3.06HT Northwood OC Success [Archive]

consumer9000

10-27-04, 02:36 AM

I decided to upgrade from my non HT 2.53GHz Northwood to a 3.06 HT Northwood. While this may seems strange to some, I did not want to go to the trouble or expense of building a whole new system to satisfy the requirements of getting either a Prescott or Athlon FX CPU. I'm running an Asus P4T533 board which uses 32bit rambus. While a fast board, it is also the most finicky I have ever dealt with. Nonetheless Rambus is very fast even now, and I've been pleased with Asus quality time and time again. To fortify the the cooling I installed a Thermalright NB1-C in place of the stock Asus active cooler. I fitted the RIMMs with Thermaltake's active memory cooling kit. A slick Vantec TMD HSF takes care of the CPU. The most important modification was the installation of a custom cooling shroud which dropped temps across the board!
Using Prime95 and 3DMark 2003 to test stability, I found cooling, RAM and Voltage to be major limiting factors. Addressing the cooling as well as possible, I bumped the VCORE to 1.65 and have been able to achieve a stable top speed of 3633MHz. While the CPU and FSB speeds jumped nicely, I had to kick down the RAM speed to keep Prime95 happy. I wonder about that program, because it will cry foul far before I notice any crashes or symptoms. Dropping the RIMMs down to 950MHz and enabling the Turbo timing mode didn't have as significant of a performance impact as I had feared... When bandwidth is needed I run my system at 3450MHz with the FSB at 600MHz and the RAM at 1200MHz. Beyond this the memory cannot keep up.
The system will boot to windows all the way up to 161*23 (3700MHz) but will be quite unstable. Overall I'm pleased with the overclockability of this CPU and this often unpredictable board and memory format.
To those of you wondering whether or not to upgrade to a Prescott, I suggest you do what I have done until the next generation of CPU's is released. HT really makes a difference, and the Northwood's 20 stage pipeline kicks ass. In Sandra my CPU at 3633MHz is on par with the Prescott chip running at 3.8GHz!
My next project is to voltmod the Radeon to get some more power out of it...at stock voltages it hardly overclocks at all. :mad:
http://www.ocforums.com/attachment.php?attachmentid=34982&stc=1

consumer9000

10-27-04, 07:17 PM

Just out of curiosity, but does anyone else have this CPU or motherboard still? If so I'd really like to know what kinda overclocks you have been able to achieve, and if you have any advice for me, as I'm sure there is always more that can be done for MORE POWER!!! :)

FIZZ3

10-29-04, 08:11 AM

Yeah I do. :)

We've got a very similar overclock too hehe. My CPU replaced a 2.0 NW some time ago. Given the slow progress in CPU speeds I'm still fairly happy with the performance I have now.

How's the beta BIOS compared to the official release? I have stuck to the latter myself.

consumer9000

10-29-04, 10:53 AM

Cool, at first I was somewhat wary of the beta release, though I've never met a Beta Asus BIOS which caused me any real trouble. I really haven't noticed any real changes in the beta bios. After contacting Asus, they were of little help in identifying what had changed. Same Promise bios, same settings options and post screen. Likely just a few bug fixes and tweaks. At the very least it isn't causing problems, and seems to have contributed to better stability when pushing the board. I also notice a "TrendMicro ChipAway" Antivirus announcement as the system posts. I'm not sure if it's my new Intel 1000MT NIC or the new bios which carries this program. Give it a try, if you've been happy with 1006, this should not dissappoint.
One thing that drives me crazy is my system will be completely stable at even more aggressive settings for everything but Prime95. That damn app will crash in 10 minutes, wheras I can play games etc for hours and hours without event?! I still use Prime95 as my standard for reliablility, as I'm sure there are situations where rounding errors in real life will lead to fun OS/Software problems down the road...

FIZZ3

10-29-04, 12:17 PM

The Trend message is from the BIOS. It's a boot sector protection that can be enabled or disabled. I've noticed that some versions of the BIOS tend to default it to enabled status, maybe that's the case for the beta bios as well.
I've used a raid module modification to change my stripe size and this required the antivirus bit to be disabled as it used a portion of its allocation. I'm currently running the normal BIOS, as the stripe size remains valid.

I know the feeling with Prime95. It's a reliable index of (CPU) stability though, as long as you're not using a dated version. I demand absolute stability from my system and therefore fare by prime95 (and memtest & 3dmark). I've seen too many problems with systems to risk the headache.

Clevor

11-01-04, 12:57 AM

Consumer, the P4T533-R has known problems with SuperPI and Prime. You can be stable in all 3D/CPU tests but it will kick out of SuperPI within 15 seconds. I wouldn't be concerned with it unless all you do is watch Prime run overnight.

I still have two boards. From what I've seen, they max out around 175-180/3x. How high you go at 4x RDRAM depends on the sticks, which are a crapshoot. I've seen sticks do 136/4x to 170/4x (and not done yet).

All the RDRAM boards are very smooth with 3D/video editing, and boot/exit WinXP inexplicably fast.

As far as the 3.06Bs, it's hard to find a bad one. I think the worse I seen topped out at 154 or just under 3.6 gig. A good one will do 161+ or over 3.7 gig on air. They usually don't need more than 1.58 volts. I have one unusual 30-cap D1 that does 158 fsb @ 1.49 actual VCORE.

I'm doing some testing now with a P4T-E Intel850 board, with just 2x256 MB PC1066 running max 156/4x the board will do. With a 2.0A, that's 3.12 gig. I'm testing the feasibility of maintaining this as my gaming rig. With a 9800 Pro overclocked to 430/400, it's handling everything at 1600x1200x32, 4xFSAA, 16xAniso, max detail, as far as relatively recent games are concerned. The previous toughies were Aquanox, Ghost Recon, Morrowind, and Max Payne. No problem, smooth and very playable. Same for lesser tier games like IL-2 and SS2nd Encounter.

Not enough memory for Far Cry and Doom3 though. Plus the 9800 Pro is gagging. And Hitman Contracts (OpenGL) is no go even at lower resolutions, primarily due to the ATI card. Even an X800XT PE has trouble with this game. The P4T-E will handle everything before these games with aplomb, though. I see no deficit with 512 MB ram, due to the ultra smoooth 3D characteristics of RDRAM.

FIZZ3

11-01-04, 04:17 AM

I've not tried SuperPI, but as far as Prime95 is concerned my own tests (long ago) have run for 9+ hours without failure. Perhaps grilling 48 hour sessions may end in a rounding error though, I'm not sure. In any case, when you're saying that the board is known to have problems with this test I think it's important to qualify that the problems concern a minute error after days of Prime95, not a 30 minute freeze.

I agree with Clevor's remarks about the performance. The biggest drawback is the limited amount of RAM it supports. It supports a lot on paper, but people have tried 2x512Mb in the past and ran into weird problems. At least back then it was said that 512Mb total ram worked best. Anyway, getting a new set of 2x512Mb today might be a bit of a quest. ;)

Clevor

11-04-04, 12:32 AM

I had to drag this post up again. I've been playing with my AMD rig the last couple of nights: NF7-S with an XP1800 at 2455 mhz, running 223x11, 2-2-2-5 with 2x512 BH-5, CPC enabled. This is a well above average AMD rig, probably ranking 9 on a scale of 10. What a bitch to run compared to my P4T-E at 156/4x with a 2.0A at 3.12 gig.

For one thing, I have to run 2.0 VCORE with watercooling. The CPU runs maybe 45-50 C. I barely have the rig 3D stable. Most are not fortunate to run this high fsb with BH-5 on the board. The point I'm making is I am getting about the same benchmarks as the P4T-E in the 3DMarks, SuperPI, etc.

The P4T-E has no VDIMM options so the ram is running a cool 2.5 volts. I am running 2.96 actual on the BH-5. No need to wonder what CAS settings or VDIMM to use with RDRAM, or which brand of ram to get. To be fair, my PC1066 can run 156/4x, well above average so I got lucky there.

Buffered bandwidth on the AMD is 3450/2440. On the P4T-E it is near 4000. Unbuffered bandwidth? 2500 on the RDRAM, maybe 1600 on the AMD.

The VCORE on the 2.0A is 1.49 actual compared to 2.0 on the AMD. CPU temps on air is around 35 C (Alpha8942).

Overall, the RDRAM rig boots up faster and is quicker overall, seat-of-the-pants. The rig seems more stable and bulletproof. While 3D is smoother on an AMD rig compared to a P4, it's not quite as smooth as an RDRAM rig.

Can you imagine the guys out there still running an AMD rig like this? The NF7-S is still a very popular board. Look at some of the forums over the web: there are more overall AMD posts than Intel posts. Life could have been much simpler for these AMD guys, if they had an obsolete RDRAM rig :p .

FIZZ3

11-04-04, 03:55 AM

Can you imagine the guys out there still running an AMD rig like this? The NF7-S is still a very popular board. Look at some of the forums over the web: there are more overall AMD posts than Intel posts. Life could have been much simpler for these AMD guys, if they had an obsolete RDRAM rig :p .

A contemporary single channel DDR PC2100 would have fared less well, I agree. Still, to be fair, the RDRAM option was more expensive at the time. It's also illuminating to realize that the "death to rambus" obsolesence is no worse than the obsolesence of slow speed DDR. Kind of proves that there really is no future-ready hardware in PC land.

Talking about obsolete stuff; I'm off to play with the P3-866 system I got and do some parts swapping with my own old P3-800. One test I plan to do is to compare the FSP 250 watt PSU with the 300W PCP&P and see how it affects max overclock... if at all! Should be interesting.

Clevor

11-04-04, 05:58 PM

A contemporary single channel DDR PC2100 would have fared less well, I agree. Still, to be fair, the RDRAM option was more expensive at the time. It's also illuminating to realize that the "death to rambus" obsolesence is no worse than the obsolesence of slow speed DDR. Kind of proves that there really is no future-ready hardware in PC land.

Talking about obsolete stuff; I'm off to play with the P3-866 system I got and do some parts swapping with my own old P3-800. One test I plan to do is to compare the FSP 250 watt PSU with the 300W PCP&P and see how it affects max overclock... if at all! Should be interesting.

Fizz, I also have a dedicated Coppermine rig. Currently I'm running a CUV266 DDR board with a cDO 850E at 144 fsb or 1223 mhz (1.90 volts). Don't laugh about the VIA chipset board; it's one of the best overclocking boards I've seen, and since it uses DDR, I can easily run 160+ fsb at 2-2-2-5 timings with 512 MB or more of ram. You don't need good DDR for this, and 3 sticks will run as easy as one. Try that with SDRAM. No real gain in performance though, as buffered bandwidth is the same as good SDRAM running 2-2-2-5 (around 1000-1100).

As I got five rigs, I can swap them in the time it takes to plug one in and hook up video cables. In 2D stuff like installing software, booting up, and internet/e-mail, this Coppermine rig is quicker than an AMD or P4 rig running default 133/100 clock, even though you're talking 1.8-2.53 gig for the latter. A P4 rig running a 2.0A at 100 fsb is DOG SLOW.

Same with a 440BX board like a P3B-F at max 150 fsb: seat-of-the-pants, it smokes. In 2D, I can't tell the diff between it and a 3.0C P4 running 3.85 gig and 3400 unbuffered.

I have this CUV266 set up as my nostalgia rig: got a Voodoo5 in it right now. The 3DFX diehards hacked some WinXP drivers and OpenGL tricks so I was actually running Doom3 on the rig! Even had a Diamond MX300 Aureal sound card in it for awhile, except they never developed WinXP drivers so I had to take it out and put in a SBLive (hadda switch to XP to try Doom3).

Again, though slower in benchmarks versus 800 fsb DDR boards, the RDRAM rigs are inexplicably the fastest in booting and exiting WinXP, even with 512 MB ram versus 1 GB of DDR.

larva

11-04-04, 09:45 PM

My favorite recipe in the BX days was a P3B-F running 150fsb 2-2-2 with the best Coppermine that would stabilze at that fsb at any point in time. I first started with 550e's on the first stepping for 825MHz, then moved to the 600e on the second (900MHz), and finally to a 700s on the C stepping that did 1050Mhz as core quality improved. These rigs were fast, and I still have a time-warp-new P3B-F and Slot1 700 awaiting a use. Add a Promise disk controller and some quality PC133 and this is still a useful system.

I also dabbled with Abit BF-6's at 160+fsb (at looser timings). They never did equal the P3B-F at 150fsb and 2-2-2. I see my current rig as the modern equivalent. I run the ram at 400MHz 2-2-2-5 with GAT. Even though it is quite feasible to run the memory bus at higher clock speed it is very hard to significantly increase the application performance. Just like it was hard to best a 150fsb 2-2-2 BX with more clock rate and looser timings.

In another replay of the past, I chose my current 2.8c CPU because it was both cheap and certain to run at my (this time) 250fsb target. I prefer to engineer the motherboard's operating parameters first, and select CPUs that are happy with the resultant CPU clock--now and then. And to be honest most times the best operating parameters are exactly what Intel intended in the first place, as these guys are some of the sharpest tools in the box.

While I could certainly achieve higher cpu and memory clock rates, I can't build an appreciably better system from a functional standpoint. And stopping short of flailing every single component for all it's worth pays the dividends of nearly silent operation, reduced heat output, and decreased power consumption. It also uniformly lowers the cost of support components like power supply and cooling gear.

I guess that has always been my goal with OC'ing. I want to elevate the FSB as much as possible without ruinous affects on stability, keep the memory clock rate low enough to allow razor-sharp memory calibrations, combined with a cpu that matches or exceeds Intel's top speed grades for a fraction of the cost of Intel's version. While my BX rigs weren't the first or the last time I achieved this type of success, they were simply outstanding systems and still occupy a warm place in my heart understood by all the old BX jockeys.

consumer9000

11-04-04, 10:16 PM

I'm glad to see that I am not alone in my quest for 850E power!!! :clap:
Thanks for the input guys, it's encouraging to hear that the issues I've been having with Prime95 may not be the result of the CPU or RAM. One amusing anomaly I have been noticing recently is that if I attempt to run two instances of Prime95 in torture test mode, one window will run for around 10 minutes and generate and error, while the other one, started at the same time running the same torture test will run for hours and hours without a single issue! Go figure. Also, the difference between 1MHz on the FSB (157-158 for example) means the difference between 20 minutes of P95 stability or hours of stability! Everything else is stable well after Prime95 decides to call it quits, though it always bugs me to think that I eventually small errors could fux0r my comp...so I err on the side of caution (what am I doing overclocking? lol) and keep my OC's Prime95 kosher.
I can run mine all day long at 23*158 without event, though I have to have my VCore hovering around 1.6-1.65. Speaking of hovering, when watching Asus probe my system will show a Vcore in excess of 1.7 at startup, but it will gradually settle down to 1.625 or so. Is this normal? I just don't wanna cook my chip to death (Idle 70F 140F load)
As for memory testing, I have tested precisely 7 different RIMMS. I used to work in a computer shop and the vendor would allow us to return unneeded hardware. I would use this to cherrypick the best parts. Anyhow, their entire inventory was 6 256MB RIMMS so I tested each one individually, and found the variations in performance amazing. Unfortunately, even two great performers, when run in pairs seem to suffer and lose the ability to overclock as well. A single RIMM is ideal, but I broke down and went for the extra ram instead...even so, it's plenty quick though 4X is out of the question. (windows will hang) Now that I'm done raping my Radeon I can resume tinkering with the CPU/Mobo... I've got some interesting plans for cooling and I'll keep you posted.

FIZZ3

11-05-04, 06:53 AM

Clevor: I have not been able to enjoy quick WinXP booting too much; my raid configuration and NIC claim 90% of the boot delay! ;) I've always wondered why it is so necessary to laboriously scan the entire PCI bus for "surprise disk additions" at every reset or why I would want a 10 second delay to "boot from network". Technology at its best. :D

larva: yeah BX boards were really smooth. I've built a system for my brother years ago on an Abit BX6 board. He's been running a 700E at 933 for years now, without any voltage changes and without more than stock cooling (and no case fan!).

consumer9000: I am running at 4x using dual 256Mb sticks. My board is 28xxx series. What serial number is your board? An older one may benefit more from a VRIMM modification, if you want to go to that length for a bit extra speed.

consumer9000

11-05-04, 10:49 AM

Here's the serial number off the box:
2CZG074627

Please enlighten me as to which generation this board is. When I purchased it I asked for the newest/final revision, and this is what the vendor shipped me.

VRIMM mod would be great, as I currently have no way of getting more juice to that stupid ram! Otherwise everything else is simmering along with hopped up voltages! :burn:

FIZZ3

11-05-04, 11:31 AM

That is a new revision of the board, you should be fine. Here's the link to the VRIMM mod (don't mind the funny URL, it's legit):

http://www.thenakedreview.com/index.php?p=showarticle&id=82&showpage=3

consumer9000

11-06-04, 01:02 PM

Thanks for the link! I had always wondered why this board, with all it's options omitted any rimm voltage tweaks. It appears that Rambus has a lower tolerance for voltage deviation according to the article, but I suspect that with adequate cooling I should be able to get away with some extra juice! ;)

FIZZ3

11-06-04, 01:43 PM

Speedy Gonzales

06-29-05, 08:51 PM

Clevor

06-29-05, 11:39 PM

Wow! Uhmmm... :confused: Did I read it right? It actually IS
possible to run 800FSB HT Pentium CPU's on the 533FSB P4T
serie boards? How is that possible then? Can you tell me the
tricks please? And would they also work for my P4T-E's?
The P4T-E's are 400FSB boards that can run 533FSB Intel
CPU's when performing a few tricks, see. Thanks very much
in the advance. :)

Wow, this post dug up again?

Nope, you can't really run 800 fsb NWs on any of the RDRAM boards. You can only run one HT CPU: the 3.06B, a 533 fsb CPU. And you can only run this on the P4T533-R (32-bit RIMM) or the P4T533-C (16-bit RIMM). Won't run right on the P4T-E or TH7II-RAID. There's also an Iwill 850E board that will run it.

I did try to run a 3.2C on the P4T533-R, but it didn't run right. I got it to 190-195/3x, but it wasn't totally stable. I figured even if I got it to default clock that is 200 mhz at 3.2 gig, but no go. Same deal with a 3.4C: 3.4 gig at 200 mhz is sweet on those boards. I doubt Asus developed a BIOS to enable this either.

Speedy Gonzales

06-30-05, 05:09 AM

Months ago, I think I've read information on a site
somewhere, of a company that manufactures and sells
slotkets with which you can put a 800FSB 478 pins P4
CPU on a 533FSB mobo... But I didn't save the site into
my favourites. Now that I am starting to get more and
more interested into that matter again though, I will try
and locate that certain site and info. And if I find it, I will
then also try and find out,if it's also compatible with the
P4T-E and P4T533 series. If I have results then I will post
them here. But don't pin me on it I'd say, cuz I'm a busy
person and it's not of my highest priorities. ;)

Clevor

06-30-05, 05:51 PM

Speedy Gonzales

07-01-05, 04:12 AM

Ok, no HT then... seems logic to me too indeed... But
maybe the P4T-E would recognise and use it as a non-HT
CPU? I'm considering trying that out by using a Turbo PLL.
But I don't know yet -when- I will do that. :-)

batboy

07-01-05, 09:07 AM

There were some i845 overclockable mobos (for example, the Abit BE7 and the Asus P4PE and others) that if you had good cooling and got a good mobo and updated the BIOS, they sometimes hit 200 FSB and run a 800 bus processor at default speed. But, that ain't gonna happen with a i850 chipset mobo.

Speedy Gonzales

07-01-05, 09:17 AM

Ah, I see. And why not then? When using a Turbo PLL to
bypass the FSB frequency limitations that even the latest
BIOS version for the P4T-E's has, then using the P4T-E's
jumpers instead of the BIOS to change the settings and
(maybe) using the voltmods that are possible to perform
on the P4T-E's, what would I need a BIOS update for?
And I could use some serious cooling in case of over-
clocking the Intel 850 chipset, I assume? Or do I miss
some crucial information here? Looking forward to your
reply and thanks in the advance.

Speedy Gonzales

07-01-05, 09:29 AM

batboy

07-01-05, 10:25 AM

Perhaps if you mod enough things, you might get close, but why bother? You still won't have HT and the AGP will only be 4X too. I have never seen or heard about a i850 running 200 FSB. My old Abit TH7-II would flake out about 157 FSB. I've heard of a few rare i850 mobos hitting approximately 170 FSB (give or take).

Ok, now let's look at the RAM situation. Admittedly, some good Samsung PC1066 RAM might hit PC1200 speeds, but not all would. Even if the RAM can handle that much O/C, the IC's on the mobo might not. By the way, what RAM do you have? Let's do the math: Since you can't use the normal 4X memory divider at those speeds, 3 X 200 FSB X 2 = PC-1200. Good luck.

Time to sell that antique mobo and RAM, then get a decent up to date mobo and nice DDR RAM.

Speedy Gonzales

07-02-05, 04:21 PM

Nah. I like playing with antigue stuff just as much as with modern stuff.

Clevor

07-10-05, 06:00 PM

The P4T533 and P4T533-C have a version of the 850 chipset
too, by the way, just like the P4T-E has a chipset out of that
serie. The only difference is that the P4T533 and P4T533-C
have the 82850E where the P4T-E has the 82850 without that
added E. Are you aware of all this? And does that E, whatever
it stands for, make such a big primary difference, that the
PT4533 is able to run a HT CPU but a P4T-E isn't? I wonder
what that difference is then? Does anybody here know the
answer please? Thanks! :)

I've seen what happens when you try to run a HT CPU on a board that does not support it: Iwill P4T533-R, an 850E board. When I installed a 3.06B with a BIOS that does not support HT, everything ran in slow motion in Windows. With I flashed to the HT BIOS, everything ran normal. The P4T-E doesn't even have a BIOS that supports HT. You could try flashing to the P4T533-C BIOS at your own risk, but I think you need the 850E chipset.

It's was not easy to find, but I still have Samsung PC1066 that run 156/4x (four sticks :)). If you do the math, that's 208/3x. That's why I tried the 3.2C in the P4T533-R, as my RIMM4200 does 158/4x (210/3x). In actuality, you won't hit your theoretical max even if you can get that high. I even have one 256 stick that does 170/4x, and I did not test higher!

BTW, Macci did use a TurboPLL to overclock a heavily modded P4T-E to something like 188/4x. He was getting tremendous bandwidth at the time (around two years ago). He had the board VDIMM modded to the max. Of course the benchmarks would pale in comparison to now. But the RDRAM boards still find utility in stuff like video editing. The only thing holding you back for gaming (you can always run a killer video card) is the difficulty of getting bigger than 2x256 sticks to do 156/4x.

I did run 4x256 PC1066 at 156/4x Memtest clean and 3D stable!!! But I did not want to risk killing a stick for whatever reason, and prefer to keep 2x256 out of the board as spares. Apparently the RDRAM boards do not have trouble running 4 sticks of ram, as you would think. But it's all in the sticks; it's very difficult to find good RDRAM, which is the reason I still keep the platform. I would get nothing for the ram and board on the used market here in Japan. It would be a shame to be offered $15 for a 256 stick that does 170/4x.