Just got my new Opty 165 and some G.Skill DDR500 delivered today!

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

5t3v0

Senior member
Dec 22, 2005
508
0
0
Prime95 has been running for 2 hours now & I'm cautiously optimistic that I've sussed it. I think I was being a bit wishful in setting tras to 6T so I've raised it 1 clock to 7T. However, this on its own wasn't enough to regain stability so I have set tref back to its spd value of 7.8µs. I tried setting tref to 7.8µs and leaving tras at 6T but this wasnt stable either. So the 2 changed settings from above are:

Min RAS# active time (Tras) = 7T
Refresh Period (Tref) = 7.8µs

Once I am confident I have got them stable, I'll post some benchmarks.
 

5t3v0

Senior member
Dec 22, 2005
508
0
0
Woke up to another BSOD this morning! Cook1, if you're still watching, pls can you let me know what you've set your other timings to, or have you left same as on your DFI post? Those settings dont work for me but I wonder if you've updated that page since changing to the 166 divider & lowering the Tcl etc. We are running the same cpu/ram frequencies so it would be really helpful to know what your timings are if you are stable. cheers m8
 

WA261

Diamond Member
Aug 28, 2001
4,631
0
0
NOt bad. I run my G.Skill pc4400 and 148 at 3.0Ghz on air. Do some more tweaking bro!
 

TrevorRC

Senior member
Jan 8, 2006
989
0
0
Originally posted by: Cook1
Was so excited all morning to come home and have two packages on my door step!

However...

G.Skill can't pass memtest at default settings :( Gah, what a way to ruin my weekend that I had planned of getting in some quality PC time.

I just recieved mine, as well--to let you know, they took a SUBSTANTIAL amount of tweaking to get them where I wanted them, but damn they're nice.

Give me a minute, and I'll post my A64 Twear/CPU-Z shot.
Still stability testing/improving the overclock, but a 20Mhz overclock on stock is bloody amazing! (Especially considering it's near TCCD speed anyway! [270Mhz!])

Also, Lower Voltage = BETTER!
If your board can do 2.5v, STICK IT AT 2.5!

http://img.photobucket.com/albums/v167/trevor_r_C/G.jpg

There, running 270 @ mainly stock timings/etc.

Any chance you can set a divider up to allow the RAM to run higher than the CPU? (on your board?)

 

Cook1

Diamond Member
Jul 11, 2004
6,315
0
86
Originally posted by: 5t3v0
Woke up to another BSOD this morning! Cook1, if you're still watching, pls can you let me know what you've set your other timings to, or have you left same as on your DFI post? Those settings dont work for me but I wonder if you've updated that page since changing to the 166 divider & lowering the Tcl etc. We are running the same cpu/ram frequencies so it would be really helpful to know what your timings are if you are stable. cheers m8

I'll post them when I get home for you. And I've ran Prime and Memtest overnight with no issues.
 

5t3v0

Senior member
Dec 22, 2005
508
0
0
I just cant get Prime95 stable at 2.5-3-3-7 1T @ 229MHz. I tried lowering vdimm from 2.6v to 2.5v & BSOD'd even quicker. I changed to 3-4-4-8 this morning to check memory wasn't goosed & Prime95 is still running now, 10 hours later so it must be OK. Damn, this memory id fiddly stuff. I read the G.Skill HZ thread on DFI Street & it seems like I'm not alone. Perhaps we should start a similar dedicated thread here for the HZ on all mobo brands.
 

JustStarting

Diamond Member
Dec 13, 2000
3,135
0
76
I don't think it's so much that the memory is freaky, but it's that the memory controller is on die. I swapped from a 146 to 170 to a 165 and my memory timings need tweaking for each CPU.

In the old days you could get your memory tweaked in with any CPU and it would stay optimized with any other CPU thereafter, but not with these Opty's.... especially the dual cores:|

If you have a dual core I just recommend relaxing your timings as much as possible within reason. Bandwidth just does not matter..... dual cores suck at bandwidth compared to single cores anyway. Just get stability;)
 

JustStarting

Diamond Member
Dec 13, 2000
3,135
0
76
system is in the sig.....

My TCCD does 1:1 up to DDR620 memtest stable (Windows stable to DDR610)) with these settings:

cpc enabled
tcl 3
trcd 4
tras 8
trp 4
trc 12
trfc 16
trrd 04
twr 03
twtr 02
twrt 04
tref 3120 or 3072
twcl 1

DQS Skew Control = INCREASE
DQS Skew Value = 0 (zero)
Dram Drive Strength= 8 or 6 (704-BT2A= weak 4 or weak 3)
Dram Data Drive Strength= 4 (704-BT2A= Level 1)
MAX ASYNCH LATENCY = 8X
READ PREAMBLE= 5X
IDLE CYLCE= 256 cycles
DYNAMIC Counter= Disabled
R/W Que Bypass = 16x
ByPass Max 7X
32 Byte Granularity = Disable 8 Bursts

with my 170 at 280 FSB and 2.8Ghz Sandra bandwidth is 7300+. My 165 at 304FSB is 7100+. Stable as heck and trying any tighter does not make a difference.

With my 146 at 292fsb my Sandra bandwidth is 7800+.
 

5t3v0

Senior member
Dec 22, 2005
508
0
0
Thanks for the post JustStarting. These G.Skills are UCCC ICs in double banks of 64Mx8 & the timings/overclocking characteristics are totally different from TCCD. 1Gb modules are supposed to be harder to overclock than 512M & these are proving to be no exception. I tried 1:1 with my cpu at 280HTT with timings of 3-4-4-8 1T but the pc intermittently failed to post, at 275 it posted but failed memtest on test 5. Didn't bother trying any lower because I wouldn't be prepared to sacrifice that much cpu speed. I went back to using the 166 divider giving 229MHz on a 280HTT but I just cant get Prime stable with 2.5 CAS latency. I was hoping that by running the ram lower than spec MHz, I could get tighter timings but the best I've managed to date is 3-3-3-7 (8 hours prime stable so far). I'll post some comparative benchies once I'm happy with the stability. Here's a full list of my best timings:

CPU Frequency = 280
CPU Multiplier = 9
DRAM Divider = 166 (5:6)
DRAM Frequency = 229
CPU voltage = 1.55v
DRAM voltage = 2.6v
Chipset voltage = 1.5v
Command Per Clock (CPC) = 1T
CAS Latency Control (Tcl) = 3
RAS# to CAS# delay (Trcd) = 3T
Min RAS# active time (Tras) = 7T
Row precharge time (Trp) = 3T
Row Cycle time (Trc) = 7T
Row refresh cyc time (Trfc) = 13T
Row to Row delay (Trrd) = 2T
Write recovery time (Twr) = 2T
Write to Read delay (Twtr) = 1T
Read to Write delay (Trwt) = 3T
Refresh Period (Tref) = 7.8µs
R/W Queue Bypass = 16
Bypass Max = 7
32 Byte Granularity = Disable
DQS Skew Control = Disable
DQS Skew Value = 0
Read Preamble Time = 5.0ns
IdleCycle Limit = 16T
Max Async Latency = 7.0ns
Dynamic Counter = Enable
 

Diasper

Senior member
Mar 7, 2005
709
0
0
Maybe you might want to try each RAM stick individually to see what they're capable of - or perhaps if there's problems with one particular stick.
 

TrevorRC

Senior member
Jan 8, 2006
989
0
0
As far as I know, these can't do DDR400 at extremely tight timings.


Mind letting us in on your divider options?

Also, can your board do 300-310 HTT?
--Trevor

You could do around 297HTT @ an 8.5 Multi, and run the RAM @ 9:10
That'll put your RAM @ 270, which is near where it will peak out (within 1-3 Mhz) and will keep your CPU at it's max, too. (I'm assuming you're running a 165?)

Also, again, low voltage > high voltage with these sticks--I can prime @ 270 with 2.5, but if I go to 2.8 I fail o_o;;
 

5t3v0

Senior member
Dec 22, 2005
508
0
0
Trevor, unfortunately the divider options below 1:1 are limited on Epox boards to 1:2, 2:3 & 5:6. The 9:10 option would be great even at 280x9 since I would at least be getting stock (252MHz) out of these rams. I have complained to Epox about their limited dividers & requested more in future BIOS releases but doubt anything will come of that. I haven't tried the cpu at 320x8 in anger, which on a 5:6 divider, would give a memory freq of 256MHz. Perhaps I should give that a whirl. I too found that high voltage made stability worse but in earlier testing, 2.5v caused BSOD faster than 2.6v so I've stuck at 2.6. BTW, I have a 144, not 165.
 

TrevorRC

Senior member
Jan 8, 2006
989
0
0
Sorry, just assumed from the clock you had a DC.

Try the high HTT, see where it goes--doubtful the board can put it out, but who knows! Best of luck.
--T
 

Job

Senior member
Jan 16, 2006
283
0
0
Recently go hold of an opteron 165, managed to overclock it to a reasonable 2.66Ghz by raising the fsb to 295 and using the ram divider to turn the memory down to 150mhz (it runs at 221mhz as a result). However, as my ram is only the value variety, my system often crashes even with the memory overclocked so modestly, so I have to lower the fsb to keep stability, with the highest speed around the 2.61ghz mark.

For some reason, my motherboard (dfi nf4 sli-d) will allow me to select ram speeds as low as ddr200, but windows will not boot with ram running slower than ddr300! I would like to get the cpu to at least 2.7 or even 2.8, is there any way of getting the ram to run slower or more stable so I can get the fsb up past 300?

Opteron 165 2.61ghz @ 1.5V
4 x 512mb value ram (pc3200) 2.5-3-3-6 @ 2.7V
DFI lanparty nf4 UT Sli-d
500W PSU
 

5t3v0

Senior member
Dec 22, 2005
508
0
0
Job, do you have the GeIL with the 2.5-3-3-6 timings (SKU GE5123200B)? I used to have that memory & clocked it up to 237MHz on stock timings & 229MHz on 2-3-3-6, both at 1T command rate. To get these overclocks needed more volts though - 3.0v in the BIOS, actual output 3.05v. This might seem high but remember they are rated by GeIL up to 2.95v. My prefered setting on this 144 was 280x9 using a 5:6 divider to give 229MHz on the memory. Any higher on the cpu needed > 1.55v which is uncomfortably high for me.

Also be careful about the timings if you've left them at Auto. I found that GeIL had some rather ambitious spd values when using the dividers < 1:1 so I set the HTT to 200 and divider to 1:1 and after a restart, went through all the timings and set each one manually to lock them at the 200 spd values.
 

Job

Senior member
Jan 16, 2006
283
0
0
Yep, those are the timings; I have two sticks of 512 geil and two sticks of generic 512 I bought on the cheap (same timings tho), they are working on 2T at the mo (I understand this is automatic for boards using all four dimms). Does this mean I can be stable if I just push up the voltage, or will I have to change the timings? If I need to change the timings, what would you recommend?
 

5t3v0

Senior member
Dec 22, 2005
508
0
0
If you're running 4 dimms then your options will be fairly limited & the max OC speed will depend on your slowest dimms, most likely the generic pair. Perhaps you could get some more GeIL to replace the generic sticks but that still might not help. I sure I've read that 4 dimms OC poorly compared to 2 dimms so you're probably stuck with DDR400. I almost bought 2 more GeIL to add to my previous pair but went for 2x1Gb in the end & sold the GeIL on ebay. Going on your experience, I'm glad I did. If you are having stability problems, then I suggest you put the matched pairs in seperately & make a note of the timings for each, then when you put all 4 together, make the sure the timings are the same as the worst pair & if not set them. 2.5-3-3-6 seems a bit high spec for generic ram - is that their correct timings? You wont get any better than 2T CPC with 4 dimms either. BTW, did you just change your sig? I'm sure you only had the 2x512 GeIL before.
 

Job

Senior member
Jan 16, 2006
283
0
0
It seems the timings for the generic pair are 2.5-3-3-8 and the geil are 2.5-3-3-6 - POST screen shows timings of 2.5-3-3-7 for the whole lot if I'm reading it right - wish I'd gone for 2x1gig now! If I can match the timings for all four sticks at a lower spec, could I get a more stable overclock? I'm stuck at 290FSB at the mo, anything more crashes during games -

Yeah, changed my sig - updated to include the extra ram

Thanks for your help!
 

5t3v0

Senior member
Dec 22, 2005
508
0
0
No worries m8. I'm afraid I'm just guessing as I dont have any experience of running 4 modules, but I'd suspect that you will find that setting the timings to the slowest modules will help. You will probably find that you'll need to change more than just tCL-tRCD-tRP-tRAS though. Even you cant get higher HTT, 9x290 is a very good OC for that chip is it not? I cant get that high on my 144 unless I use 1.65v vcore which is just asking for trouble - hence why I settled for 9x280 on 1.55v (I am currently testing 9x288 on 1.6v). You may find you've just hit the wall of your cpu. What vcore are you using? I presume you are using a 133 (2:3) divider with that HTT? My recommendation would be to set your HT multiplier to x2 and your memory divider to 133 (2:3) and see if you can raise it above 290.
 

Job

Senior member
Jan 16, 2006
283
0
0
I'm still very happy with the overclock, its just endlessly annoying to know that you can go just a bit further! THe HT multi is set to x3 at the mo, and for some reason windows wont boot when I lower the multi on the memory below ddr300 (3:4) - I'm pretty certain its the RAM, the cpu has plenty of Vcore (1.5 - 1.55) and still idles at 25 degrees C even with the 800mhz overclock - I'll try changing the timings of the geil to match the generic (2.5-3-3-8) and see what happens!
 

5t3v0

Senior member
Dec 22, 2005
508
0
0
A 150 divider on 9x290 = 217.5MHz memory frequency. It's unlikely you would get that on 4 sticks of any ram, let alone value+generic ram. Have you tried the 143 multiplier? That would give 200.9 MHz which should work, or to be safe, the 133 multi would give 186MHz. Defo change those timings to the worst sticks though. I wouldn't recommend using a high vcore if you dont have to, it wear the chip out quicker even if the heat is OK. Leave it at 1.5v if you can. I am reluctantly using 1.55v but I'd prefer to be lower.
 

Job

Senior member
Jan 16, 2006
283
0
0
yeah, the 150 is working fine for the moment at 217.5, but its the problem where any lower divider wont let me boot into windows thats causing me the problem - my board gives me several lower dividers than 150, but none of these get past the windows boot screen - any idea why this might be? - can't seem to find any info on the net about this specific problem
 

5t3v0

Senior member
Dec 22, 2005
508
0
0
Sorry misunderstood your previous post. That is strange that the memory wont work at lower frequency. The only possible cause I can think of is that the BIOS is adjusting the memory timings to something the ram cant handle. Did you do as I suggested before and manually set the timings at their values on 217MHz? When I had the timings on my GeIL rams set to "Auto", the spd adjusted them to be more agggressive (e.g. tCL from 2.5 to 2.0) when I tested on the 133 divider (186MHz). The GeILs on their own could handle this, but if yours is doing the same with the generic ram, that might cause it to fail POST. Basically, dont leave any settings on Auto & see if that helps.
 

Job

Senior member
Jan 16, 2006
283
0
0
Set the timings to 3-3-3-8 and still not able to post below 150 (my board gives me 140, 130, 120, 100 options) - There are a helluva lot of mermory options here, this is what I set them to:

TCL=3 (There was no .5 integer available)
TRCD=3
TRAS=8
TRP=3
TRC=8
TRFC=AUTO
TRRD=2
TWR=2
TWTR=2
TRWT=3
TREF=3120 cycles
TWCL=auto

I guess you can appreciate how long it would take to test all these options!! (there are several others too) - CPU-z tells me that the Geil changes its settings when set at a lower frequency, which means its understandable the generic RAM wouldn't be able to handle the drop; but now I've set the timings manaully it should be OK - but still, no post! Is there any other way of forcing the bastards to slow down so they dont crash?

Many thanks!