Don wrote:
Here is my question on the G.Skill forums posted yesterday. No
responses yet. A comment from Paul would be great -- he is always
helpful. Help from anyone is also appreciated. I need it.
Thanks. See below ...
Don
*****
Help from G.Skill or anyone else is appreciated. FYI – A RATHER LONG
NOTE. Proceed with caution.
Please let me know ALL of the optimal memory settings for
F3-12800CL7D-4GBRM memory, a.k.a. G.SKILL Ripjaws Series 4GB (2 x 2GB)
240-Pin DDR3 SDRAM DDR3 1600 (PC3 12800) Desktop Memory. The build
project started on or about May 1, 2010.
Here are all of the current memory related settings in BIOS for my
GA-785GMT-USB3, Version 1.0:
The BIOS version is F2. Maybe this is the culprit.
On the first screen, I select “MB Intelligent Tweaker (M.I.T)”.
The “Memory Clock” setting is set to [x8.00] by me.
The “System Voltage Control” on the same screen is set to [AUTO].
Then I select “DRAM Configuration” just below the “Memory Clock” setting.
On that next screen is the following:
DCTs Mode [Unganged]
DDR3 Timing Items [Manual]
CAS# latency [7T] (set by me)
RAS to CAS R/W Delay [8T] (set by me)
Row Precharge Time [7T] (set by me)
Minimum RAS Active Time [24T] (set by me)
1T/2T Command Timing [2T] seems O.K. or try 1T
TwTr Command Delay [6T] seems O.K. or try 5T
Trfc0 for DIMM1 [110ns] try 88ns on all
Trfc2 for DIMM2 [90ns]
Trfc1 for DIMM3 [110ns] (occupied by one stick of F3-12800CL7D-4GBRM)
Trfc3 for DIMM4 [90ns] (occupied by one stick of F3-12800CL7D-4GBRM))
Write Recovery Time [12T] seems O.K. or try 10T
Precharge Time [6T] seems O.K. or try 5T
Row Cycle Time [40T] try 33T or 26T
RAS to RAS Delay [5T] seems O.K. or try 4T
Bank Interleaving [Enabled]
Channel Interleave [Enabled]
Your Assistance:
Please comment on ALL of the settings, not just “7-8-7-24T” (please /
thanks). Are these settings O.K. ? What should I change ? The system
is going through setup pains with occasional “one long beep no boot”
issues (hence the current use of DIMM3 and DIMM4 to get the thing going
again – what a pain). Also, one episode of the “gray vertical lines
screen of death” after installing the Sapphire card (see below –
wonderful, a brand new card), but subsequently FurMark 1.8.2 has run for
1 hour on two separate occasions without incident on abusive settings
(68 degrees C max during the FurMark tesing). Memtest86 V3.5 SMP
(floppy version) shows no issues after running for 12 hours. The system
will periodically stop booting, and then I switch the memory sticks
around between DIMM1/2/3/4 to get it going again. Lots of fun.
Additional Information:
OS: WIN7 32 bit RC. Soon to expire, but I want everything working
correctly before I install the paid for version of WIN7 Professional.
Processor: AMD Athlon II X2 250 Regor 3.0GHz 2 x 1MB L2 Cache Socket AM3
65W Dual-Core Processor.
Video Card: SAPPHIRE 100287VGAL Radeon HD 5670 (Redwood) 512MB 128-bit
DDR5 PCI Express 2.0 x16 HDCP Ready CrossFireX Support Video Card. Using
VGA at the moment. HDMI later. CCC 10.4 Drivers, set at factory GPU
(775) and Memory (1000) settings, auto fan control. Never overclocked.
I tried underclocking at (725) and (950) but that lead to periodic black
screen lockups. Back to factory settings. FurMark was run on factory
settings. Great.
The on-board Video is turned off.
1 SATA hard drive.
1 SATA + 1 IDE optical drives.
1 floppy drive.
Power Supply: COOLMAX "EZ Wire" CU-400T 400W ATX Modular Power Supply.
Maybe this is the culprit.
1 D-Link DWL-G520 IEEE 802.11, 802.11b/g 32bit PCI2.2 High Speed
Wireless Adapter.
1 AVerMedia AVerTV Combo PCIe ATSC/NTSC/QAM TV Tuner Card (White Box)
w/L-P Bracket, PCI-Express x1 Interface.
That is it. If I forgot anything, let me know.
P.S.
This setup is replacing a TYAN S2865 that went up in flames, complete
with scorch marks, so maybe the case has bad mojo. I think it was the
fan-less video card (GIGABYTE GV-NX73T256P-RH GeForce 7300GT 256MB
128-bit GDDR2 PCI Express x16 SLI Support Video Card) that overheated
the motherboard, but who can say. Those components are all way out of
warranty as far as I know. I saw it happen – sparky electrical noises
for a second or two, followed by a plume of whitish grayish smoke the
size of my forearm shooting up followed by a complete shutdown of the
system followed by me going into a moment of shock. Unplug the power
supply. That was in January. Here I am today with a new dodgy system
and several hundred lighter.
Thanks for any and all assistance. I have built a number of systems
since 1998, and have never run into this type of series of issues
before. Or any issues for that matter.
Don
http://www.xbitlabs.com/articles/vid...0_6.html#sect0
HD 5670, table near the bottom, 28.7W for their "3D max" test case.
So the video card should not be flattening the power supply.
*******
http://www.gskill.com/products.php?index=243
7-8-7-24-2N DDR3-1600 (PC3 12800) Test Voltage 1.6 Volts
Personally, I don't adjust more than the stated timings, as I'm too
lazy to spend days and days, adjusting things where I don't know
what they do. Basically, I verify (using CPUZ), that the motherboard
is actually using the timings the memory is rated for, then start
testing.
The RAM doesn't need excessive voltage boost. DDR3 is 1.5V nominal.
1.6 volts is not really boosted, the way some memory products are.
At that kind of speed, I might be tempted to stay with Command Rate 2T,
as at high speed, the address bus needs additional settling time.
*******
Since you have integrated video, you could start by testing the system,
without the HD 5670 present. I would simplify the system, removing
as much other hardware as possible. The idea is, we want to be able
to point the finger at some piece of defective hardware. If the
system is still operating sub-par, with the rest of the stuff
unplugged, then that reduces the number of candidates for replacement.
The Prime95 program can be used as a combined CPU and memory stress tester.
If the system is not stable with that configuration, then you're down to
CPU, memory, or motherboard that isn't working right. Prime95 will start
a test thread per core.
http://www.mersenne.org/freesoft/
I hope you're using a new power supply. The smoke and sizzling sound,
could have been coming from bad capacitors inside the previous power supply.
Prime95 is a better test than memtest86+, in terms of deciding whether the
RAM is really good or not. A test thread will stop, on the first error
detected. Prime95 can't test the entire memory, but only the section
not used by the OS itself. Memtest86+ tests closer to all the memory,
but can't touch the BIOS reserved area. You have to move the DIMMs
around in some way, to actually get all the memory tested, with
memtest86+. You'd need single channel mode, and swap DIMM positions,
to try to move the memory around enough, for absolutely complete
memtest86+ coverage.
Once you've
1) Verified settings in CPUZ in Windows. Just to prove the
BIOS settings (auto or manual), are actually being used.
2) Run memtest86+ test cases, to verify all locations have
no stuck at faults. But for stress testing, you need other test
cases.
3) Use Prime95 and run its stress test. I run mine for up to four
hours, as proof everything is fine. You may need to extend the
test time, for larger arrays of memory. No errors are acceptable.
4) If you're failing (2) or (3), you'll need to either adjust
something, or return the memory. A bit more voltage helps, in
marginal cases (like only one error in memtest86+). Using
slack timing (bump by +1) might also help. Then, you have to
decide whether the failure is worthy of returning or not.
Bump up Tcas or Trcd.
5) There is a remote chance the processor is at fault. If memory errors
always happened at the same address, you'd know for sure it was memory.
For random memory locations in error, it could be just about anything
which is at fault. Prime95 doesn't give failure addresses, so is
no help in that regard. Prime95 is a "system acceptance test", the
kind of thing you'd run, before delivering a system to a customer.
I sometimes combine Prime95 with a concurrent graphics test, as
a more stressful test case, but we're not there yet.
6) After you've completed the basics, you can add the video card and
do your 3D tests. If the system behaves entirely differently, then
the video card could be bad. Add other hardware and retest as needed.
*******
Your power supply. Amazing - it even has -5V on it!
http://www.newegg.com/Product/Produc...82E16817159047
+3.3V @ 30A, +5V @ 40A, +12V @ 18A, -5V @ 1A, -12V @ 1A, +5VSB @ 2.5A
I think you've got enough amps on +12V, for the hardware. You have
a 65W processor, a 28.7W video card, so the load isn't that great.
As long as the supply is "healthy", and isn't the one from the
failed system, it is probably OK. You can either check the hardware
monitor page in the BIOS, for voltage readings, or dig out a
multimeter and check them if you're curious. Sometimes, the
hardware monitor readings are wrong (on one system here,
checking the +12V with a multimeter, revealed it was right
on the mark, while the BIOS said it was way off).
Some video card drivers, have special case code to handle Furmark.
If they detect you using Furmark, some throttling is applied, to
prevent burnout. At one time, you could damage a video card with Furmark.
HTH,
Paul