Q9550 8GB build - unstable at stock speeds

catalinr · Post by **catalinr** » Mon Dec 15, 2008 3:30 pm

Just completed a new build (refresh of my old PC, some components are reused):

PSU Enermax 625W Seria MODU 82+, ATX v2.3
MoBo Gigabyte GA-EP45-DS3R
CPU Intel Core2 Quad Q9550 2.83 GHz Yorkfield Box
RAM DDR2-800 8GB
Heatsink Thermalright HR-01 Plus
CASE P150

Fans: 90mm Nexus IN, 120mm Nexus OUT.

First I left the HR-01 fanless. First Prime 95 tests crash within seconds. Strange thing is that the core temps were all under 60 C. I also got a reset while I left the PC idle unattended.

Placed an older Scythe 120mm fan on the heatsink and discovered the FAN_CPU header on the mobo can only control PWM fans. My voltage-controlled fan stays at 1200rmp - loud.

Ran Prime95 and after 6 minutes - same error. Core temps all under 60 C.

I ordered a PWM 120mm Nexus to put on the heatsink but I doubt that is the reason. As I see it, it's either a faulty CPU or the MoBo.

Any other things I should try? Memory Test in Vista finds no problems with the memory.

Thanks,
Cata

bgiddins · Post by **bgiddins** » Mon Dec 15, 2008 3:55 pm

Pull 6GB of RAM and try starting it with 2GB? May be dodgy RAM? If it crashes again with 2GB, switch the 2GB over for another stick - you may have one bad stick and there's a 1 in 4 chance that's the one you'll try first.

xan_user · Post by **xan_user** » Mon Dec 15, 2008 4:26 pm

skip vista memory check and use memtest+ for at least 24 hours before deciding its good ram.

edit QVL memory linky.
http://www.gigabyte.com.tw/FileList/Mem ... 5-ds3r.pdf

Nick Geraedts · Post by **Nick Geraedts** » Mon Dec 15, 2008 5:01 pm

What program are you using to monitor the temperatures?

I'd also suggest running Memtest86+ on each of the sticks individually. Run at least one full pass on each to make sure they're ok.

You also haven't said what kind of RAM it is, or what the settings are at in BIOS.

Post by **NeilBlanchard** » Mon Dec 15, 2008 6:40 pm

Hi,

To add to what others have wisely suggested: are you using ECC RAM, and are you running it at stock voltage?

hybrid2d4x4 · Post by **hybrid2d4x4** » Mon Dec 15, 2008 7:23 pm

I have a Q9450 and 8GB OCZ Platinum in a 4x2GB config, and also had some instability as well right out of the box. On the auto setting, the mobo will attempt to run the mem at 1.8V (which is rated to run at 800 w 5-4-4-12 timings at 2.1 or 2.2V) and may not be stable with 4 sticks. Go into your BIOS and up the voltage on the ram a bit (I now run mine at 1.85V at the stock speed and timings, but ymmv). If you're running it at the manufacturer's specified voltage and it fails the Memtest the above posters mentioned, then RMA the RAM. Otherwise, it's more likely the Mobo than the CPU (at least due to thermals).

catalinr · Post by **catalinr** » Tue Dec 16, 2008 12:45 am

Thanks for all the advice.

The memory: 2x A-DATA 4GB DDR2 800MHz CL4 Vitesta Extreme Dual Channel Kit. Not ECC. It is a brand of memory I have at work on maybe 20 computers in 4GB and 8GB configurations, paired with E8400 and E6600 dual cores. This is the first Quad + 8GB though.

The BIOS settings are all defaults & AUTOs. CPUZ shows the memory at CL5 1.8V. I usually (after stability testing) set it at 2.0V & CL4 as spec'ed, but CL5 1.8V is the spec default.

I am using CPUZ, SpeedFan, CoreTemp to monitor the system and the latest Prime95 (with multithreading support) to stress test. Tried RealTemps but it showed 5-10 less degrees per core temp than the others so I discarded it. CORE0 consistently shows about 5-9 more degrees (C) than the others. I am not sure if that is relevant in any way.

I attempted to test with Memtest86+ but the boot DVD hangs when starting with the "Loading............" message. I don't think at that point it started testing memory yet so I suppose the MoBo isn't supported.

I got the PWM fan today. I will mount it tonight to make sure the cooling is adequate and I will start tweaking the memory BIOS settings. If that doesn't work I will try with a couple of known-to-work RAM sticks. If that doesn't work I will send the mobo back...

I don't know what else I should try, if have other suggestions...
Cata

Post by **NeilBlanchard** » Tue Dec 16, 2008 2:21 am

Hiya,

Have you tried booting with just one of the sticks?

It sure sounds like (at least) one of them is NG.

catalinr · Post by **catalinr** » Tue Dec 16, 2008 4:47 am

I determined that the latest Memtest86+ hangs loading on another 8GB config (E8400, GA-EP35C-DS3R). So it may be a 8GB issue. I guess I'll try tonight taking memory sticks out and testing separately. We'll see.

catalinr · Post by **catalinr** » Tue Dec 16, 2008 11:33 am

I can't believe it: the mobo was setting CL4 timings for the memory WITHOUT setting the voltage to 2V but leaving it at the default 1.8V instead.

Testing at 2V as we speak...

catalinr · Post by **catalinr** » Tue Dec 16, 2008 12:08 pm

That was IT. Damn, I'm dumb.

Prime95 ran OK until I got bored. No problems. Excellent. I removed the fan and all went well until the cores arrived around 80 C where I stopped it. I don't see any point in going further since I don't foresee many occasions where all 4 cores will run at 100% for over 10 minutes...

Now if only I can find something to replace the ruined (old) elastic HDD suspension bands in my P150...

Thanks for all the help,
Cata

Nick Geraedts · Post by **Nick Geraedts** » Tue Dec 16, 2008 12:27 pm

I would still run prime95 on blend for at least 6 hours (preferably 12) to make sure things are stable. Just because temperatures seem "ok" (80C is a little high by my standards) doesn't mean that the system is fully stable.

protellect · Post by **protellect** » Tue Dec 16, 2008 1:15 pm

I'd run it at 5-5-5-15 @ 1.8v. I've had really crappy luck with multiple computers running 8GB of memory, if I don't use the default voltage. [Both Corsair Dominator and Crucial Ballistix]

These are computers that do heavy multitasking [vmware, cs3, sharepoint, mysql, mssql, etc etc]

Run 5 timings @ 1.8v, and enjoy the stability.

catalinr · Post by **catalinr** » Tue Dec 16, 2008 3:14 pm

I've been quite happy with this memory brand so far. I have them running on a lot of PCs in both 4GB and 8GB configs, both 1.8V CL5 and 2.0 CL4 and they were rock-solid (other than a few sticks found faulty on arrival).

I got 80C in Prime95 only when fanless. With a fan I get up to about 50-56 which is more than OK. The problem though is that this fan is PWM and the mobo's algorithm starts it at 1200RPM which is LOUD.

So that is why I am thinking to either:
1. go fanless&quiet, being careful not to reach 100% usage for long times, eventually ducting the CPU heatsink to P150's exhaust fan which is at a constant 5V.

2. place a voltage-controlled constant 5V fan on the CPU; still not ideal but better

Of course all this is not helping the fact that my 2 HDDs sagging now in the crappy old P150 elastic bands are touching the case and transmitting vibrations... Maybe I should use the silicone screws instead? But I am afraid that the cables from HDDs are ALSO moving vibes.

Cata

Nick Geraedts · Post by **Nick Geraedts** » Tue Dec 16, 2008 3:56 pm

protellect wrote:Run 5 timings @ 1.8v, and enjoy the stability.

Why sacrifice the performance if the RAM can handle it and/or is rated for it? I've got 8GB of G-Skill DDR2-800 in my system running at 4-4-4-12 @ 2.1V and everything is just peachy. The DIMMs are warm to the touch when under load, but not hot.

catalinr wrote:I got 80C in Prime95 only when fanless. With a fan I get up to about 50-56 which is more than OK. The problem though is that this fan is PWM and the mobo's algorithm starts it at 1200RPM which is LOUD.

So that is why I am thinking to either:
1. go fanless&quiet, being careful not to reach 100% usage for long times, eventually ducting the CPU heatsink to P150's exhaust fan which is at a constant 5V.

I would NOT recommend this path. There are times when you can't always control how much CPU power is used (updates, malware scanning, HD playback). Your system should be able to handle 100% load without problems, even if it's never run that hard "most of the time".

catalinr wrote: 2. place a voltage-controlled constant 5V fan on the CPU; still not ideal but better

A much safe idea. Fully passive CPU cooling is something I'd never recommend, although ducting a fairly low power CPU to a nearby exhaust fan is sometimes acceptable.

catalinr wrote: Of course all this is not helping the fact that my 2 HDDs sagging now in the crappy old P150 elastic bands are touching the case and transmitting vibrations... Maybe I should use the silicone screws instead? But I am afraid that the cables from HDDs are ALSO moving vibes.

Some people swear by stretch magic (although it can be tricky to tie off), but I prefer using plain old shock cord. You can often find this in bulk at a local textiles store (think sewing stuff) for dirt cheap. Unless the cables are under tension, they shouldn't be transferring any large amounts of vibrations to the case.

catalinr · Post by **catalinr** » Wed Dec 17, 2008 12:31 am

Looks like I spoke too soon. With memory at CL4 2.0V, Prime95 did eventually FAIL on the hottest core (I am not sure that is a coincidence). Changed it to CL5 1.8V and it's been running without errors for about 8 hours now and still going. Both with the PWM-controlled fan installed on the CPU.

I think I'll go with the fan on CPU fixed at 5V then. I am afraid that this won't give the same stability under stress because it won't be able to increase speed with temperature, but it should be a good compromise, and SILENT (I think the Nexus is about 500 rpm at 5V)

Thanks so much,
Cata

protellect · Post by **protellect** » Wed Dec 17, 2008 7:58 am

Nick Geraedts wrote:
protellect wrote:Run 5 timings @ 1.8v, and enjoy the stability.
Why sacrifice the performance if the RAM can handle it and/or is rated for it? I've got 8GB of G-Skill DDR2-800 in my system running at 4-4-4-12 @ 2.1V and everything is just peachy. The DIMMs are warm to the touch when under load, but not hot.

I've built about 6 computers at work running a intel quad-core and 8GB of DDR2. Any of them I have put above 1.8v to get better memory timings, the sticks have produced BSODs within six months. Returning to 1.8v brought back stability and passed memory tests.

Additionally, any gained performance from better memory timings are fairly minimal, and to me [and someone else experiencing problems] not worth stability issues. See this review by tom's hardware.

http://www.tomshardware.com/reviews/ram ... ,1807.html

Post by **NeilBlanchard** » Wed Dec 17, 2008 9:10 am

Hi,

I can't see how you know which core caused the failure?

catalinr · Post by **catalinr** » Wed Dec 17, 2008 9:28 am

Prime95 creates 4 working threads, one for each core. Each thread has its own window, it reports the core it has the affinity set to and various status messages, including failure.

catalinr · Post by **catalinr** » Wed Dec 17, 2008 12:21 pm

Drives suspended, fans placed, running Prime95 beautifully.

One thing is not clear to me: the Nexus Real Silent 120mm fan comes with Molex plug. This plug does NOT use the yellow wire. It only sends a Red and a Black wire to the fan motor.

It was my understanding that Red = 5V and Black is ground. So does that mean that the fan is already running at 5V and thus I shouldn't put it through a 12V/5V molex splitter?

Cata

Nick Geraedts · Post by **Nick Geraedts** » Wed Dec 17, 2008 3:51 pm

protellect wrote:I've built about 6 computers at work running a intel quad-core and 8GB of DDR2. Any of them I have put above 1.8v to get better memory timings, the sticks have produced BSODs within six months. Returning to 1.8v brought back stability and passed memory tests.

Additionally, any gained performance from better memory timings are fairly minimal, and to me [and someone else experiencing problems] not worth stability issues. See this review by tom's hardware.

http://www.tomshardware.com/reviews/ram ... ,1807.html

I'm not suggesting that he overclocks his RAM beyond the rated timings and speed, but some RAM is designed for operation at voltages higher than 1.8V. I've got some OCZ RAM here where the label clearly indicates 5-4-4-12 timings @ 2.1V. If the manufacturer gives you ratings, the RAM should operate at that speed. If it doesn't, it's defective and should be returned for RMA. My G.Skill RAM is rated for 4-4-4-12 and 2.1V, so that's what my system is set to. For anyone who's interested or concerned about burning out the RAM, the memory heatspreaders are warm to the touch, but by no means hot (I can touch the sticks indefinitely).

The other thing that Tom's fails to test in any memory benchmark is latency. There are several programs out there where fast latency will give better performance than overall throughput (including Windows itself).

Post by **NeilBlanchard** » Wed Dec 17, 2008 4:16 pm

Hello Cata,

Does the fan's Molex plug into the yellow wire on the power supply plug, or the red wire from the power supply?

protellect · Post by **protellect** » Wed Dec 17, 2008 6:41 pm

Nick Geraedts wrote:I'm not suggesting that he overclocks his RAM beyond the rated timings and speed, but some RAM is designed for operation at voltages higher than 1.8V. I've got some OCZ RAM here where the label clearly indicates 5-4-4-12 timings @ 2.1V. If the manufacturer gives you ratings, the RAM should operate at that speed. If it doesn't, it's defective and should be returned for RMA. My G.Skill RAM is rated for 4-4-4-12 and 2.1V, so that's what my system is set to. For anyone who's interested or concerned about burning out the RAM, the memory heatspreaders are warm to the touch, but by no means hot (I can touch the sticks indefinitely).

Thats not exactly true. the JEDEC standard states that DDR2 Memory voltage should be 1.8V +- .1V. Any voltage beyond those specifications is overvoltage. There is no memory designed to do 2.0V+.

A manufacturer is enticing you with better timings and more voltage, effectively betting that your memory running at 2.0+V, and provide you with faster timings, will not fail, and if it does will warranty the product so.

Just because the manufacturer says it will do 2.1V doesn't mean that it should. In a similar fashion, one can over clock there core2duo processor possibly with some increased voltage, but it is NOT the manufacturers specification.

You can read about the JEDEC DDR2 specification here.

catalinr · Post by **catalinr** » Thu Dec 18, 2008 2:01 am

Darn, you're right: the fan's red wire was going into the source's yellow wire. So 12V. I changed to use the 5V splitter and now it's even quieter. AND it didn't go over 65 C after 1 hour of Prime95. Not bad!

An interesting note, maybe it will help someone. Another source of noise was coming from the suspended HDD. Turns out the SATA cable is stiff and at some point it was touching the case where it was transmitting all the vibrations. Easy solved moving and securing the cable.

As for the memory - it's also about the interaction with the MoBo and CPU. I have tried with the same memory the following configs:
- 8GB, E8400, CL4 2V - OK
- 4GB, Q9550, CL4 2V - OK
- 8GB, Q9550, CL4 2V - FAIL (1 thread/core, after 33 mins in Prime95)
- 8GB, Q9550, CL5 1.8V - OK

I will go for the last config for now. I could try more, see if increading the voltage helps, etc. We'll see.

Anyway, my build is done now. THANK YOU VERY MUCH for all the help! SilentPCReview comes to the rescue once again...

Cata

Nick Geraedts · Post by **Nick Geraedts** » Thu Dec 18, 2008 2:04 am

I'd say that it depends on your point-of-view. I'd personally call the specifications written on the DIMM the design specs, regardless of what JEDEC says it should be.

ATX is another specification, but look at all the ATX 2.0+ power supplies that SPCR has reviewed in the past year - do any of them actually have fully independent 12V lines as the documentation says they should? No. Should the PSU be able to provide the power that they claim on their sticker? Yes . In my opinion, the true maximum power output should actually be slightly higher than what they claim (factor-of-safety from the engineer in me). I think you'd be hard pressed to find a 300W+ PSU today that has fully independent 12V rails.

Specifications are useful to help set boundaries, but they're rarely followed to the letter.

EDIT - Glad to hear that things are working. Sorry that we kinda took this offtopic....

protellect · Post by **protellect** » Thu Dec 18, 2008 9:39 am

Great to hear its working =) Sorry about off topic, haha.

All jokes, all jokes.

Nick Geraedts · Post by **Nick Geraedts** » Thu Dec 18, 2008 1:30 pm

Hahaha!!! I applaud you, sir, on your excellent use of XKCD.