Did you have to increase the volts to memory at all?
btw im gonna try like you just did...gonna start out a bit conservative and see how the memory goes...
1st try 10x220=2200 10,4,3,2.5 sandra 3257/3019.....which is better than what I was getting...before 3024/2842 with 11x200 6,3,2,2 so thats pretty cool with me so far =P
2nd try 10x225=2255 10,4,3,2.5 sandra 3328/3086..better still !
3rd try 10x230=2305 10,4,3,2.5 sandra 3406/3158...BUT ut2k3 locked up on me =/....so I backed the fsb down to 228
4th try 10x228=2286 10,4,3,2.5 sandra 3380/3135 efficiency was 92% .. played a scim on compressed....btw we won..=p.....ran solid...I think im gonna leave it here till I hear from some of you guys and see if this is it. I can tell a big difference in the way things run goin this route with higher fsb and lower clock mult. Its like this rig just opend up a whole new level...I cant thank you guys enough for the help and suggestions. This is exactaly the way I was wanting to run ut2k3......wait hold for it........WOOT!!!! *grin*:D
**played several games of ut2k3 and ran the 3dmark2003 and temps are running cool my asus probe sayin its 95-105 F
Download memtest which run in DOS only. But the good thing is that they have an ISO to burn on a cd, just boot with that and you are on. It starts automatically but you can also change settings in there.
I always use max memory voltage, no matter what fsb or timings a use.
Straight_ManGeeky, in my own wayNaples, FLIcrontian
edited November 2003
Actually, you CAN run the bases async, but what happens is as here:
You use VERY conservative RAM timings if the FSB is boosted more, beacuse those timings are relative to the FSB and CPU and the RAM will overheat faster after a certain point than the CPU will. So, if you have RAM that was OC'd originally, you need to INCREASE timing values to compensate for a pumped FSB as the OC already pushes the RAM beyond its module native rates.
Note, for example, that dual channel RAM is NOT intended to be used in a SIMULTANEOUS manner, really, it is alternately polled by a small offset. Thus, JEDEC timings for dual channel are 8,4,4,2 if you use RAM that is native module synced for the purpose and the FSB you are using. The more you pump the CPU multiplier, the more conservative your RAM settings need to be, and the more you pump the FSB the more conservative the RAM settings (in most cases, depends on ratio) need to be.
FSB sets base clock tick rate versus CLOCK chip (also CMOS on most modern machines, combined IC). The CPU is counting ticks faster than RAM when you pump teh FSB and not the RAM, but OC'd RAM cannot be base rate pumped as high. Mild OCing is common, DRAM chips are underspeced often and the RAM setting numbers recommended rise accordingly from a non-OC'd RAM to an OC'd RAM module.
So, yes, balancing FSB to RAM base is most stable way, but there is another way: lets say you get, instead of PC3200, PC3500 (this is RAM that is stable at a faster rate than the PC3200, and is less likely to be OC'd from say a PC2700 to PC3200 out of box), OR you get RAM known to be very conservatively made and intended for overclocking. The first is easier to find than the second, and generally a TIB less expensive, as the SECOND needs a lot of expensive, engineer managed QC or you get uneven quality and some users having very bad luck and others having very good luck out of the same RAM stick model number.
What is happening in most cases, is NOT that the CPU itself cannot be pumped, but that either the RAM or Video subset or bridges cannot keep up without burning up slowly adn going very unstable. This assumes very aggressive CPU cooling. My Barton 2500+, even at stock rates, likes a lower voltage than MSI specs as default, I DROP the voltage .025 to .050 volts versus stock to make CPU stable with a lesser fan and heatsink. 2 week runtimes on 98 SE between reboots with 60% resources free??? Yep.
Try a 1.575 voltage on the CPU with a 12x190 multiplier, and let the box cool down before you restart also. Set RAM as discussed earlier for now. Feed the RAM base rate a 166 or a 200, if you feed 200 and the RAM is stable with heatsinking, than try dropping the 11, 3, 3, 2 to 8, 3, 3, 2 0r even 8, 2, 2, 2.
DDR ram never gets to hot, not even at 3.2 volts. I have tried it and many others have as well. Heatspreaders on ram is actually counterproductive. They make the ram warmer than without.
On an AMD rig, NEVER run cpu and memory assyncronous. That defeats the whole design and advantage of Amd processors.
0
Straight_ManGeeky, in my own wayNaples, FLIcrontian
edited November 2003
Right, in a sense. You can only run async if you overcompensate with extremely conservative RAM timings as what those timings are is CPU\FSB VS RAM native rates. CPU can be OC'd more than RAM, relatively. Good RAM is native at 3.2-3.3 volts, slightly OC'd out of box sticks need a bit less to stay cool. BUT, the reason for spreaders is not that the RAM actually likes or needs them itself, but that board designers cannot mount the RAM at a 45-55% right down angle(ANTIclockwise rotation of X plane of mount, CLOCKWISE of Y plane similarly) to match airflow native to case to make the LONG axis the axis of flow of air between the sticks. So, DIMM one side one overheats more than DIMM 3 side 2 and one way out if your board has three DIMM slots is to run DIMM 1 in DIMM slot 1 and DIMM 2 in DIMM Slot 3. Essentially, if the spreader is epoxied on and joined with metallic like liquid solder, the heat from the hotter side migrates to cooler side with conduction. Right GENERAL idea, wrong exact fastening method as tape has too little metallic or ceramic conductive content and acts as a conduction barrier. In essence, the material used on the underside of the space shuttle makes agood heat dissipative, though the extreme heat the underside of the space shuttle encounters measn thay also have to be partly heat ablative.
Here is what top-end Corsair looks like with a review from Anandtech, and the outer coating is not normal enamel, it is a high-carbon coating.
Cheap modules overheat a LOT, as the cheap ones tend to be oversped to begin with. Samsung makes decent RAM, Corsair, and Kingston HyperX is decent.
Do not tape, use Arctic Silver Alumina Epoxy to mount the spreaders. BEST heatsinks for RAM are actually baked on epoxied with a high metallic content on the epoxy (let me find a link, think it is a new Corsair line) or a metallic+carbon EPOXY heatsink for the module. The TAPE is what makes the heatsinks counterproductive, not the fact of the heatsink itself, and a SOLID joined spreader will work better than a taped one by a whole bunch.
I have tried upping the voltage, still no go. I'm going to try to lower the mulit to 10, it is at 10.5 right now, and then raise the voltage to see if I can push out 2.ghz its a 1.95 right now.
Have tried running async?
For testing purposes only.
If you run CPU and mem async you can pinpoint the problem.
Would be nice to know whether it is the CPU or the memory.
Firstly, sorry if I unintentionally bumped an otherwise dead thread. Ageek, if you're still part of this forum, how did you attain your knowledge?
I'm extremely new to overclocking principles, as I've spent the last three years separated from computers and ipod-mania. I'm quite dedicated, however, when it comes to filtering information. Seeing as how I'll probably run into memory-related issues in the future, I'd like to know what to expect of my hardware.
Did you browse the web for the information that you presented the the viewers of this thread? Or, are you certified in memory architecture?
I'd appreciate it if you posted several links, references to books, articles in journals if you could.
I have tried several assynch setting just for the fun of it. Wouldn't boot over 270 fsb on ANY dividers which just kills the whole idea. NForce just isn't designed to run at assynch, simple as that.
I'll say it again, heatspreders on ram is just bs and it will always be bs on DDR at least.
I'll say it again, heatspreders on ram is just bs and it will always be bs on DDR at least.
But the heatspreaders look good.
I have found a couple of links regarding async performance on the nForce2.
http://www.extremetech.com/article2/0,3973,808388,00.asp
Quote:"DDR400 and DDR333 Memory Support: Nvidia noted that running the
clocks in "sync" with the frontside bus speed
of the Athlon XP delivers the best performance; the system BIOS of all the
tested motherboards supported this sync mode, though none by default."
http://www.vr-zone.com/reviews/EPoX/8RGA+/page5.htm
Quote:"More tests are done at different CPU and Memory FSB settings and we
can conclude that running Sync at 133/133Mhz and 166/166Mhz deliver better
memory performance than Async 133/166Mhz and 166/200Mhz. For best
performance, we ran 8RGA+ at Dual Channel DDR400 for the remaining tests."
Greetings, all. Over a year ago, I got ahold of a DFI LAN PARTY NFII ULTRA board (first rev) and a Barton-core XP 2500+. As far as I can tell, the multiplier appears unlocked, but it does not seem to like certain settings - like multipliers below 9.5 at high FSB (214+.. not sure how low, exactly). However, that doesn't really disturb me much.
I'm running 2x256MB of Crucial PC2100 CL2.5 in Dual-Channel mode (rated for 2.5V, not certain about timings), at VDIMM 2.7V, 7-3-3-3 @ 178 (synchronous with FSB - CPU running at 11x178). I was running the RAM async at 142 (5-2-2-2.5) earlier, due to skepticism that 178 FSB could take advantage of dual channel - but that became a non-issue when I found a way to push the RAM higher (and the CAS is lower in ns, if not in clock cycles). The RAM was originally intended to go into a nForce 420 board with my old 1.1 Thunderbird, but I never actually GOT a nForce 420 board, moving directly from KT133 to nF2 Ultra.
I was wondering if anyone had any advice for pushing the RAM higher, or if that's a good idea? I see some mentions for pushing the VDIMM higher on this forum and I'm meaning to try that (to 2.8V, the board's limit), but I just wanted to hear if anyone had any experience with similar RAM, or, basically, the opinions of the more enlightened.
Any more info you need, I'll try to dig up... any info you can give would be much appreciated..
Comments
btw im gonna try like you just did...gonna start out a bit conservative and see how the memory goes...
1st try 10x220=2200 10,4,3,2.5 sandra 3257/3019.....which is better than what I was getting...before 3024/2842 with 11x200 6,3,2,2 so thats pretty cool with me so far =P
2nd try 10x225=2255 10,4,3,2.5 sandra 3328/3086..better still !
3rd try 10x230=2305 10,4,3,2.5 sandra 3406/3158...BUT ut2k3 locked up on me =/....so I backed the fsb down to 228
4th try 10x228=2286 10,4,3,2.5 sandra 3380/3135
**played several games of ut2k3 and ran the 3dmark2003 and temps are running cool my asus probe sayin its 95-105 F
What memory test should I be using???
I always use max memory voltage, no matter what fsb or timings a use.
This is fun isn't it?
right now my fsb is 185, I can't get it to do 200 stable without that horrible siren going off. It's a 1700+ JIUHB
Is your NF7-S a rev 2.0?
and have you tried upping the voltage on the memory and processor?
You use VERY conservative RAM timings if the FSB is boosted more, beacuse those timings are relative to the FSB and CPU and the RAM will overheat faster after a certain point than the CPU will. So, if you have RAM that was OC'd originally, you need to INCREASE timing values to compensate for a pumped FSB as the OC already pushes the RAM beyond its module native rates.
Note, for example, that dual channel RAM is NOT intended to be used in a SIMULTANEOUS manner, really, it is alternately polled by a small offset. Thus, JEDEC timings for dual channel are 8,4,4,2 if you use RAM that is native module synced for the purpose and the FSB you are using. The more you pump the CPU multiplier, the more conservative your RAM settings need to be, and the more you pump the FSB the more conservative the RAM settings (in most cases, depends on ratio) need to be.
FSB sets base clock tick rate versus CLOCK chip (also CMOS on most modern machines, combined IC). The CPU is counting ticks faster than RAM when you pump teh FSB and not the RAM, but OC'd RAM cannot be base rate pumped as high. Mild OCing is common, DRAM chips are underspeced often and the RAM setting numbers recommended rise accordingly from a non-OC'd RAM to an OC'd RAM module.
So, yes, balancing FSB to RAM base is most stable way, but there is another way: lets say you get, instead of PC3200, PC3500 (this is RAM that is stable at a faster rate than the PC3200, and is less likely to be OC'd from say a PC2700 to PC3200 out of box), OR you get RAM known to be very conservatively made and intended for overclocking. The first is easier to find than the second, and generally a TIB less expensive, as the SECOND needs a lot of expensive, engineer managed QC or you get uneven quality and some users having very bad luck and others having very good luck out of the same RAM stick model number.
What is happening in most cases, is NOT that the CPU itself cannot be pumped, but that either the RAM or Video subset or bridges cannot keep up without burning up slowly adn going very unstable. This assumes very aggressive CPU cooling. My Barton 2500+, even at stock rates, likes a lower voltage than MSI specs as default, I DROP the voltage .025 to .050 volts versus stock to make CPU stable with a lesser fan and heatsink. 2 week runtimes on 98 SE between reboots with 60% resources free??? Yep.
Try a 1.575 voltage on the CPU with a 12x190 multiplier, and let the box cool down before you restart also. Set RAM as discussed earlier for now. Feed the RAM base rate a 166 or a 200, if you feed 200 and the RAM is stable with heatsinking, than try dropping the 11, 3, 3, 2 to 8, 3, 3, 2 0r even 8, 2, 2, 2.
John.
John.
On an AMD rig, NEVER run cpu and memory assyncronous. That defeats the whole design and advantage of Amd processors.
Here is what top-end Corsair looks like with a review from Anandtech, and the outer coating is not normal enamel, it is a high-carbon coating.
http://www.anandtech.com/memory/showdoc.html?i=1867
If you go to the link below and look on the right side, there are many links to other review places for the XMS Pro series.
http://www.corsairmicro.com/
Cheap modules overheat a LOT, as the cheap ones tend to be oversped to begin with. Samsung makes decent RAM, Corsair, and Kingston HyperX is decent.
Do not tape, use Arctic Silver Alumina Epoxy to mount the spreaders. BEST heatsinks for RAM are actually baked on epoxied with a high metallic content on the epoxy (let me find a link, think it is a new Corsair line) or a metallic+carbon EPOXY heatsink for the module. The TAPE is what makes the heatsinks counterproductive, not the fact of the heatsink itself, and a SOLID joined spreader will work better than a taped one by a whole bunch.
For testing purposes only.
If you run CPU and mem async you can pinpoint the problem.
Would be nice to know whether it is the CPU or the memory.
But it sure looks like a very bad JIUHB.
BTW Is it a DLT3C, DUT3C or something else?
I'm extremely new to overclocking principles, as I've spent the last three years separated from computers and ipod-mania. I'm quite dedicated, however, when it comes to filtering information. Seeing as how I'll probably run into memory-related issues in the future, I'd like to know what to expect of my hardware.
Did you browse the web for the information that you presented the the viewers of this thread? Or, are you certified in memory architecture?
I'd appreciate it if you posted several links, references to books, articles in journals if you could.
Thanks
I'll say it again, heatspreders on ram is just bs and it will always be bs on DDR at least.
But the heatspreaders look good.
I have found a couple of links regarding async performance on the nForce2.
http://www.extremetech.com/article2/0,3973,808388,00.asp
Quote:"DDR400 and DDR333 Memory Support: Nvidia noted that running the
clocks in "sync" with the frontside bus speed
of the Athlon XP delivers the best performance; the system BIOS of all the
tested motherboards supported this sync mode, though none by default."
http://www.vr-zone.com/reviews/EPoX/8RGA+/page5.htm
Quote:"More tests are done at different CPU and Memory FSB settings and we
can conclude that running Sync at 133/133Mhz and 166/166Mhz deliver better
memory performance than Async 133/166Mhz and 166/200Mhz. For best
performance, we ran 8RGA+ at Dual Channel DDR400 for the remaining tests."
I'm running 2x256MB of Crucial PC2100 CL2.5 in Dual-Channel mode (rated for 2.5V, not certain about timings), at VDIMM 2.7V, 7-3-3-3 @ 178 (synchronous with FSB - CPU running at 11x178). I was running the RAM async at 142 (5-2-2-2.5) earlier, due to skepticism that 178 FSB could take advantage of dual channel - but that became a non-issue when I found a way to push the RAM higher (and the CAS is lower in ns, if not in clock cycles). The RAM was originally intended to go into a nForce 420 board with my old 1.1 Thunderbird, but I never actually GOT a nForce 420 board, moving directly from KT133 to nF2 Ultra.
I was wondering if anyone had any advice for pushing the RAM higher, or if that's a good idea? I see some mentions for pushing the VDIMM higher on this forum and I'm meaning to try that (to 2.8V, the board's limit), but I just wanted to hear if anyone had any experience with similar RAM, or, basically, the opinions of the more enlightened.
Any more info you need, I'll try to dig up... any info you can give would be much appreciated..