E4400/DDR2-667 on P35-DS3R --> FSB wall?

[VirtualLarry]

Originally posted by: watek
Probably fsb wall on your cpu chip. You just got one that does not oc well.

Same with my E4500. I can boot 340 x 10, all day long, orthos hours and hours but can't do any FSB above 350.

I vote for the same. Your RAM should be good for 800 (400Mhz doubled).

I have a couple of P35-DS3Rs also, with E2140s in them, and they are running at 8x400 at default voltages on the RAM, MCH, and VTT. The board can handle 400Mhz easy, your RAM should too (perhaps with +0.2v on vdimm). So it's gotta be your chip, man.
Especially looking at your voltage curves, your chip is looking like a dud for OCing already.

I would check one other thing, is performance enhance set to "standard"? It should be if you are overclocking.

 

[nullpointerus]

Yeah, "performance enhance" is set to "standard."


Good news...!!

I *may* be able to talk someone (not an o/c-er) into buying my current MB+CPU+RAM now that I'm confident it's stable at stock speeds. So I really might get a Phenom BE soon. I know that's not the most cost-effective solution, and o/c-ing is very sporadic on that platform -- hey, some of us just like to be different! -- but do you have any advice about overclocking in case I happen to get a decent chip?

Here's what I'm looking at:

Phenom 9600 BE 2.3 GHz AM2+
MSI K9A2 Platinum AM2+ 790FX

w/ a type of RAM modules (total: 4GB)

1x OCZ Platinum 4GB (4 x 1GB) DDR2-800 (4-5-4-18 @ 2.1v)
1x Patriot EP 4GB (4 x 1GB) DDR2-800 (4-4-4-12 @ 2.2v)

-or-

Practically anything w/ heat spreaders from here:
http://www.newegg.com/Product/...mance+or+Gaming+Memory

-or-

2x WINTEC AMPX 2GB DDR2-800 (5-5-5-15 @ 1.9v)
2x G.SKILL 2GB DDR2-800 (5-5-5-15 @ 1.8v-1.9v)
2x GeIL Value Series 2GB DDR2-800 (5-5-5-15 @ 1.8v)

-or-

2x WINTEC AMPX 2GB (2 x 1GB) DDR2-1066 (5-5-5-15 @ 2.2v)
2x Kingston HyperX 2GB (2 x 1GB) DDR2-1066 (5-5-5-15 @ 2.2v)
2x OCZ Micron D7-based 2 GB (2 x 1GB) DDR2-1066 (5-5-5-15 2.1v - 2.3v)

While the BE CPU has a multiplier that is upwards unlocked, I have heard that the Phenom platform *really* likes extra RAM bandwidth. Should I be aiming for 800 or 1066 MHz RAM? I'd want 4 GB, but probably no chance of upgrading beyond 4 GB on that rig, so either 2 x 2GB or 4 x 1GB would be fine. (Heck, 4 x 2GB would be fine, if it's *really* cheap.) I do not want to go *very* high on the memory voltage, at least not past the specs (up to 2.2v).

FYI, my new CPU cooler would hang over the RAM slots, so no high RAM heatspreaders, OK?

 

[watek]

Are you not satisfied running 340 x 10?

You have a decent system. 3.4 Ghz is good enough for anything right now.

Maybe give some time until you have money to save up for another cpu. You have a good m/b.

 

[watek]

Are you not satisfied running 340 x 10?

You have a decent system. 3.4 Ghz is good enough for anything right now.

Maybe give some time until you have money to save up for another cpu. You have a good m/b.

 

[nullpointerus]

Originally posted by: watek
Are you not satisfied running 340 x 10?

You have a decent system. 3.4 Ghz is good enough for anything right now.

My max. stable o/c is 2.85 GHz. As soon as I hit 2.9 GHz, voltage requirements go *way* up, and by 3.0 GHz my CPU requires 1.55v (BIOS). I have decent cooling (< 50C Prime95 Small FFT). There's just no way I can hit 3.4 GHz with it... It's just a dud...

Maybe give some time until you have money to save up for another cpu. You have a good m/b.

Yes, I'm sure it's a good MB for overclocking. :thumbsup:

My only real problem with the P35-DS3R is the small passive heatsink that resides under the PCI-E 16x slot. This blocks most of the SB's airflow, and there's no heatpipe on the MB to link the SB to a bigger or, at least, a better-positioned heatsink.

My thinking with the Phenom BE is that it's cheaper (for quad core); it has an unlocked multiplier; a C2D loses certain optimizations (e.g. dynamic branching?) in 64-bit mode; my case has tons of cooling; my current MB is going to get pulled out anyway to install a new CPU cooler, and, well, I've just wanted to play with a new AMD chip.

Who knows? Maybe I'd get lucky.

The problem is selecting the best RAM for the Phenom platform. Is DDR2-1066 worth the price premium? IOW, does it add significant performance in real-world applications?




Does anyone here have experience with Phenom?

 

[Tempered81]

lol

e4400 do this: 333 FSB x 10 multi 1.51-1.54 volts vcore in bios
Divider 1:1 Ram speed 667mhz Ram voltage 1.9-2.0volts
MCH +1
FSB +1

3.3with adequate cooling

 

[nullpointerus]

Originally posted by: jaredpace
lol

e4400 do this: 333 FSB x 10 multi 1.51-1.54 volts vcore in bios
Divider 1:1 Ram speed 667mhz Ram voltage 1.9-2.0volts
MCH +1
FSB +1

3.3with adequate cooling

My chip would never post at 333x10 with 1.55v in BIOS. I've tried it...

As I said earlier:

FYI, I've done many tests in the past, and here are my stable results with max CPU o/c:

10 x 266 = 2660 MHz @ 1.375v (BIOS)
10 x 275 = 2750 MHz @ 1.387v (BIOS)
10 x 280 = 2800 MHz @ 1.400v (BIOS)
10 x 290 = 2900 MHz @ 1.450v (BIOS)
10 x 300 = 3000 MHz @ 1.550v (BIOS)
> 3000 MHz (not stable at all)

...so this exponential curve implies that I've properly identified the max. CPU frequency for this chip.

(Before anyone posts common suggestions for 3+ GHz, forget it -- this chip just won't go that high. )

 

[Tempered81]

Im pretty sure there is No reason why it would not if you did what i said - unless you have too much vdroop / vdrop

 

[Tempered81]

What you could do, if you think your chip requires an insane ammount of vcore just to boot over 3.0ghz, which could be, is just create a profile in your gigabyte bios with the settings i showed you. (make sure you arent manually borking your memory timings, let it run by SPD) Start at a low fsb like 295 and try to boot, make note of the minimum Cpu Vcore it requires to boot completely into windows. Test this, and keep increasing your FSB, Find the point where you need to go back into the bios to give it more Vcore to boot to windows. Once you find the relationship, and make notes of min required vcore, while keeping the 10x multi, increase setp by step vcore/FSB accordingly til you hit 1.55 in bios.

Tell me what your max "xxxFSB x 10 multiplier" is, and what voltage it took.

like ex:

295 x 10 = 2950mhz 1.425 vcore in bios
298 x 10 = 2980mhz 1.44 vcore
300 x 10 = 3000mhz 1.45 vcore
303 x 10 = 3030mhz 1.47 vcore in bios

 

[nullpointerus]

Hmm... I'll try it once more.


Successful Windows Boots

288 x 10 = 2880mhz @ 1.41250v
291 x 10 = 2910mhz @ 1.41250v
294 x 10 = 2940mhz @ 1.41875v
297 x 10 = 2970mhz @ 1.41875v
300 x 10 = 3000mhz @ 1.43125v
303 x 10 = 3030mhz @ 1.45000v
306 x 10 = 3060mhz @ 1.45625v
309 x 10 = 3090mhz @ 1.46875v
312 x 10 = 3120mhz @ 1.48750v
315 x 10 = 3150mhz @ 1.51875v
318 x 10 = 3180mhz @ 1.51875v
321 x 10 = 3210mhz @ 1.55000v

(All voltage is what's listed in the BIOS, not the actual vcore.)


Notes

-- Usually when Windows boot fails, I'll get a BSOD code 124 "uncorrectable hardware error."

-- On unstable settings, my drive detection may also be unusually slow (2x to 3x as long) in the AHCI/RAID BIOS.

-- When running 300 x 10 = 3000mhz, Windows boot succeeded at 1.45000v and 1.46250v and I even started stress-testing, yet Windows boot later failed at 1.47500v...?! Wierd.


*** I'm currently stress-testing 300 x 10 = 3000mhz @ 1.48750v (BIOS) ***

 

[Tempered81]

try 300x10, and opent cpu-z and tell me the speed & timings of your ram

also, at 300x10, report the idle Vcore, Load Vcore (you can see these in cpu-z) and Idle & load temperatures... load up two instances of prime, and get a cold idle temp & and the highest load temp you see after about 3-10 minutes (if you can prime that long)

edit: also, what do you see if you select 321x10 with bios vcore of 1.51875

 

[Doclife]

Have you updated your motherboard with the latest BIOS ? I'm using the DS3R mobo and before the BIOS update I can only go to 3GHz max with the E4500 & HP 667Mhz RAM. After the BIOS update, I can go to 3.2GHz.

 

[nullpointerus]

Originally posted by: jaredpace
try 300x10, and opent cpu-z and tell me the speed & timings of your ram

300.0 MHz
1:1
5-5-5-14
2T

also, at 300x10, report the idle Vcore, Load Vcore (you can see these in cpu-z) and Idle & load temperatures... load up two instances of prime, and get a cold idle temp & and the highest load temp you see after about 3-10 minutes (if you can prime that long)

idle: 1.456v / 25C (17C per core)
load: 1.424v / 57C (51C per core) -- after 1 hr. Small FFT w/ Tuniq fan on auto (2100rpm @ load)

edit: also, what do you see if you select 321x10 with bios vcore of 1.51875

I see the same Windows boot error (BSOD code 124 "uncorrectable hardware error").

Originally posted by: Doclife
Have you updated your motherboard with the latest BIOS ? I'm using the DS3R mobo and before the BIOS update I can only go to 3GHz max with the E4500 & HP 667Mhz RAM. After the BIOS update, I can go to 3.2GHz.

Yes, I'm using the F11 BIOS.

But I have the rev. 1.0 board, which may be why I saw no change.

 

[lopri]

RAM working DDR2-800 @stock setting (in this case 200FSB 1:2) doesn't necessarily guarantee that it will work at the same frequency at 400FSB 1:1. Unfortunately on Intel platform, everything is so tightly woven together and "divide and conqure" doesn't work. So yeah it could be RAM or it could be the board. Then again, I have seen so many E2xxx/E4xxx stopping around 1)350FSB 2)425FSB, so it won't surprise me if it's the CPU itself that has a wall.

 

[hokiealumnus]

Sorry, I've got nothing on why your chip can't get to 3.0GHz with much lower Vcore if you're doing it with a 1:1 CPU : DRAM ratio. What is your chip's VID?

The one thing I can answer for you is that E2xxx & E4xxx chips do NOT like high FSBs and lowered multipliers. As an example, mine will only make it to 380, regardless of what multiplier I use. I was able to get the chip to 3.5GHz @ 1.565Vcore (set, 1.52 after Vdroop) but only with the default multiplier of 10.

For reference, the OC in my sig is set Vcore and it runs at 1.41 after Vdroop.

 

[nullpointerus]

Originally posted by: lopri
RAM working DDR2-800 @stock setting (in this case 200FSB 1:2) doesn't necessarily guarantee that it will work at the same frequency at 400FSB 1:1.

Thank you!! Now everything makes sense!

It also ties in with the thread that demonstrates memory dividers over 1:1 do little for real-world performance at the same FSB. The CPU simply can't push the RAM hard enough with the lower FSB. Things seem decoupled, and there is some improvement with DDR2-667 memory at a 266FSB, but it's not the same in real world usage as 333FSB would be.

My ideal setup (w/o swapping hardware) would be:
-- CPU variable from 333*6 (1998) to 333*9 (2997) MHz w/ EIST/C1E
-- FSB of 333 MHz @ stock v.
-- MCH @ stock v.
-- RAM at 333*2 (666) MHz 5-5-5-15 2T 1.8v (i.e. completely stock)

Results:
-- I know the 333*9 will run a little hotter than 300*10, but I have great cooling.
-- The higher FSB would allow my CPU to fully stress my RAM, which is excellent.

However, in its natural state, the EIST/C1E feature, plus a 333FSB, would put the CPU up to 333*10, which is much too high for this chip. Is there any way to edit the EIST/C1E profiles so that they only range from 6x to 9x? I think I saw a program that could do this, but I can't remember the name. And it might not work in Vista x64...

I suppose a fixed 333*9 w/o EIST/C1E is the next best solution.

Unfortunately on Intel platform, everything is so tightly woven together and "divide and conqure" doesn't work. So yeah it could be RAM or it could be the board. Then again, I have seen so many E2xxx/E4xxx stopping around 1)350FSB 2)425FSB, so it won't surprise me if it's the CPU itself that has a wall.

Cool, that's just what I needed to know.

To go any higher than about 3.0 GHz, I'd have to:
-- upgrade CPU (E8400, Q6600, or get luckier w/ another E4x00)
-- upgrade RAM (DDR2-800 or DDR2-1066 for 1:1 w/ high FSB)



That's fairly pricey...

If I were spending that much, I'd rather go Phenom (just because it interests me!).

 

[perdomot]

nullpointerus,
Couple of things. You said your ram timings are 5-5-5-14 which seems a bit odd as most ram would be 5-5-5-18. Might want to change that setting. I used this article to help with my settings. There is one there for e4400 as well but the e4300 shows all the settings: http://www.legionhardware.com/document.php?id=610

 

[BonzaiDuck]

Can't read through all these posts, and maybe there's already something here to refute me.

But:

1) Just because it passes MEMTEST86+ or MEMTEST86 @ 825 Mhz doesn't mean the system will post. I've seen this happen with three systems, and two different motherboards, using high-quality DDR2-667 memory kits (Crucial Ball'x).

In one case, I'd run MEMTEST86+ for about 8 or 9 hours without a single error, and several hours under iteration of test #5. But "no cigar" in Windows.

2) With "Value-RAM" or RAM rated below the speed you're trying to achieve, you're taking a gamble of some sort unless some reviewer at a serious benchtest review web-site (like CD Info, Tech Report, X-bit Labs, etc.) has proven the elasticity of the RAM at its stock (or tighter) settings for speeds exceeding the spec and voltages within the memory's rated recommended warranty maximum.

So I'm not saying "don't do it." I'm offering the caveat that you take a risk doing this unless you've seen reliable evidence that it will work. Ultimately, it's not the rated spec for the memory that counts so much as its ability to meet your over-clocking expectations, speed-range and memory multiplier/divider/CPU-to-RAM-ratio.

Couple weeks ago, I got two kits of Crucial Ballistix DDR2-667 2GB 2x1GB modules for an after-rebate price of $25. There were two rebates allowed per household. As the orders streamed in to NewEgg, the reseller bumped up the price on 'em twice!! So $25 became $30, which became $35.

There are definitely bargains to be had. Since Crucial asked me if they could use my "RMA-thank-you" letter in promotions which I wrote in response to their e-mail issuance of my RMA number, I'd like to imagine they'll give me some "freebie," but I'm not holding my breath. In fact, I should feel lucky for how they turned around my DDR2-1000 Balli-sticks, which had about eight months of stressful mileage on them.

AND -- TO SPEAK OF THAT EXPERIENCE: You think that pushing the bus-speed above spec can stress memory rated at a lower speed? I was running those DDR2-1000's at a [DDR] speed of 712 Mhz. VDIMM was at -- and for a longer time, below -- the warranty recommended max.

 

[Tempered81]

hrm so you went from

"300 x 10 = 3000mhz @ 1.43125v (All voltage is what's listed in the BIOS, not the actual vcore.)"

to

also, at 300x10, report the idle Vcore, Load Vcore (you can see these in cpu-z) and Idle & load temperatures... load up two instances of prime, and get a cold idle temp & and the highest load temp you see after about 3-10 minutes (if you can prime that long)

idle: 1.456v / 25C (17C per core)
load: 1.424v / 57C (51C per core) -- after 1 hr. Small FFT w/ Tuniq fan on auto (2100rpm @ load)

You must have seriously increased your vcore in the bios to tell me those readings.

sounds fishy, maybe you are better off with the intriguing phenom

 

[Tempered81]

If your goal is for greatest overall cpu speed with stability. I would go for 333x10 and run the mem at 1:1 1.55 in bios should be more than adequate and you have nice load temps with that tuniq. might need to bump up the ram volts a little and the mch. You're on a great board for OCing.

If you don't like it, im sure you could get a great deal for the sell of the mobo +cpu combo in the for sale forums.

 

[nullpointerus]

Originally posted by: jaredpace
hrm so you went from

"300 x 10 = 3000mhz @ 1.43125v (All voltage is what's listed in the BIOS, not the actual vcore.)"

to

also, at 300x10, report the idle Vcore, Load Vcore (you can see these in cpu-z) and Idle & load temperatures... load up two instances of prime, and get a cold idle temp & and the highest load temp you see after about 3-10 minutes (if you can prime that long)

idle: 1.456v / 25C (17C per core)
load: 1.424v / 57C (51C per core) -- after 1 hr. Small FFT w/ Tuniq fan on auto (2100rpm @ load)

You must have seriously increased your vcore in the bios to tell me those readings.

sounds fishy, maybe you are better off with the intriguing phenom

What sounds fishy?



I said I was testing 300*10 @ 1.48750v, and then you asked me to open CPU-Z and tell my results...right?

If your goal is for greatest overall cpu speed with stability. I would go for 333x10 and run the mem at 1:1 1.55 in bios should be more than adequate and you have nice load temps with that tuniq. might need to bump up the ram volts a little and the mch. You're on a great board for OCing.

It takes 1.55v to get 321*10 MHz to boot into Windows...

There's NO way I can run 3.3 GHz stable at 1.55v. It doesn't even POST there.

 

[nullpointerus]

Originally posted by: perdomot
nullpointerus,
Couple of things. You said your ram timings are 5-5-5-14 which seems a bit odd as most ram would be 5-5-5-18. Might want to change that setting. I used this article to help with my settings. There is one there for e4400 as well but the e4300 shows all the settings: http://www.legionhardware.com/document.php?id=610

My RAM's actually rated for DDR2-667 5-5-5-15 @ 1.8v, so it should work fine at 600mhz 5-5-5-14 @ 1.9v. The -14 part is a bit odd, but that's where the BIOS calculates the timings to work at the reduced speed of 600 vs. 667 stock with its DDR2-667 SPD of 5-5-5-15.

In any case, the extra 0.1v on the RAM while underclocked should take care of any (minor) timing issues, and even if that's not the case, the CPU voltage is what's directly affecting the stability of my CPU o/c-ing, so I'm not limited by RAM until *after* 333FSB.

Good advice, though, in the general case.

Originally posted by: BonzaiDuck
Can't read through all these posts, and maybe there's already something here to refute me.

But:

1) Just because it passes MEMTEST86+ or MEMTEST86 @ 825 Mhz doesn't mean the system will post. I've seen this happen with three systems, and two different motherboards, using high-quality DDR2-667 memory kits (Crucial Ball'x).

In one case, I'd run MEMTEST86+ for about 8 or 9 hours without a single error, and several hours under iteration of test #5. But "no cigar" in Windows.

Cool, that's already been said, but further confirmation is much appreciated! :thumbsup:

2) With "Value-RAM" or RAM rated below the speed you're trying to achieve, you're taking a gamble of some sort unless some reviewer at a serious benchtest review web-site (like CD Info, Tech Report, X-bit Labs, etc.) has proven the elasticity of the RAM at its stock (or tighter) settings for speeds exceeding the spec and voltages within the memory's rated recommended warranty maximum.

So I'm not saying "don't do it." I'm offering the caveat that you take a risk doing this unless you've seen reliable evidence that it will work. Ultimately, it's not the rated spec for the memory that counts so much as its ability to meet your over-clocking expectations, speed-range and memory multiplier/divider/CPU-to-RAM-ratio.

Yep, since the CPU and RAM are both o/c duds, I'd have to upgrade both to increase o/c potential.

If I can keep 333*9 with this CPU+MB+RAM, then upgrading's not cost-effective...for me.

Couple weeks ago, I got two kits of Crucial Ballistix DDR2-667 2GB 2x1GB modules for an after-rebate price of $25. There were two rebates allowed per household. As the orders streamed in to NewEgg, the reseller bumped up the price on 'em twice!! So $25 became $30, which became $35.

There are definitely bargains to be had. Since Crucial asked me if they could use my "RMA-thank-you" letter in promotions which I wrote in response to their e-mail issuance of my RMA number, I'd like to imagine they'll give me some "freebie," but I'm not holding my breath. In fact, I should feel lucky for how they turned around my DDR2-1000 Balli-sticks, which had about eight months of stressful mileage on them.

I see a lot of rebates. I guess the best strategy would be to just keep my eyes open. DDR2 is not going anywhere soon because DDR3 is so extravagantly priced for little to no actual performance advantage, let alone widespread support. Any DDR2 I pick up should be able to work just as well in my *next* rig.

AND -- TO SPEAK OF THAT EXPERIENCE: You think that pushing the bus-speed above spec can stress memory rated at a lower speed? I was running those DDR2-1000's at a [DDR] speed of 712 Mhz. VDIMM was at -- and for a longer time, below -- the warranty recommended max.

So you're saying the RAM required a higher (normalized) voltage at a higher bus speed while below mem. spec?

 

[nullpointerus]

After testing Prime95 Blend for 5 hr. w/ the following:

CPU: 300 * 10 = 3000 @ 1.48750v (1.424v/1.456v load/idle)
FSB: (stock v.)
MCH: (stock v.)
RAM: 300 * 2 = 600 5-5-5-14 2T @ 1.9v (1.85v/1.87v load/idle)

...the system appears to be quite stable (so far). :thumbsup:

I had to keep the +0.1v on the RAM because actual voltage in SpeedFan 4.33 drops to 1.87v (idle) when the system is overclocked w/ 4 DIMMs and 1.85v (load). There's no way I'll trust my system to run 24/7 with undervolted value RAM!!




Now I'll switch to 333 * 9 (CPU) and 333 * 2 (RAM) w/ same voltage/timings and retest overnight.

(These are my target settings for the long term. )


A big thanks to all who participated, regardless of whether we agreed!

The suggestions and debates help keep things real and concrete.:thumbsup:

 

[nullpointerus]

After reading some reviews, I don't see the point in disabling EIST to get a +33FSB. The performance benefit simply won't be there, except in a few synthetic benchmarks. I'll just keep this rig running at the settings in the previous post (i.e. 300*10) and let the stability testing programs identify any potential problems.

 

[nullpointerus]

nvm