The *Official* MSI K8NGM2-FID GeForce 6150 Motherboard Thread

[Artanis]

Quality 2x512 should be enough. You'll need 2GB in the future, just sell your memory and buy a 2x1GB kit.

 

[renethx]

Originally posted by: BernardP
I am asking this because I want to go with 1 GB in my new system for now. Is it better to go for 2x512 Gb (dual channel) and suffer the memory frequency penalty *IF* I upgrade in the future.....OR.....Go with a single 1Gb stick now (and suffer the no-dual-channel penalty) AND get the dual-channel performance gain at full DDR 400 *ONLY IF* I upgrade to 2 Gb in the future?

Performance depends on applications. The below is the tests I did last year (Abit AX8 Motherboard, Its nice.):

Iran SuperPI 8M with FSB 311 MHz and CPU multiplier 9 (= 2800MHz) in various settings of memory frequency and timing to determine the best combination of freq and timing, because with high freq (DDR266) + low latency (2.5-3-3-8-1T) the system was not stable in the previous test. Curiously this time the system became stable in this combination. (Maybe memory has been burning in?) So the test became meaningless (use this combination!) However, as I have done the test, I will give the results just for reference anyway.
BIOS: 1.1
OS: Windows 2000 Professional (roughly x 1.25 slower than Windows XP in SuperPI)
VCore: 1.5500, VDDR: 2.60, VNB: 1.65, VSB: 2.65, VHT: 1.35, VDDRRef: 0mv

CPU Multiplier 10

FSB = 280 MHz, CPU = FSB x 10 = 2800 MHz, LDT = FSB x 4 = 1120 MHz

DDR266 = CPU/15 = 186.7 MHz, 2.5-3-3-8-1T: 06m 35s

CPU Multiplier 9

FSB = 311 MHz, CPU = FSB x 9 = 2800 MHz, LDT = FSB x 3 = 933 MHz

DDR200 = CPU/18 = 155.6 MHz, 2.5-3-3-8-1T: 07m 02s
DDR200 = CPU/18 = 155.6 MHz, 2.5-3-3-8-2T: 07m 19s
DDR200 = CPU/18 = 155.6 MHz, 3-4-4-9-1T: 07m 14s
DDR200 = CPU/18 = 155.6 MHz, 3-4-4-9-2T: 07m 24s

DDR266 = CPU/14 = 200.0 MHz, 2.5-3-3-8-1T: 06m 32s
DDR266 = CPU/14 = 200.0 MHz, 2.5-3-3-8-2T: 06m 47s
DDR266 = CPU/14 = 200.0 MHz, 3-4-4-9-1T: 06m 40s
DDR266 = CPU/14 = 200.0 MHz, 3-4-4-9-2T: 07m 05s

SuperPI is heavily dependent on mem bandwidth.

I am not sure why you need 2GB memory. Are you gaming, or do you use Photoshop? Otherwise you won't need 2GB. Buying a stick of 1GB and buy another later is not recommended because even the same model number does not guarantee the same memory chip and dual channel may not work.

 

[renethx]

Yeah, Artanis' idea is much better than using four sticks.

 

[pctwo]

Originally posted by: josh2e

I have the MSI K8NGM2-FID running 3.1 w/ MIT's MyHD card for capturing OTA HD. It seems to be working fine but unfortunately I can't seem to get nVidia to allow me to output over the supplied component output bracket. when i go into the video properties settings (xp pro) - there isn't any sign of a 2nd monitor (or TV). when i go into advanced and go to the GForce 6150 tab, i have checked the troubleshooting box that says, "if you have a tv connected and it is not being deteced, select the checkbox below" - Force TV Detection. i am currently only getting a signal to my monitor over RGB. i haven't tried hooking up my TV w/ RGB eventho it does have that option.

does anyone have any suggestions?

don't have an answer for you but look at this thread

http://www.avsforum.com/avs-vb/showthread.php?t=601451&page=1&pp=30

if anyone knows, it'll be the HTPC folks over there.

 

[BernardP]

Originally posted by: renethx
I am not sure why you need 2GB memory. Are you gaming, or do you use Photoshop? Otherwise you won't need 2GB. Buying a stick of 1GB and buy another later is not recommended because even the same model number does not guarantee the same memory chip and dual channel may not work.

Thanks. I am a casual gamer. Will eventually have a GeForce 7600 series card in my system. No Photoshop. I mostly work on audio files. I want to keep the option of 2Gb open "just in case". From your test results, I see that being forced to go from DDR400 to DDR333 if I had to eventually get 2 Gb memory would not be a catastrophe.

No 1x1Gb for me then.

There are currently 42 processes (including SMS crap) running on my system at work(P4 2.66 with 512 Mb memory) and it is going smoothly. So a more powerful "recreational" computer with 1 Gb memory should not fare too badly.

 

[imported_derekn]

renethx,

Thanks for compiling the valuable information about the MSI K8NGM2-FID. You saved me a lot of time.

I haven't built a PC since 2004 when I overclocked my Athlon XP 2500 (Barton) on my Abit Nf7 board to an Athlon XP 3200. I was able to order the MSI K8NGM2-FID from Amazon (1 left in stock) and the rest of the parts from Newegg.com. (AMD X2 3800, Corsair 1 GB Corsair Value Select, etc).

Feels weird to branch away from Asus, Abit, Chaintech, ECS, Tyan.

I have had the best luck with Asus (a7n8x deluxe, a7v8x-x) and Abit (2 Nf7 v2).

ECS (2 K7s5a), Chaintech (7VJL6) and Tyan (s2390b) were awful experiences.

Let's see what happens. It's great to see user experiences with the MSI.

-Derek

 

[renethx]

New Contents Added

Choosing Components > Memory > 2 x 1 GB vs 4 x 512MB

 

[Raydur]

Originally posted by: Artanis
I confirm Memtest96 errors with 3.2 BIOS, in spite of running fine Prime95 and SuperPI, and having no BSOD. It's sad because the temperatures were corrected in this version, but for safety I flashed back to the unofficial 3.13 version.

Can you elaborate on "the temperatures were corrected"?

Thanks

 

[Ivan05]

Originally posted by: bigfootindy2k6
Kind of a bump from my previous post - can anyone confirm that putting 4 dimms in definately decreases the speed from DDR400 to DDR333? Is this a "feature" of the chipset or just the board? Is there a way around it?

Everybody's hinted at this, but I seen haven't this directly adressed in this thread...

It is a limitation based on the onboard memory controller of the A64 architecture. 2 sticks will run at DDR400, 4 sticks must run at DDR333. There is no way around it, other than OC'ing to make up for the reduced speed.

See this thread.

And this White Paper at AMD.

 

[renethx]

Ivan05, thanks for information. I added the reference to the main post.

 

[Artanis]

Originally posted by: Raydur

Originally posted by: Artanis
I confirm Memtest96 errors with 3.2 BIOS, in spite of running fine Prime95 and SuperPI, and having no BSOD. It's sad because the temperatures were corrected in this version, but for safety I flashed back to the unofficial 3.13 version.

Can you elaborate on "the temperatures were corrected"?

Thanks

Reported temperatures were closer to the reality (and a bit lower than before).

Originally posted by: Ivan05
Everybody's hinted at this, but I seen haven't this directly adressed in this thread...

It is a limitation based on the onboard memory controller of the A64 architecture. 2 sticks will run at DDR400, 4 sticks must run at DDR333. There is no way around it, other than OC'ing to make up for the reduced speed.

See this thread.

And this White Paper at AMD.

But DFI somehow managed to get 4 DIMMs in 1T with LANParty UT RDX200 mainboard. More here: http://www.anandtech.com/mb/showdoc.aspx?i=2572&p=5

"However, what is truly unique is that the DFI LANParty UT RDX200 is the first board designed to run four double-sided 512MB or 1GB DIMMs at the fastest timings and a 1T Command Rate. In fact, AMD specifications tell us that this is impossible. To his credit, Oskar Wu and DFI found a way to accomplish the impossible."

 

[bigfootindy2k6]

Thanks for the replies - seems like if I overclock it would be possible to effectively get near DDR400. More than likely, I'll have to get 2x1GB dimms.

 

[Ivan05]

Originally posted by: bigfootindy2k6
Thanks for the replies - seems like if I overclock it would be possible to effectively get near DDR400. More than likely, I'll have to get 2x1GB dimms.

I was in the same situation before I built my K8NGM2-FID box. Unless you're running XP 64 bit AND have an app that will benefit from the extra memory, its really not worth the extra expense.

Remember, XP (32 bit) won't address RAM above, IIRC, 3.5 GB.

I ended up going with XP Pro and 2GB based on the fact that driver support, IMHO, for 64 bit is still not what it should be and the fact that the only app that I run, Photoshop CS, would have required an upgrade that I wasn't prepared to spring for. Needless to say, I went with 2GB.

 

[mikeaco]

Hi renethx,

Thanks for compiling all of this information. It was really helpful in helping me make my final decision on which components to get as well as making sure I wasn't forgetting anything while setting up my first system. I was pretty happy to see that, with very little effort, a modest overclock would be obtainable.

The parts I chose, along with the MSI mainboard, included the 3200+ Venice E6 (I actually ordered the ADA3200BPBOX (E3) from newegg, but received the E6 instead) and 2 x 512MB Corsair VS DDR400 in dual-channel mode.

After getting everything installed and the OS loaded (Win XP Pro), I started playing around with overclocking to see what I could get. I started out with your "quick recipe" settings and the v3.0 BIOS. Eventially, I ended up flashing to v3.13.

My results (and it didn't seem to matter whether or not it was v3.0 or v3.13 BIOS) are as follows:

CPU-LDT Frequency=250MHz, CPU=2.5GHz, Mem=208.33MHz
- System would POST, but got a quick BSOD after Windows XP initial splash screen. The system would then reboot.

CPU-LDT Frequency=240MHz, CPU=2.4GHz, Mem=200.00MHz
- System would POST and boot into Windows XP.
- Ran SuperPI (8M) and got an "Error (NOT EXACT IN ROUND)"

CPU-LDT Frequency=230MHz, CPU=2.3GHz, Mem=191.67MHz
- System would POST and boot into Windows XP.
- Ran SuperPI (32M) with a PASSED result.
- Ran Prime95, Torture Test, Small FFT with a FAILED result.

I haven't tried any other (slower) clock speeds since I ran out of time last night. Plus, I was a bit discouraged that I am not able to overclock as easily as you make it sound So, I wanted to here first to see if you or anyone had any ideas first.

Could any of the failures be attributed to bad RAM? This morning, I ran Memtest86 and found that several tests failed. All failures were because of the same bad bit (bit 4). I tried swapping the DIMM's, but the failure remained on bit 4. Could this be an indication that there is a DIMM slot, memory bus or other mainboard hardware problem? I have the DDR Voltage set to the default 2.55V. However, I've seen somewhere that the normal voltage for this memory should be 2.5V. Would this be causing the problem? Would reducing the Memclock Value to 133MHz remove the possiblity that the memory is the problem?

Anyway, any suggestions would be greatly appreciated.

Thanks again!
Mike

 

[Artanis]

Increase the VDIMM to at least 2.65V, and make sure you have 3.13 BIOS version and CPU to NB Frequency at 800 MHz. Your succes in overclocking the CPU depends mainly on luck

 

[renethx]

mikeaco, first I recommend to follow Zebo's articel strictly to identify the problem.

ISOLATE

1. Max HTT/FSB (= "CPU to NB Frequency" and "NB to SB Frequency")...
2. Max Chip capable...
3. Memory MAX...

CONSOLIDATE

...

Boosting CPU Voltage by 0.05V by the method I wrote or raising DDR Voltage (2.70V is safe for value RAM) is a good idea.

If CPU turned out to be the problem, there are two possible reasons.

1. You drew a bad core.
2. Many say that E6 < E3 for overclocking in AT Forums > CPU/PROCESSORS AND OVERCLOCKING. I don't know if it's true.

 

[mikeaco]

Thanks guys. I will try that out later tonight. I do still need to read those overclocking how-to's that are linked on the first page of this thread. I guess I was hoping that I would be able to use the "quick recipe" solution and get "lucky". I guess it will be a little more work than that.

One question on the memory failures in Memtest86...Could the failures be attributed to improperly set timings. When everything is set to "Auto" in the Memory configuration section of the BIOS, CPU-Z reports the timings as 2.5-3-3-7 on one DIMM and 2.5-3-3-8 on the other. Would manually relaxing the CAS to 3 clocks help? Or, is the problem more likely to be fixed by up'ing the DDR voltage, as you guys suggested?

Thanks again..I'll keep you guys posted!

 

[mikeaco]

Oh...and I'll probably try to get better overclocking results by boosting the CPU Voltage. However, long term I don't think that I'll leave it in that configuration. I want to be able to have Cool'n'Quiet running to reduce noise from the fan. As you say, boosting the voltage disables this feature.

 

[Artanis]

Get the VDIMM to 2.65 or 2.7V first.

 

[pctwo]

Originally posted by: mikeaco
Oh...and I'll probably try to get better overclocking results by boosting the CPU Voltage. However, long term I don't think that I'll leave it in that configuration. I want to be able to have Cool'n'Quiet running to reduce noise from the fan. As you say, boosting the voltage disables this feature.

you don't need CnQ to reduce cpu fan noise. just use the board's smart fan control feature. it's in the H/W Monitor BIOS menu

 

[renethx]

Yeah, agree with Artanis. Memtest failure may indicate not enough voltage to mem. Raising DDR Voltage may (or may not) fix the problem instantly.

 

[cupajava]

I have that ram there is no way it will scale as high as 2.5ghz without at least 2.70 vdimm. Any higher didn't seem to help. Vcore will also likely need to raised 1.45-1.50.

 

[mikeaco]

Unfortunately, according to renethx's guide on page 1, the default Vcore of 1.35V can only be increased 0.05V, to 1.40V, in an indirect way within the BIOS for this board.

I'm not sure what you mean by the RAM scaling as high as 2.5 GHz. The processor runs at 2.5 GHz and then you use the divider technique to run the RAM at 208.33MHz. At least that's what I would do if I could

I'll try the DDR Voltage Adjust and hope that helps!

 

[Olive Yew]

Mikeaco, Renethx meant well with his mini overclocking guides. But in those guides he makes several rather large assumptions about the components in your system, which in my experience can never be taken for granted.

The 1st step for any new system build should *always* be to test the memory. I can't stress that enough. It is very bad practice to start stress testing other components in your system without first determining if your memory is functioning without errors. If you jump ahead to test the CPU or the limits of your HTT, you'll never know if any errors that pop up are due to your memory or the limitations of your other components. Even if you've tested the memory you intend to use previously on another system, you still need to test it again. No mobo in existence will work flawlessly or optimally with all available types of memory. The K8NGM2-FID is no different.

As for specific practices, here's how to get started testing memory on this particular mobo. These instructions are with regard to the 3.13beta BIOS:




Before you get started, head over to http://www.memtest.org/ and download the latest copy of Memtest 86+. Burn/Write yourself a bootable floppy or CD before you proceed.

1. Under the "Advanced Chipset Features" menu, make the following changes
a - MCT Timing Mode -> Manual
b - CAS Latency (CL) -> As per spec for your RAM
c - TRAS -> As per spec for your RAM
d - TRP -> As per spec for your RAM
e - TRCD -> As per spec on your RAM

Most RAM sticks are labeled with their timings such as 2.5-3-3-7 or 3-4-4-8. Those numbers are listed in order as CAS Latency-tRCD-tRP-tRAS.

f - CPU to NB Frequency -> 800mhz
g - NB to SB Frequency -> 800mhz
g - CPU to NB LinkWidth -> 16 16
h - Primary Graphics Adapter -> PCI Express if you're using an separate vid card (recommended)
i - Onchip and PCIe VGA Selection -> Disable Onchip VGA if have PCIe VGA if you're using a separate vid card (recommended)
j - Onchip VGA Frame Buffer Size -> 16MB
k - CPU LDT Frequency, Mhz -> 200mhz

Note: It's highly recommended to use a separate vid card and disable the onboard video while testing memory. As the onboard video always uses a portion of system memory, it's impossible to test 100% of the memory while it's active.


2. Under the "Cell_Menu", make the following changes:
l - Cool N Quiet -> Disabled
m - Voltage Control Function -> CPU Frequency Configuration -> FID Change -> Manual
n - Voltage Control Function -> CPU Frequency Configuration -> Processor Frequency Multiplier -> Set this value as default - 3. On a Venice 3200+ with a default 10x multiplier you would set this at 7.
o - Voltage Control Function -> Adjust DDR Voltage(V) -> As per spec for your RAM
p - Spread Spectrum -> CPU Spread Spectrum -> Disabled
q - Spread Spectrum -> PCIE Spread Spectrum -> Disabled
r - Spread Spectrum -> SATA Spread Spectrum -> Disabled

Ok, now you're ready to get started. Make sure you've set options to boot from CD or floppy as appropriate. Insert your copy of Memtest86+. Save your settings in BIOS and reboot. After posting, Memtest86+ will load automatically and start running. If by chance your system does not post, hold the insert key, reboot, and enter BIOS. Then go back to step 2o and raise VDimm incrememtally until your system does post. The k8NGM2-FID is capable of raising Vdimm up to 3.1. While nice, it is not extreme by any measures so don't worry about frying your RAM. You really need in excess of 3.4v or so to pose a major health risk to your memory.

Now back to Memtest. Let Memtest run through one complete pass. If no errors are detected, go back to step 1k and raise your LDT by 5mhz. Or if your memory is rated at better than DDR400 speeds, raise the LDT accordingly. Run Memtest again for one complete pass and go back and raise your LDT if no errors are deteced. If Memtest detects some errors, you have the option of raising VDimm in Step 2o or relaxing your timings in steps 1b->1e in order to push your RAM a little further. Whatever you do, I highly recommend you test to at least 5mhz PAST the maximum speed at which you intend to run your memory. If you are planning on raising the LDT to 250mhz and using a 166mhz divider, then you'll need to test your memory to at least 210mhz.

Once you think you've determined the maximum stable speed and configuration for your RAM, then it's time to go back to Memtest. This time, you need to let Memtest run for 32 complete passes minimum. Why 32 passes?

Test 4 [Moving inv, 32 bit pat, cached]

This is a variation of the moving inversions algorithm that shifts the data pattern left one bit for each successive address. The starting bit position is shifted left for each pass. To use all possible data patterns 32 passes are required. This test is effective in detecting data sensitive errors in "wide" memory chips.

If Memtest completes 32 passes without detecting any errors, then congratulations! Your memory has tested completely error free and you can move on to testing other components without worry of possible memory problems. However, if any errors are detected, you'll have to back down the LDT, raise Vdimm, or relax timings and test again in order to find the stable limit of your memory.


Edit - The above was written assuming you are using DDR400 memory or better and your system is capable of booting this memory at 200mhz. In this case, the LDT mhz = RAM mhz and no further consideration are required. If your memory will only boot at DDR333 or lower, then LDT mhz != RAM mhz. If you don't know how to calculate the speed of your memory based off your CPU frequency and memclock, then boot into Windows and check with an app such as CPU-Z.

 

[cupajava]

Mikeaco,

To clarify my previous response, when I meant "scale to 2.5ghz", I meant scale that high and be stable i.e. memtest86, prime95. Even though you're using a divider, more vdimm is required to support the higher FSB.

I will be acquiring this board very soon and I didn't realize vcore could only be raised an additional 0.05.