Solved 8-Core, overclocked machine slow...

[JoesMorgue]

I would test RAM stick individually, and run the test overnight, and if an error is found with the RAM stick, place the same RAM stick in a different motherboard slot (repeat for all the slots), if an error is found in all the slots for the same RAM stick then you have a faulty RAM module.

However, if no errors are found for the RAM module in a different slot, then you have a bad motherboard DIMM slot. Remember, to make sure the contacts on the RAM modules are clean and the motherboard slots are not covered or full of dust.

Run the test for least 6-7 passes.

I started testing before I read your post. I pulled two, moved my SSD, and added another HD to my system for VMs. Ran the test, and almost INSTANTLY, I got an error. Shutdown, swapped out the first stick and running again. 49% thru the test without errors. I'm guessing it's safe to say it's the stick no the slot...

 

[x BlueRobot]

Keep testing in all the slots, the tests will take time, but least you will sure what RAM stick or DIMM slot is causing the problem.

 

[JoesMorgue]

also note that most java and flash around are mostly single or two-threaded programs. That means all cores beyond the 4th are there just to keep time because there are only two "streams" of instructions to process. Having more cores means that each core taken alone is weaker than a processor with less cores but still running at the same frequency. Given modern needs, it's not a so good idea, as lots of programs (and games) aren't going to multithread over 4 cores, go figure over 8. So you have half of your processing power that is basically wasted.

AMD stuff shines when you are doing heavily multithreaded things (basically workstation or server stuff), but for gaming and personal use the best one is around the i5 3750k, performance-wise.

Also, now that you mention overclocking, AMD cores do a trick when reaching dangerous temp, and this could cause the CPU to run like a crappy i3 as overheated cores downclock to 1 Ghz or so. Your processor is a Bulldozer series, which is known for heavy power use (and high thermal output).

Check if your cores are cooled well enough to keep their stated max frequency even under real loads with programs like CPUz. This page tells more details, although it's for a slightly more modern processor of the same family.
It could be likely that you need to decrease your overclock or get better cooling to keep them from auto-downclocking.

Right now, with the memory issues I'm having now, I have the overclocking turned off, but I am slowly proceeding with a new project, and I have taken into consideration the fact that 1/2 of my cores are quite unused.

Once I positiviely identify the defective stick [I think I have] I will work on getting a replacement,but will move that stick to the very back to limit the number of crashes since my replacement will travel by mail.

I am going to install a VM of XP [because I have one I can use] and give it it's own keyboard and mouse to see if I can use it while someone else uses the rest of the Windows 7 machine. I am undecided on 1 or 2 cores for the VM, but I'm going to lock it into Core 8, and maybe core 7.

 

[JoesMorgue]

Keep testing in all the slots, the tests will take time, but least you will sure what RAM stick or DIMM slot is causing the problem.

Yes sir! [Bows, then salutes!]

 

[JoesMorgue]

I started testing before I read your post. I pulled two, moved my SSD, and added another HD to my system for VMs. Ran the test, and almost INSTANTLY, I got an error. Shutdown, swapped out the first stick and running again. 49% thru the test without errors. I'm guessing it's safe to say it's the stick no the slot...

The first stick is now in the back. [All in reference to the front of the case.] I am re-running Memtest86+ and the error came up almost instantly. When I swapped out the one stick, it went thru the entire test without errors.

Now what?

 

[Britton30]

Run each memtest for 7 passes minimum, unless there is any errors, you can stop if so. I suspect the one RAM module is bad as you do. You might check it alone in each slot of the board to confirm.

 

[x BlueRobot]

Test that RAM module in every slot, it does seem that one RAM module may be issue, if less RAM is being displayed by Windows.

 

[JoesMorgue]

Ok. Updates! I found something!

I tested each stick by itself. Three of the sticks said 400 MHz DDR-800, while the forth stick said 667 MHz, DDR-1334.

I think that is the issue...

 

[bobafetthotmail]

Once I positiviely identify the defective stick [I think I have] I will work on getting a replacement,but will move that stick to the very back to limit the number of crashes since my replacement will travel by mail.

If you have 16 GB in 4x4GB banks you can safely live without 4GB for even a month. Unless you do stuff that specifically demands it, you won't be able to fill 8 GB of ram.

The first stick is now in the back. [All in reference to the front of the case.] I am re-running Memtest86+ and the error came up almost instantly. When I swapped out the one stick, it went thru the entire test without errors.

The position of the memory stick is irrelevant unless you were testing the slots themselves, as all ram is going to be used regardless of the slot.
Ram tests are usually done on a single ram bank, just to make sure that if there are errors they 100% come from that bank.

400 MHz DDR-800, while the forth stick said 667 MHz, DDR-1334

400 mhz ram? Isn't it a bit outdated for a Bulldozer processor? :sarc:

And yes, mixing rams working at different speeds could result in issues. if the stick giving errors was the 667Mz one, try removing all the others and placing it in the first slot, then repeat the test.
If it comes out clean it was one of such "mixing ram issues", not a defective stick.

I am undecided on 1 or 2 cores for the VM, but I'm going to lock it into Core 8, and maybe core 7.

I'd give it more, 3 or 4 cores if it can use them. Remember that each core is around half the power of a good quad-core's core.

 

[JoesMorgue]

Once I positiviely identify the defective stick [I think I have] I will work on getting a replacement,but will move that stick to the very back to limit the number of crashes since my replacement will travel by mail.

If you have 16 GB in 4x4GB banks you can safely live without 4GB for even a month. Unless you do stuff that specifically demands it, you won't be able to fill 8 GB of ram.

Well, the OS is x64, and I think the MB is too. Would that cause troubles because of the missing stick?

The first stick is now in the back. [All in reference to the front of the case.] I am re-running Memtest86+ and the error came up almost instantly. When I swapped out the one stick, it went thru the entire test without errors.

The position of the memory stick is irrelevant unless you were testing the slots themselves, as all ram is going to be used regardless of the slot.
Ram tests are usually done on a single ram bank, just to make sure that if there are errors they 100% come from that bank.

400 MHz DDR-800, while the forth stick said 667 MHz, DDR-1334

400 mhz ram? Isn't it a bit outdated for a Bulldozer processor? :sarc:

And yes, mixing rams working at different speeds could result in issues. if the stick giving errors was the 667Mz one, try removing all the others and placing it in the first slot, then repeat the test.
If it comes out clean it was one of such "mixing ram issues", not a defective stick.

The sticks I'm using are labeled as: 1866MHz, Corsair said the DRAM speed should be set at 1600 for my CPU. I bought them as a set of four from Amazon. I do not understand why it's showing at 400, or even 667.

I am undecided on 1 or 2 cores for the VM, but I'm going to lock it into Core 8, and maybe core 7.

I'd give it more, 3 or 4 cores if it can use them. Remember that each core is around half the power of a good quad-core's core.

It appears that XP Pro is only seeing a single processor. I'm want to wait until I get all the updates, but the Windows [XP] update part of Microsoft is having errors, so I think I'm stuck on SP1 for now...

 

[bobafetthotmail]

Well, the OS is x64, and I think the MB is too. Would that cause troubles because of the missing stick?

Nope. The mobo doesn't care of OS architecture (the processor does, but even my prehistoric one can support x64 OS).
And the OS has only a lower and max ram limit ( x64 Win7 needs around 2 GB to run, but makes no sense to install it if you have less than 3 GB anyway)

There is something called "dual channel" memory usage, which generally does favor having a even number of ram sticks from the same manufacturer as it basically puts the rams in a RAID.

Check your mobo manual about this. I've seen some that manage to run dual-channel with paired rams AND single-channel with the odd lone ram stick.

Won't make a huge performance difference, around 5% in running very memory intensive tasks.

The sticks I'm using are labeled as: 1866MHz, Corsair said the DRAM speed should be set at 1600 for my CPU. I bought them as a set of four from Amazon. I do not understand why it's showing at 400, or even 667.

Is that a reading from the BIOS? Try going in the bios and set them manually at 1600 or if it doesn't work set them to 1333, also check their required voltage (manufacturer's site) and adjust it from the bios as needed, sometimes they require more voltage than what the BIOS thinks.

It appears that XP Pro is only seeing a single processor. I'm want to wait until I get all the updates, but the Windows [XP] update part of Microsoft is having errors, so I think I'm stuck on SP1 for now...

What the virtual machine sees isn't an indication, see by task manager of the host machine if the cores you assigned are working. I know that issue with updates, it is murderously annoying. Google around and you'll find plenty of possible solutions (it can be caused by various different things). The best solution is getting a XP disk with SP2 already on it.
I tend to like VirtualBox as the VM program.

 

[JoesMorgue]

Well, the OS is x64, and I think the MB is too. Would that cause troubles because of the missing stick?

Nope. The mobo doesn't care of OS architecture (the processor does, but even my prehistoric one can support x64 OS).
And the OS has only a lower and max ram limit ( x64 Win7 needs around 2 GB to run, but makes no sense to install it if you have less than 3 GB anyway)

There is something called "dual channel" memory usage, which generally does favor having a even number of ram sticks from the same manufacturer as it basically puts the rams in a RAID.

Check your mobo manual about this. I've seen some that manage to run dual-channel with paired rams AND single-channel with the odd lone ram stick.

Won't make a huge performance difference, around 5% in running very memory intensive tasks.

The sticks I'm using are labeled as: 1866MHz, Corsair said the DRAM speed should be set at 1600 for my CPU. I bought them as a set of four from Amazon. I do not understand why it's showing at 400, or even 667.

Is that a reading from the BIOS? Try going in the bios and set them manually at 1600 or if it doesn't work set them to 1333, also check their required voltage (manufacturer's site) and adjust it from the bios as needed, sometimes they require more voltage than what the BIOS thinks.

It appears that XP Pro is only seeing a single processor. I'm want to wait until I get all the updates, but the Windows [XP] update part of Microsoft is having errors, so I think I'm stuck on SP1 for now...

What the virtual machine sees isn't an indication, see by task manager of the host machine if the cores you assigned are working. I know that issue with updates, it is murderously annoying. Google around and you'll find plenty of possible solutions (it can be caused by various different things). The best solution is getting a XP disk with SP2 already on it.
I tend to like VirtualBox as the VM program.

Feeling lazy tonight, so I'm not seperating the quotes:

The motherboard is Dual Channel, which is why I thought I had to do pairs. The motherboard does support single channel operations. I will pull the last stick, and see what happens.

I set the DRAM manually to 1600 per Corsair before testing each stick by itself. As far as required voltage, I will go and check the site, but considering 3 of them tested fine, I feel confident that the board detects the fine.

I removed the assignments on the VM, during the install, Core 7 was pegged at 100% and the thing was like installing XP on a low end 386. The little bit of time I've tried running XP, it BARELY put any load on my [real] CPU. It told me I had a 7% processor load with both running, but doing very little. Between the RamDrive, Win7, and Windows XP, I was using 55% of my 16 Gigs of Ram.

I'm running VirtualBox on Win7 and my MacBook and both have an update that can be installed, just havn't done it yet. My problems with the VM will be posted in a new thread...

 

[JoesMorgue]

Update:

I called Corsair [again] and they wanted to check things first. We checked my settings, and he said 1.49 was a little low for voltage with 4 sticks, and to raise it up to 1.65 [which I did] and the rest of the settings were OK. I ran MemTest past 10% without errors [unlike before] so I told him I would run it overnight, after using the computer the rest of the day. He said if I get a BSOD, or it fails the test [and he wanted at least 3 passes] they would replace it.

We left, leaving the computer on. While out, I had a BSOD. I ran the test while I tried to sleep, and it appeared to be the third time through before it started finding errors.

I called Corsair [yet again] and they wanted to try more things. I said I just wanted to advance exchange it. The new memory is on it's way...

First, I want to thank x BlueRobot, and Bobafetthotmail [I added to your reps] for their time/knowledge.

Second, do I mark this as solved? I started this thread on a performance issue, and we ended up solving a memory/BSOD issue. I don't KNOW if they are related/solved.

 

[Dave76]

EDIT: Don't mark it solved until you get a stable system.

Put two matched RAM cards in the blue slots and the other card in the black slot. This is the recommendation from your motherboard manual.
When you're running memtest86+ and you see a % of a test, that is one of several tests it runs in a single pass. Each pass runs different tests and it will run indefinitely. The number of passes is in the statistics row, which is just above the 'Red' errors (if you have errors).

Can you download CPUZ and post a snip of the CPU, Mainboard, Memory, and SPD tabs.
For posting the information , enter 'Snipping tool' in the Start button > Search box, hit enter. Select the area you want to post and save to a convenient place.
How to Post a Screenshot in Seven Forums
Go to your BIOS and tell us what the CPU, RAM and the CPU integrated memory controller voltages (Should be CPU/NB voltage for your motherboard) are.
Usually you need to bump the IMC and the DRAM for 16GB RAM.

EDIT:
Just saw your last post.
You should let me know what the IMC voltage is set at, I have adjusted a lot of RAM with good results, sometimes they will become stable with a IMC and DRAM voltage bump, but usually not with only one or the other.

Are you still running the RAMDrive?

What SSD do you have?
You really don't have to worry about writes to a SSD. The rumors about them being fragile are grossly exaggerated.
Check this thread, these guys are running non-stop writes to several SSDs to see when they fail.
SSD Write Endurance 25nm Vs 34nm
The graphs on the first page are out of date and only cover some of the first SSDs that started the tesing.
The average desktop user writes between 7-10 GB worth of information per day, check the total writes these guys are getting.
For example the Samsung 830 256GB.

Samsung 830 256GB Day 237

(GiB) 5,887,996
(TiB) 5,749
(PiB) 5.65

These are the total writes so far, and it's still going.

5,887,996 GiB = 6,361,334 GB
5,749 TiB = 6,362 TB
5.65 PiB = 6.36 PB

Even if you figure 100GB per day it's 174 years.

 

[JoesMorgue]

EDIT: Don't mark it solved until you get a stable system.

Put two matched RAM cards in the blue slots and the other card in the black slot. This is the recommendation from your motherboard manual.
When you're running memtest86+ and you see a % of a test, that is one of several tests it runs in a single pass. Each pass runs different tests and it will run indefinitely. The number of passes is in the statistics row, which is just above the 'Red' errors (if you have errors).

I moved the stick back to a black slot. The rest I knew/understand. I stopped the test after 10% because of the time it takes, and my kids were coming home from school, and one of them would want to be on the computer. It would be better for this household to run the test all night.

Note: INTERESTING... Before writing this response, I shutdown a large [LONG] operation, then started over after moving the stick back to it's origional location [Farthest from the CPU] and the time remaining on the operation is MUCH lower than before [by AT LEAST 30%, AND it's running over my network. [I think it will make a 100GIG file by the time it's done...]

Can you download CPUZ and post a snip of the CPU, Mainboard, Memory, and SPD tabs.
For posting the information , enter 'Snipping tool' in the Start button > Search box, hit enter. Select the area you want to post and save to a convenient place.
How to Post a Screenshot in Seven Forums
Go to your BIOS and tell us what the CPU, RAM and the CPU integrated memory controller voltages (Should be CPU/NB voltage for your motherboard) are.
Usually you need to bump the IMC and the DRAM for 16GB RAM.

EDIT:
Just saw your last post.
You should let me know what the IMC voltage is set at, I have adjusted a lot of RAM with good results, sometimes they will become stable with a IMC and DRAM voltage bump, but usually not with only one or the other.

Are you still running the RAMDrive?

Yes. I do wonder if I am getting a performance boost from it...[and yes, I know it will be unseen if it is, but I can afford the memory]

What SSD do you have?
You really don't have to worry about writes to a SSD. The rumors about them being fragile are grossly exaggerated.
Check this thread, these guys are running non-stop writes to several SSDs to see when they fail.
SSD Write Endurance 25nm Vs 34nm
The graphs on the first page are out of date and only cover some of the first SSDs that started the tesing.
The average desktop user writes between 7-10 GB worth of information per day, check the total writes these guys are getting.
For example the Samsung 830 256GB.

Samsung 830 256GB Day 237

(GiB) 5,887,996
(TiB) 5,749
(PiB) 5.65

These are the total writes so far, and it's still going.

5,887,996 GiB = 6,361,334 GB
5,749 TiB = 6,362 TB
5.65 PiB = 6.36 PB

Even if you figure 100GB per day it's 174 years.

This is the drive I bought from Amazon: Amazon.com: OCZ Vertex 2 120 GB SATA II 2.5" Solid State Drive (OCZSSD2-2VTXE120G): Electronics

I admits it...I messed up and didn't look at which SATA ver it was. I SHOULDA gotten a FASTER drive...

 

[Dave76]

Sometimes just re-seating the RAM cards will solve problems.
With 16GB RAM you will probably need to bump the IMC (Integrated Memory Controller, in your BIOS it's CPU/NB Voltage) and DRAM voltage.
For the IMC use the 'CPU/NB Offset Voltage' this will allow any speed control functions to still work. The IMC needs more voltage when running 4 RAM cards or more than 4-8GB RAM.

Change the CPU & NB Voltage mode to ‘Offset’.
CPU/NB Offset Voltage to 1.20v, select the offset voltage of +0.100v if the current setting is 1.100v.
DRAM Voltage change to 1.550v.
Save & Exit BIOS.
These are conservative increases, but should be stable, if not we can bump them a little more.

Test the setting by running memtest86+ for 7 full passes, best to run overnight.
If you get any errors you can stop the test and let me know.

Update your BIOS to version 1501, there are two update since 1202 both are for 'Improve system stability'.

I wouldn't bother with the RAMDrive, it was intended to speed up HDDs, your SSD should be fast enough on it's own.
Yes, a SATA 6Gb/s SSD would be better but the 3Gb/s should be fast.
If/when you upgrade, my personal preference, avoid SSDs with the SandForce controller. Too many problems.
Check at the OCZ site for a firmware update, install if not current, all the info you need is on their site.

 

[JoesMorgue]

Sometimes just re-seating the RAM cards will solve problems.
With 16GB RAM you will probably need to bump the IMC (Integrated Memory Controller, in your BIOS it's CPU/NB Voltage) and DRAM voltage.
For the IMC use the 'CPU/NB Offset Voltage' this will allow any speed control functions to still work. The IMC needs more voltage when running 4 RAM cards or more than 4-8GB RAM.

Change the CPU & NB Voltage mode to ‘Offset’.
CPU/NB Offset Voltage to 1.20v, select the offset voltage of +0.100v if the current setting is 1.100v.
DRAM Voltage change to 1.550v.
Save & Exit BIOS.
These are conservative increases, but should be stable, if not we can bump them a little more.

Test the setting by running memtest86+ for 7 full passes, best to run overnight.
If you get any errors you can stop the test and let me know.

Update your BIOS to version 1501, there are two update since 1202 both are for 'Improve system stability'.

I wouldn't bother with the RAMDrive, it was intended to speed up HDDs, your SSD should be fast enough on it's own.
Yes, a SATA 6Gb/s SSD would be better but the 3Gb/s should be fast.
If/when you upgrade, my personal preference, avoid SSDs with the SandForce controller. Too many problems.
Check at the OCZ site for a firmware update, install if not current, all the info you need is on their site.

I have memory being shipped to replace what I have. As long as the machine don't crash every 10 minutes [Been running for 30 hours straight now] I'm OK.

Then again, making changes to my machine now would be a good time to learn. <Evil Grin> Currently, my DRAM Voltage is 1.65 [per Corsair]

Can you elaborate on the point of the CPU/NB offset voltage? [Start with what is NB?]

I JUST tried to update to the newer BIOS, this is the second one that failed to install. Monday, I am going to call Asus...

 

[Dave76]

CPU/NB is the Integrated Memory Controller (IMC), part of the CPU, it controls the RAM.
Previously the Memory Controller was on the North Bridge (NB).
When you have a lot of RAM installed it is normal to bump the IMC voltage to help the RAM run properly.
This is more important than bumping the DRAM voltage in most cases, surprised Corsair didn't suggest that, they usually do.

Sometimes just re-seating the RAM cards will solve problems.
With 16GB RAM you will probably need to bump the IMC (Integrated Memory Controller, in your BIOS it's CPU/NB Voltage) and DRAM voltage.

Memory controller

Northbridge (computing)


Your DRAM voltage is probably higher than is necessary, bumping the IMC voltage will let you lower the DRAM voltage.

 

[JoesMorgue]

CPU/NB is the Integrated Memory Controller (IMC), part of the CPU, it controls the RAM.
Previously the Memory Controller was on the North Bridge (NB).
When you have a lot of RAM installed it is normal to bump the IMC voltage to help the RAM run properly.
This is more important than bumping the DRAM voltage in most cases, surprised Corsair didn't suggest that, they usually do.

Sometimes just re-seating the RAM cards will solve problems.
With 16GB RAM you will probably need to bump the IMC (Integrated Memory Controller, in your BIOS it's CPU/NB Voltage) and DRAM voltage.

Memory controller

Northbridge (computing)


Your DRAM voltage is probably higher than is necessary, bumping the IMC voltage will let you lower the DRAM voltage.

OK, I looked and the current CPU offset is at 1.262 Volts, and the CPU/NB is at 1.250. Both auto-set by the motherboard.

 

[Dave76]

Bump the CPU/NB offset to 1.300v.
If it's still not stable, bump the CPU/NB offset to 1.350v.

When you get it stable, lower the DRAM voltage to 1.60v, check for stability. If stable lower the DRAM voltage to 1.550v check for stability.
When it is not stable bump the DRAM voltage up by +0.020v until stable.
Don't go below 1.500v.

A good quick stability test is OCCT.
OCCT stability checking tool
Run the CPU:OCCT test for at least 30 minutes, be sure to monitor your CPU and GPU temps.
Set the time before you start.