Dual channel poll

Here

Thought I'd get some more coverage here. Please delete if I'm doing anything wrong by posting this here.

Who's going to complain about you man. You help people day in day out around here.

Gautam said:

Also post if there's something else I should add in this poll.

Click to expand...

The operating system of course.
You're not going to get the same scores in XP as in 98/Me I don't think.

Thanks, c627627. Bumpity bump. Come on, I know that more of you guys are running in dual channel. In theory, all operating systems should report identical bandwidth, but theory doesn't always equal reality. It would be interesting to see how OS's measured up to one another. I know that in 3dMark 01 and 03, it can be a large consideration.

Gautam said:

In theory, all operating systems should report identical bandwidth, but theory doesn't always equal reality. It would be interesting to see how OS's measured up to one another.

Click to expand...

Windows XP is more stable but the only place you can see Windows XP beating Windows Me is benchmarks on web sites.

Install a dual boot Windows XP/Me on the same system and tell me that XP feels faster... you would be the first since 2001.

I have yet to install a dual boot Windows XP/Me where XP is faster in any benchmark, including your tests. It's a choice between speed of Windows Me and reliability of Windows XP.

1. 256MB PC2100 Kingston (Hynix chip) ; 256MB PC2100 Crucial ; 256MB PC2100 Kingston (Hynix chip)
2. 208 FSB with just two Kingston Hynix PC2100s but 204 FSB with all three
3. DIMM voltage regulator at default 2.63V
4. 6-2-2-2
5. nForce2 Epox 8RDA+ 1.1 (A2/A2)
6. CPU voltage regulator = 1.800 V

7. Sandra memory bandwidth score (Dual boot system, 3 x 256MB PC2100 @ 204 FSB):
Windows Me: Test 1: 2270 MB/s Test 2: 2258 MB/s
Windows XP: Test 1: 2207 MB/s Test 2: 2158 MB/s

Bandwidth efficiency (Dual boot system, 3 x 256MB PC2100 @ 204 FSB, tested, rebooted, tested again):
Windows Me: Test 1: 70% Test 2: 69%
Windows XP: Test 1: 68% Test 2: 66%

Proving once again that Windows XP is slower.

8. Added PC2100 Crucial in the middle, PC2100 Kingston Hynix stayed in slots 1 and 3.

_____________________
2100+ AIUHB 0248
[204] FSB x 11 = 2244 MHz
at 1.8 V @
37(min) C to 46(max) C Winter
43(min) C to 50(max) C Summer
Epox 8RDA+
Thermalright SK-7 with variable speed 80x25mm YS-Tech FD1281259B-2F
1x 256MB Crucial PC2100 +
2x 256MB Kingston (Hynix chip) PC2100 7 3 3 3 @ 6 2 2 2
BFG GeForce4 Ti4200 8X 128MB; Antec SX-835II case ; 350W Antec SmartPower SL350
Five 80x25mm Ultra Quiet 18CFM Case Fans (NMB C3110KL-04W-B19)

i hate ME so i guess i will just have to deal with the lack of speed

Yes yes of course. The only reason I'm using it is 1 minute 52 second restore with this:
http://forum.oc-forums.com/vb/showthread.php?s=&threadid=214369


Gautam, I'm guessing twinmos/winbond ch-5 pc3200 is the best RAM $ can buy.

What do you think 2 x 256MB of those can get me over my current setup?
Extra 50Mhz-100Mhz?
It's difficult to spend $100-$120 just for that so that's why I didn't bother.

Besides MHz, where else would I see improvement?

c627627... DUDE ! !
What in the name of Lucifer are you running on your rig to give you such low scores on sisoft? You're running three mem modules on your Epox board, so of course you're going to get low mem scores.... that's not even dual channel any more... try removing the stick of mem from that slot, and run your test again in accordance with epox's placement of the dual channel module placement. BTW, WinME blows chunks in comparison to WinXp, period. You'll also see your bandwidth efficiency go up...
Obviously there's a bit of ongoing background activity running on your rig while you're bench marking your sisoft scores. tsst tsst.

2nd set of results on my rig, which i've stepped down to 2.5GHz
Corsair XMS LowLatency 256mbx2

SiSoftware Sandra

Int Buff aEMMX/aSSE (Integer STREAM) Results Breakdown
Assignment : 3489MB/s
Scaling : 3489MB/s
Addition : 3469MB/s
Triad : 3472MB/s
Data Item Size : 8-bytes
Buffering Used : Yes
Offset Displacement Used : Yes
Bandwidth Efficiency : 95% (estimated)

Float Buff aEMMX/aSSE (Float STREAM) Results Breakdown
Assignment : 3488MB/s
Scaling : 3296MB/s
Addition : 3154MB/s
Triad : 3110MB/s
Data Item Size : 8-bytes
Buffering Used : Yes
Offset Displacement Used : Yes
Bandwidth Efficiency : 89% (estimated)

Test Status
Memory Used by Test : 256MB

Chipset 1
Model : Abit Computer Corp nForce2 AGP Controller
Front Side Bus Speed : 2x 228MHz (456MHz data rate)
Width : 64-bit
Maximum Bus Bandwidth : 3648MB/s (estimated)

This thread will be very useful for me then.

For starters I’ll remove the PC2100 Crucial stick from the middle and leave the Kingstons in slots 1 and 3 allowing me to crank up the FSB to 208.

Will do tests in both XP and Me, this time not running any applications in background. These are PC2100 sticks...

Agreement here about Windows ME being worse overall.
I have both XP and Me on my rig, it takes me like 50 seconds to switch between the two. All I'm saying is that I've used Me 80% of the time since 2001 if not more because it simply feels faster.

I would never do Photoshop or anything complicated on Me. You realize very quickly that it's more worth it to switch to XP and do it there.

For everyday use though, if I may exaggerate to make a point, switching to XP is like switching from a 3200+ to a 1600+ CPU for me. You go from instantaneous program starts to a little bit less than instantaneous. It's there for you to actually feel it if you install both OS on the same rig.

XP felt a lot faster and smoother for me, but that's kind of subjective anyways. But in any case, yes, there is something up with your scores. Are you running in sync? Because running PC2100 that high with such low voltage doesn't seem possible, and if you are running syncronously, it would explain the low efficiency. The Twinmos stuff really is just about the best. You'd be able to run at over 220mhz in sync, and expect bandwidth in the 3400's. Whether or not its worth it, is up to you.

When you get your finalized results, be sure to post them in the thread in the memory section.

Will do.

Forgoing the OS discussion, I feel that the efficiency discussion instead is very important to me and probably to many other people who overclocked their CPUs but do not fully have a grasp on how important memory efficiency is to overall performance.


PC2100 Crucial stick removed from the middle and two PC2100 Kingstons left in slots 1 and 3 allowing me to crank up the FSB to 208:

Windows Me: Ram Int Buffered 2458 MB/s / 208 = 11.82 / 16 = 0.74
So now it's 74%

Repeated the test, this time with all the apps that were originally running in the background and got a 73% efficiency.

Windows XP: Ram Int Buffered 2486 MB/s / 208 = 11.82 / 16 = 0.75
This time Windows XP outperformed by 1%.

Note that my system is thoroughly Prime95 tested.


I can post this in memory section if anyone has a problem with this being here:


1. "There is something up with your scores. Are you running in sync? Because running PC2100 that high with such low voltage doesn't seem possible, and if you are running syncronously, it would explain the low efficiency."

Perfect, all the things I always wanted to know about. What does running in Sync mean?
BIOS > Advanced Chipset Features > Memory Frequency:
Is that it?
Mine says Auto.
Other options are by SPD and percentages.
If this is what we're talking about, would much appreciate if you could say in your own words what they mean.


2. The third stick lowers my efficiency 3%-5% but I'd like to keep it because I noticed the difference when I added it.

Windows Me:
3 x 256MB PC2100 @ 204 FSB: 70%
2 x 256MB PC2100 @ 208 FSB: 73%

Windows XP:
3 x 256MB PC2100 @ 204 FSB: 68%
2 x 256MB PC2100 @ 208 FSB: 75%

Now am I to understand that 70% or 75%, regardless, these are really low scores because they should be over 90%?


3. "The Twinmos stuff really is just about the best. You'd be able to run at over 220mhz in sync, and expect bandwidth in the 3400's. Whether or not its worth it, is up to you."

Another issue I always wanted to know about.
My CPU is out of the first batch of 2100+ T-bred B's at newegg when they came out. It runs at 208 x 11= 2288.
I understand that overall performance is not just about 208 x 11 but I do not understand the nature of what it is that I am missing.

What does it mean to have the bandwidth in the 3400 vs. my 2400s?
What does it mean to have 220 FSB vs. my 208 FSB?

Those answers can be helpful in the final question of whether $120 is worth that difference.

Yes... Auto is not what you want. Looks to me that you're running in a 5/4 ratio, meaning that your memory is actually only running at 166mhz, which is what your mobo would run at set to auto. So, you're only running your memory at 166mhz, in which case a score of 2486 would be 2486/2656=93.6%, much more in line with whats to be expected. What mobo do you have? You need to find out how to run your current memory faster, if possible. It'll probably top out at 180mhz or so. See whether your BIOS allows for percent of fsb, or fsb/dram ratio, or a seperate data rate option for the memory altogether.

Weakest points many overclockers have is knowledge about memory so I'm going to have to bother you a bit more with them:

"So, you're only running your memory at 166mhz, in which case a score of 2486 would be 2486/2656=93.6%"

When you say 166, do you mean even though it's set to 208 FSB, it's actually 166 because of the auto setting??

nForce2 Epox 8RDA+

"It'll probably top out at 180mhz or so. See whether your BIOS allows for percent of fsb, or fsb/dram ratio, or a seperate data rate option for the memory altogether."

What does running in Sync mean?
BIOS > Advanced Chipset Features > Memory Frequency:
Mine says Auto.

Other options are by SPD and percentages.

Top out when I set it to SPD or percantages? What should I set it as?

Regarding "seperate data rate option for the memory altogether."
snapshot from pdf manual:
http://www.pbase.com/image/19181832/original

Is there one?


Would much appreciate if you could say in your own words what SPD means and percentages mean.

How are these questions different:
What does it mean to have the bandwidth in the 3400 vs. my 2400s?
compared to
What does it mean to have 220 FSB vs. my 208 FSB?


Those answers can be helpful in the final question of whether $120 is worth it because the alternative is not slightly cheaper memory but free after rebate Kingstons.


P.S. I will delete all my posts and paste them in another thread if this is messing up your Dual channel poll thread. I kept them here becasue I guess you wouldn't mind this thread being bumped up.

I don't really mind at all. The one in memory section is the real thread anyways. And I would like it bumped . What you would need to do is change the memory frequency to either 6/6 if thats available, or to 208mhz. But 166mhz is already pushing PC2100 quite hard, and 208mhz is pretty much out of the question. So the actual question is what does it mean to a have a memory bus of 166mhz vs one of 220mhz; in which case the difference would be tremendous. Your front side bus is at at 208, but your memory is only at 166mhz, tremendously bottlenecking it. In games and such, the near 50% increase that could result, could be very helpful.

What you would need to do is change the memory frequency to either 6/6 if thats available, or to 208mhz

1. What is "by SPD"??

My options: By SPD,50%,60%,66%,75%,80%,85%,100%,120%,125%,133%,150%,166%,200%,Auto

2. Are percentages under 100% underclocking memory??

3. What is the difference between the memory bus and the front side bus?

4. You're saying my FSB is at 208 but my memory bus is at 166,
How come no one ever lists their memory bus in their sigs or talks about it? Do they all just run theirs at defaults??

5. Take any game as an example, how does higher CPU speed affect it vs. higher FSB vs. higher memory bus??

6. How do I test memory bus stability, Prime95?

"by SPD" keeps the memory at stock, or as close to it as possible. The percentages are the percent of your front side bus. For example, 75% of 208mhz would result in a 156mhz memory bus. Yours is currently most likely at 80%. If you set it to 100%, then your memory would run at 208mhz. The front side bus is rate of communication between the northbridge and the processor; the memory bus is between memory bus and northbridge. Both go hand in hand and effectively limit each other. I'm guessing that most people run their memory bus and front side bus syncronously, so they'd be the same. Since your memory is bottlenecking the front side bus, having your front side bus at 166mhz would most likely yield identical performance. Look here to see how front side bus affects 3Dmark2001 scores.

I feel I'm making a lot of progress by reading this.

1. What is the difference between "Auto" and "By SPD"

2. FSB doesn't mean anything if you have low rated memory.
There I was thinking my 208 FSB was something but real world performance was 166, all this time it was like I was at 166 FSB, this is a big deal, in a negative sense.

3. This then is what I'm looking at with 208 FSB:
80% for 166
or
85% for 177

100% is out of the question you're saying.

OK so I should set it to 80% then Prime95 for half a day, is that the idea? Then up it to 85% then Prime95 half a day...

Setting to 100% would be like overclocking your CPU unrealistically high, then trying to boot with it, in other words 100% is not even up as an option unless I have what kind of RAM at least, in your opinion?

4. Would you say 208 FSB was unneccessarily stressing my system and that contrary to my previous understanding, people should go for a higher MHz clock with lower FSB and a higher multiplier if they have lower rated memory?

PC1600 = (100 MHz Operating Speed) x (2x Rising & Falling) x (64-bit Bus) / (8 bits per byte) = 1600 MB/s available bandwidth.

PC2100 = (133 MHz Operating Speed) x (2x Rising & Falling) x (64-bit Bus) / (8 bits per byte) = 2128 MB/s available bandwidth.

PC2400 = (150 MHz Operating Speed) x (2x Rising & Falling) x (64-bit Bus) / (8 bits per byte) = 2400 MB/s available bandwidth.

PC2700 = (166 MHz Operating Speed) x (2x Rising & Falling) x (64-bit Bus) / (8 bits per byte) = 2656 MB/s available bandwidth.

PC3000 = (183 MHz Operating Speed) x (2x Rising & Falling) x (64-bit Bus) / (8 bits per byte) = 2928 MB/s available bandwidth.

PC3200 = (200 MHz Operating Speed) x (2x Rising & Falling) x (64-bit Bus) / (8 bits per byte) = 3200 MB/s available bandwidth.

That's from Memory Sticky, weren't there higher speeds released recently and how is this info completed with their speeds?


I'll take this to memory section and start a new thread after your next answer.. thank you very much for all this usefull info.

"By SPD"and "Auto" should each do exactly the same thing.

2. FSB doesn't mean anything if you have low rated memory.
There I was thinking my 208 FSB was something but real world performance was 166, all this time it was like I was at 166 FSB, this is a big deal, in a negative sense.

Click to expand...

Unfortunately, this is true. Dual channel should theoretically double your bandwidth, but in reality it doesn't seem to do much. It does give you a boost, but nothing that dramatic. The memory still does very much limit your system performance, and the front side bus cannot make up for it.

100% is out of the question you're saying.

OK so I should set it to 80% then Prime95 for half a day, is that the idea? Then up it to 85% then Prime95 half a day...

Setting to 100% would be like overclocking your CPU unrealistically high, then trying to boot with it, in other words 100% is not even up as an option unless I have what kind of RAM at least, in your opinion?

Click to expand...

100% would need memory that could run at 208; mid-level quality PC3200. If you really are at 80% right now; this was just a conjecture on my part, it would be a great overclock for PC2100, from 133mhz to 166mhz. For it to go further is not to be expected, but who knows. You could try increasing the memory voltage to around 3v and see how high you can make it go.

4. Would you say 208 FSB was unneccessarily stressing my system and that contrary to my previous understanding, people should go for a higher MHz clock with lower FSB and a higher multiplier if they have lower rated memory?

Click to expand...

Your understanding is correct, but this really only applies to when the front side bus and memory bus are in sync. Having your memory run at 166mhz is really only as effective as having your front side bus running at 166mhz.

Well I am back to square 1 of overclocking.

Step 1 is therefore to find out max memory bus then set FSB to equal max memory bus then crank up the multi to the highest stable resulting MHz. Right or wrong?