for the help.
I guess I'd prefer to characterize it in terms of the relative amount by which performance increases. Dual channel results in a much larger relative performance gain than, say, going from DDR2-533 to DDR2-800. Enabling dual channel is something that I think would always be recommended. Going with DDR2-800 is more of a "If you've got it, there's no reason not to use it".

It seems to be a natural desire of people to want to always buy the fastest memory which their budget will allow. But focusing on faster memory alone is not the most effective way to improve the performance of a Core 2 based PC. The real-world performance (as opposed to a "synthetic" memory benchmark) gained by increasing the memory speed alone is not typically worth the extra money spent on it.

That said, I'm still find myself wondering. Would paying an extra 20% or so to get DDR2-800 instead of DDR2-667 be worth it for other reaons? Potential perceived resale value? Or that old phantom we all seem to chase, "future proofing"? Am I an idiot or what? :oops: :roll: :wink:

FWIW, I also remember seeing a puzzling benchmark that showed (on at least one Core 2 system) running the FSB at the stock 266MHz and the memory at DDR2-667 resulted in a slight performance drop, the opposite of what you'd expect. :? 8O Wish I could remember where I saw that ... somewhere on AnandTech I think. I also wonder if they ever followed up on it to find out in more detail why it was happening. At this point I think that result will end up as just a briefly lived Core 2 urban legend. Maybe I should look for it on www.snopes.com ? :)

-john, the ostensibly clueless redundant legacy dinosaur
 

zjohnr

Distinguished
Aug 19, 2006
577
1
18,980
Here's the article. http://www.madshrimps.be/?action=getarticle&number=1&artpage=1962&articID=472
The "synchronization latency" explanation sounds plausible.
Thanks, Mondoman. While that is not the article I remember reading ... I really think it was somewhere on AnandTech ... I actually liked the overall presentation in this article better. Much easier to follow because the performance differences are all scaled to percent changes.

If the OP hasn't read this article, I'd definitely encourage him to. If nothing else at least skip to the end and read the conclusions. :wink:

Maybe I will go with DDR2-800 after all. :) Or at least if I do get DDR2-667 I'll run it at 1:1 speeds (i.e. DDR2-553 unless/until I decide to try to overclock).

-john, the ostensibly clueless redundant legacy dinosaur
 

Mondoman

Splendid
Actually, the conclusions in the article are poorly written and contradicted by the actual data shown in the article, so ignore the conclusions section. I should have included that caveat when I posted the link; sorry!
 

zjohnr

Distinguished
Aug 19, 2006
577
1
18,980
Actually, the conclusions in the article are poorly written and contradicted by the actual data shown in the article, so ignore the conclusions section.
Well, it went right past me. Could you expand a little on what is wrong in the conclusions section?

Edit: I'm not sure I agree with the assertion that
By running the memory synchronous to the FSB you have the least amount of latency and thus performance is at its best.
but that's more because I feel I don't really have a handle on the dynamics of how DRAM memory latency interacts with the MCH interacts with the FSB then because I know anything is wrong in that assertion. :? :)

-john
 

RJ

Distinguished
Mar 31, 2004
655
0
18,980
Why is it that supposedly ram works best at 1:1 ratio with the FSB?
Do prebuilt systems most always come set as a 1:1 ratio?
If so, are vendors such as Dell only taking your money when they offer high speed memory with their systems? Is it not being utilized?
Currently I am running 800mhz ram with my e6600. I read that my mobo's FSB tops out around 310mhz. Does this mean that on this mobo I will never utilize 800mhz memory? Even if I OC?
If so would it be advisable to buy some 667 with tight timings? How much of a difference will it make going from cas5 to cas3?
Also, I only hvae a single stick of ram right now. Is there a performance increase running dual channel? Or is it rather insignificant?
Thanx for any help/info.
Who says your mobo is only capable of 310 mhz? What exactly is it? How old is the article you read? When the first reviews of C2D's were done, they were using F5 cores. F6 cores clock much higher. Also there have been tons of BIOS upgrades to address overclocking stability. I suggest finding your motherboard in www.newegg.com and reading what some of those people are hitting for an overclock. Don't take it for "gospel" but you can read between the lines for what to expect. One chipset I know of that doesn't like a high fsb is the G965 chipset. I think it's because the PCI-E and FSB are linked somehow and PCI-E really doesn't tolerate a higher-than-stock clock. There also aren't voltage settings usually found on a G965.

Running a 1:1 ratio with DDR2-800 lets you run the CPU clock @ 400 mhz, and the memory is still stock speed. In reality with a 965 chipset, it's better to run 401 mhz, as I've come to learn recently. The link in my sig goes to an article on why it's so.
 

RJ

Distinguished
Mar 31, 2004
655
0
18,980
...
FWIW, I also remember seeing a puzzling benchmark that showed (on at least one Core 2 system) running the FSB at the stock 266MHz and the memory at DDR2-667 resulted in a slight performance drop, the opposite of what you'd expect. ...
Here's the article. http://www.madshrimps.be/?action=getarticle&number=1&artpage=1962&articID=472
The "synchronization latency" explanation sounds plausible.

In layman's terms, a 1:1 ratio means there's no multiplication/division that has to be done to the speed of the transfer between the FSB and the memory. It's like you getting onto the highway at 65 in a 65 mph zone. Nothing has to slow down to mesh.

I hope that helps.
 

Mondoman

Splendid
OK, they say there that
... in our testing we found performance actually decreases a bit going from PC4200 (533) to PC5300 (667)!
However, if you look at the various tests, that's not true. What they DID find was that DDR2-667 *CL5* was sometimes slower than most/all of the DDR2-533s tested. However, they also showed that DDR2-667 *CL3* was *always* faster than all the DDR2-533s tested, and faster than one or both of the DDR2-800s tested. DDR2-667 CL4 was in-between in performance.

IMHO, what this testing really showed was that latency (or at least CLx) is fairly important in C2D real-world performance. Stay away from CL5.
 

zjohnr

Distinguished
Aug 19, 2006
577
1
18,980
IMHO, what this testing really showed was that latency (or at least CLx) is fairly important in C2D real-world performance. Stay away from CL5.
I'd say that I think I see where you're going, but feel that would be giving me too much credit. But I do think I see some of what you're saying. I now see a problem with how I was viewing the charts. They are ordered in terms of increasing frequency and decreasing latency within a frequency. There is nothing on the charts to highlight when both the frequency and the latency increase which is what happens at the DDR2-667 "problem" points.

But there is still something unexpected going on here. If you compare DDR2-533(4-4-4-12) to DDR2-667(5-5-5-15) the total latencies should be approximately the same. Yet there is still a drop in the results for the Super-Pi and the Photoshop benchmarks. Strange.

-john, the ostensibly clueless redundant legacy dinosaur
 

Mondoman

Splendid
... If you compare DDR2-533(4-4-4-12) to DDR2-667(5-5-5-15) the total latencies should be approximately the same.
Yes, excellent point.
Yet there is still a drop in the results for the Super-Pi and the Photoshop benchmarks.
I think the idea is that with the mismatched bus freqs, there is an additional extra latency involved, perhaps in the memory controller as it converts input to output at a slightly different speed. Even faster memory bus speeds can make up for this, but the DDR2-667 isn't enough faster on throughput to make up for this.
 

jjknoll

Distinguished
Sep 13, 2006
225
0
18,680
I am running a P5N-32 sli se deluxe but it could possibly have been from older articles. But I thought that I had read from recent forum postings that this mobo is not the greatest for oc.
 

TRENDING THREADS