Team Xtreem 2GB (2 x 1GB) DDR2 SDRAM DDR2 1300 (PC2 10400)Dual Channel Kit + eVga680i

[Don_Dan]

I've never been a big memory overclocker but I've really become fascinated by it lately, I'm trying to learn what I can. There are so many combination's that can help or hurt that it's a bit overwhelming. One thing is for sure though, this RAM has exposed all of my motherboards weaknesses.

Actually it was C4 at 1110Mhz that was able to break 10K in Everest. While this RAM seems to get along alright with my 680i it doesn't like certain dividers. It refuses to boot into Windows period at 1:1 although the 680i does seem to have a couple of FSB holes below 1333 . Maybe next week I'll pick up a EP45-UD3P.

I see you're lowering tRTC, tRP and tRC with Memset, that's one of the things I would have given you as an advice. Are you booting at 400 MHz FSB? If yes, then try to boot at lower FSB speeds and clockgen to 400MHz. That's because the strap change at 400MHz will not take effect if you pass this mark in Windows and it will keep on running on the faster 333MHz strap.
On 680i it's all about picking the correct divider, the best are 2:3 and 9:11 and maybe 9:10.

Look at my screen, that was the best result I've ever made on 680i, because my CPU doesn't like more than 400MHz FSB...



Only 600MHz on the RAM, but almost 11k in Everest's read test, all my changes in Memset and the used voltages are visible, maybe my tips will help you a bit.

[hecktic]

Ill have to revist this with the e8400 chip sometime.

That cpu was a custom purchase confirmed to oc up to 6.2ghz

So I should have a lot of head room to play around with.

I usually just overclock without messing with the straps until I have no other choice so we will see how things go.

[Mr Roboto]

I see you're lowering tRTC, tRP and tRC with Memset, that's one of the things I would have given you as an advice. Are you booting at 400 MHz FSB? If yes, then try to boot at lower FSB speeds and clockgen to 400MHz. That's because the strap change at 400MHz will not take effect if you pass this mark in Windows and it will keep on running on the faster 333MHz strap.
On 680i it's all about picking the correct divider, the best are 2:3 and 9:11 and maybe 9:10.

Look at my screen, that was the best result I've ever made on 680i, because my CPU doesn't like more than 400MHz FSB...

Only 600MHz on the RAM, but almost 11k in Everest's read test, all my changes in Memset and the used voltages are visible, maybe my tips will help you a bit.

I can boot into Windows without issue at up to 500Mhz FSB. Anything beyond that takes a bit of work. I almost never use 8x multiplier for that reason. I checked ClockGen out a while back but it didn't show support support for 680i so I deleted it. I saw someone mention a beta version in a previous post in this thread but Google turned up nothing. Do you have the beta version of ClockGen you could upload somewhere?

I noticed with all these memory timings that most are better left alone unless you have a great understanding of them. There are a lot of variables. The advice I've seen posted on the web ranges from the ridiculous to hearsay and guesstimates. I think what it comes down to is researching, benching, trial and error.

I do have a few questions though. When using MemSet are there some values that if you run too low you actually risk physically damaging the RAM or is it just like CPU\GPU overclocking where you'll just BSOD and have to reboot? Mostly I'm talking about tRFC, I noticed that lowering it gets some of the best bandwidth increases. What about running at 1T command rate? I know you have to downclock considerably. Is there any value in adjusting RMS? I was always interested in it because the values are so different from the rest being measured in µs instead of cycles. Thanks.

[CryptiK]

How old are your TG's?

I found with mine, the heatspreaders are cool for awhile until the IC's bed into the thermal tape. I attached foldback clips to the heatspreaders to apply inward pressure while they ran for awhile.

Now even at 2.1v they get quite warm, but nowhere near as warm as my cellshocks which use thermal paste, so there is some thermal transfer issues.

[Mr Roboto]

How old are your TG's?

I found with mine, the heatspreaders are cool for awhile until the IC's bed into the thermal tape. I attached foldback clips to the heatspreaders to apply inward pressure while they ran for awhile.

Now even at 2.1v they get quite warm, but nowhere near as warm as my cellshocks which use thermal paste, so there is some thermal transfer issues.

I just got them Friday. Yeah I was thinking that because the material they covered the heatspreaders with, while it looks nice it doesn't seem like it's the best choice for dissipating heat. At first I thought they were covered with some sort of plastic or carbon fiber, to me it seems like it's trapping the heat instead.

Now that you mention it I'm going to pop them off right now and see how warm they get underneath with some voltage. Were your Cellshock chips D9GKX too?

[CryptiK]

It's not so much the sinks themselves, being as thick as they are is actually a good thing as they can absorb more heat out of the IC's, it's the thick thermal tape they use that I think is the issue.

My Cellshock sinks are very thick also and they get very hot without good cooling.

My Cellshocks are GKX, my TG's are GMH.

[Mr Roboto]

It's not so much the sinks themselves, being as thick as they are is actually a good thing as they can absorb more heat out of the IC's, it's the thick thermal tape they use that I think is the issue.

My Cellshock sinks are very thick also and they get very hot without good cooling.

My Cellshocks are GKX, my TG's are GMH.

I see what you're saying now. I took one of the sinks off the other day when I first got them just to have look see. They came off easily with no force. After a day of breaking them in I just tried to take them off. They're not coming off now. The one side came off with a little effort but the larger half wouldn't. I stuck them in the oven for a few minutes and the ends started to loosen but still wouldn't budge. Instead of doing something I would regret I left them in the oven for a few more minutes to let the pads get nice and soft and put them back together with two clips, one on each side like you said. I'll leave the clips on for a day or so or at least a few on\off cycles. Once that thermal strip sets it's very hard to get off.

I didn't see that you're using the PC2-8500's. The 9600's were supposed to come with those Thermalright HR-07 heatsinks but they didn't. I contacted MemoryC.com but they have yet to email me. I probably wouldn't have used them but I was actually looking forward to having them bare. I think with a fan blowing on them they'd be cooler.

That strip actually pulled the printing off of the top two IC's in the picture.

[Don_Dan]

I can boot into Windows without issue at up to 500Mhz FSB. Anything beyond that takes a bit of work. I almost never use 8x multiplier for that reason. I checked ClockGen out a while back but it didn't show support support for 680i so I deleted it. I saw someone mention a beta version in a previous post in this thread but Google turned up nothing. Do you have the beta version of ClockGen you could upload somewhere?

I noticed with all these memory timings that most are better left alone unless you have a great understanding of them. There are a lot of variables. The advice I've seen posted on the web ranges from the ridiculous to hearsay and guesstimates. I think what it comes down to is researching, benching, trial and error.

I do have a few questions though. When using MemSet are there some values that if you run too low you actually risk physically damaging the RAM or is it just like CPU\GPU overclocking where you'll just BSOD and have to reboot? Mostly I'm talking about tRFC, I noticed that lowering it gets some of the best bandwidth increases. What about running at 1T command rate? I know you have to downclock considerably. Is there any value in adjusting RMS? I was always interested in it because the values are so different from the rest being measured in µs instead of cycles. Thanks.

Sorry, I don't have such a version, I always used Asus' own software, AI Suite. Maybe you can try SetFSB, but I don't know if your motherboard is supported. However, it's not about booting very high FSBs, but rather booting with a low FSB speed and then raising it in Windows. The chipset will set the strap according to the FSB set in BIOS, and once above 400MHz internal chipset latencies are very loose, and thus your performance will go down. If you can, test 395MHz and 405 MHz FSB speed and leave the rest of the configuration unchanged to see it for yourself.

When I tested 680i, the sub timings that showed bandwidth gains were tRFC and tRC, you just have to test how low you can run them on your memory. Of course it isn't guaranteed, but I never killed memory because of tight latencies, and I've never read about such a case. Most likely you will get a bluescreen instantly if the timings are too tight. If you want to run CR 1T you will need to run excessive volts on the chipset, I reached about 425MHz CL3 1T relatively stable but in my opinion performance is not optimal. For 680i it's better to go for highest possible CL4 frequency.

I can recommend the lecture of this thread by eva2000 if you haven't done so already. He is really the God of 680i tweaking.

[CryptiK]

You won't physically damage your ram by setting timings tight etc. Voltage, current and heat will however damage them.

Oh, and eva2000 is a ram tweaking expert, on any platform.

[Don_Dan]

Oh, and eva2000 is a ram tweaking expert, on any platform.

I know!

[Mr Roboto]

It's not so much the sinks themselves, being as thick as they are is actually a good thing as they can absorb more heat out of the IC's, it's the thick thermal tape they use that I think is the issue.

My Cellshock sinks are very thick also and they get very hot without good cooling.

My Cellshocks are GKX, my TG's are GMH.

After heating up the thermal tape and allowing them to set they work much better now. Even at a moderate 2.2v they get really warm now and give off a lot of heat. Seems they just needed to gel a bit.

I also tried some 4x1GB action with these (since I bought two kits) and was somewhat successful. I can get them up to 1155Mhz at 5-5-5-15 but they need damn near 2.45v which I don't feel comfortable running daily. Maybe it's just this motherboard but I'm surprised at how much voltage these sticks like. At 1280 5-5-5-15 and 2.5v I can feel this RAM wanting more juice.

[d3fct]

hi guys i gotta question for ya i have 4x1gb ddr2 1300 team xtreems on a ud3r with a 9650, i can get to 4.5 stable with no issues(500mhz) im running 1:1 btw. Now if i try to get past 500, anywere from 500-600 i have to mess with performance levels and refresh cycle time just to get a post but locks before windows loads. it wont even make it into memtest at that speed and tough to get a post, am i missing a subtiming or two thats making this so tough? the best i could get it so far was a post and into windows (then freeze) at 525mhz was refresh at 59 and performance level 9, that will get me to the windows login screen but freezes. any info would be greatly appreciated im trying to hit 5g's or better. i have loosened the timings way up, didnt help, put volts to max, then min didnt help. i have added vcore and nb volts none seemed to make any differance.

sorry for asking in this post, but it seems that this group of guys are pretty familiar with this kit.

system:
ud3r gigabyte
q9650 @ 500mhz 4.5g, 1.4v in bios, 1.37v in cpuz, 1.36v at full load(cpuz)
4x1gb team xtreem ddr2 1300 @ 2.45v 5-4-4-13
single stage phase -52 @ idle, -48 @ full load

[Don_Dan]

I don't think that the RAM is to be blamed in this case. This kit is one of the best OCing kits ever made.

So running a quad and 4GB of high-clocked memory can be a problem for the Northbridge, but since increasing the Vmch didn't get you any further, my bet is that you found the FSB-wall of your chip! 525MHz is a very good result for a quad already, maybe upping the tRFC to 60 will get you a bit further, but I don't think that will be more than maybe one MHz or two.

If you want to make absolutely sure that it's your processor limiting you, run only one stick in single-channel mode, CL5 and 1:1 and increase the FSB.


// edit: You'll find more answers here.

[d3fct]

well i ran 1 stick @ 530 4-4-4-12 2.45v and did some screwing around with the subtimings and was able to boot into windows and run a superpi 1m bench at 9.8sec. didnt prime it or anything still tryin to get further.

[d3fct]

i think your right don, the highest i can get along with stabilty is 515, i can get into windows at 520/525/530 but is no wheres near stable, no biggie i guess ill have to get a big dual if i want to see 5 gigs. just got this cpu and mobo last week just wanted to see what its limits were, guess i found the cpu's limit.

[Don_Dan]

i think your right don, the highest i can get along with stabilty is 515, i can get into windows at 520/525/530 but is no wheres near stable, no biggie i guess ill have to get a big dual if i want to see 5 gigs. just got this cpu and mobo last week just wanted to see what its limits were, guess i found the cpu's limit.

Yeah, I guess you found the limit of your processor, the memory shouldn't be limiting you even at 530MHz CL4.
Now get an E8600 and be ready for some high FSB action!

[hecktic]

looks like my 780i board killed one stick of the TG set, even though it was underclocked to 1066mhz cl4-4-3-4 or something like that with about 1.875v or so.

now I gotta figure out how to get these replaced before I can try them again and revisit this project.

Anyone know the RMA process for TG ram?

[cdawall]

these results aren't all that amazing

PC4300 CL4 is stock







i also have some team D9GMH 667 that can clock to 1300 cas5 along with a dominator C3D kit

[Don_Dan]

these results aren't all that amazing

PC4300 CL4 is stock

i also have some team D9GMH 667 that can clock to 1300 cas5 along with a dominator C3D kit

Stock speed doesn't matter at all as long as it's D9

BTW, you can post your results in the D9 chip thread!