NF4 UltraD 1st successful OC, need help with OCZ RAM

[ed_jacobson]

My main priority with this build has been to get a stable, quiet system at stock speeds. I’ve got that - although my USB ports are still reading as 1.1, but that's another story to sort out.

 

Now, I wanted to take a look at OCing, with the objective of getting stable speeds at 2.0 GHz and 2.2 GHz, roughly equal to the stock speeds of the AMD 3200+ and 3500+ cpus. For the most part I expect to run at stock speed of 1.8 GHz, but OCing may come in handy as I desire more speed over time. I’m new to this, as this was my first ever computer build venture, nevertheless OCing, and I’d appreciate some feedback on my technique, results, and especially regarding OCing my RAM.

 

I was noodling around with OCing a bit over the past week and ran into some problems, especially with my RAM. There are so many variables I was having a hard time keeping track of them and I felt I was randomly making changes and there was no clear way to keep track of cause and affect, so I decided to start at square 1 and a constant set of basic assumptions. I set the BIOS to all optimal default values with the following global changes that apply to all test variations unless otherwise noted below: (1) turn off CPU throttling; (2) set PCI payload to 256mb; (3) disabled Cool’n’Quiet; (4) disabled bank interleave; (5) set FSB Bus to 200; (6) set LTD/FSB to x5; (7) set CPU/FSB Freq Ratio to x9; (8) manually set memory timings to rated timings 2-3-2-6; (9) manually set DRAM Frequency to 200 1:1; (10) manually set Command Per Clock = Enabled = 1T).

 

Step 1: Test Optimal stock settings w/ global changes for stability and benchmarks.

 

Test Purpose: Test stability at 1.8 GHz and document stock speed and memory bandwidth benchmarks.

 

Test Procedure: Test stability of base settings for stability using Memtest and Prime95. Test and document the following benchmarks: Memtest bandwidth; SuperPi 1M, 2M, and 8M; Aquamark3; SiSoft Sandra Lite Memory Bandwidth, CPU Arithmetic, CPU Multi-media, Cache and Memory; Everest Memory Read, Write, and Latency.

 

Result: Stability: Memtest (2 passes), Prime95 (12+ hours) , and SuperPi 8M.

Speed: documented results of benchmark tests.

 

Conclusion: Stable at stock speed and memory settings.

 

Step 2: Calculate Max HTT

 

Purpose: Isolate HTT by underclocking CPU and memory to calculate max HTT capable.

 

Assumptions/Changes to Settings: (1) set CPU/FSB Freq Ratio to x7; (2) set DRAM Frequency to 100 1:2

 

Test Procedure: Once the CPU and memory have been isolated and taken out of the equation, raise FSB incrementally until cannot boot into Memtest.

 

Result: Able to boot into Memtest through FSB = 310. At FSB=320, booted into Memtest but then froze up.

 

Conclusion: Approx. max HTT = 310. (Is there a better way of testing max HTT? Is this in the ballpark of what to expect?)

 

Step 3: Calculate Max CPU

 

Purpose: Isolate CPU by underclocking HTT and memory to calculate max CPU speed.

 

Assumptions/Changes to Settings: (1) set LDT/FSB Freq to x4; (2) set CPU/FSB Freq Ratio to x9; (3) set DRAM Frequency to 100 1:2; (4) set CPU VID Control (vcore) to 1.4; (5) set CPU VID Special Control to x110%; (6) Change LDT/FSB Freq to x3 once FSB >250 (to keep HTT < 1000).

 

Test Procedure: Once HTT and memory have been isolated and taken out of equation, raise FSB incrementally until unstable in either SuperPi 8M or Prime95 (5 minutes).

 

Result: Able to raise FSB and calculate SuperPi 8M and 5 min stable in Prime95 through FSB=250. At FSB-260, SuperPi 8M calculated successfully, but Prime95 fatal error (rounding was 0.5 expected less than 0.4) after 1 minute.

 

Conclusion: Max CPU speed approx = 250x9 = 2.25GHz. (I did not test between 250 and 260). Smartguardian indicated CPU volts = 1.48v. I could have probably increased max CPU speed by increasing vcore or Special Control to 123%, but 2250 MHz met my expectations, so that was good enough for now. (I know I’ve seen other people get the 3000+ Winnie higher than 2.25, in the range of 2.4 which is about 270x9. Is there a better way to test max CPU speed and get higher clocks?)

 

Step 4: Calculate max memory overclock.

 

Purpose: My memory (PC3200 DDR400) is rated at 200MHz which will limit OC if I plan to run FSB > 200 at 1:1. Isolate CPU and HTT in order to test ability of RAM to run at higher FSB than 200 and calculate max memory overclock.

 

Assumptions/Changes to Settings: (1) set CPU/FSB Freq Ratio to x7; (2) set LDT/FSB Freq to x4; (3) set DRAM Frequency to 200 1:1; (4) memory timings to 2-3-2-6.

 

Test Procedure: Once CPU and HTT have been isolated and DRAM Freq set to 1:1, starting at 200, raise FSB incrementally to determine max RAM overclock.

 

Results:

Test 1: FSB =210; Result: 1.8 million errors in Memtest 5; manually shut down.

 

Test 2: FSB=210; Increase DRAM Volts from 2.6 to 2.7; Result: 600000+ Memtest 5 errors; manually shut down.

 

Test 3: FSB=210; Increase DRAM Volts from 2.7 to 2.8; Result: 1142 Memtest 5 errors, 6037 Memtest 8 errors after 1 complete pass. Better, but still, RAM no likey.

 

Test 4: FSB=210; Change memory timings to 2.5-4-4-10; reset DRAM Volts to 2.6; Result: No Memtest errors after 1 pass! Aha! RAM no likely high FSB and tight timings!

 

Test 5: As in Test 4 above; FSB=220; Result: Memtest has funky gibberish text, Pass # rapidly goes to 2354 and then locks up.

 

Test 6: As in Test 5 above, DRAM volts increased from 2.6 to 2.7; Result: Memtest has random text gibberish, locks up.

 

Test 7: As in Test 6 above, DRAM volts increased from 2.7 to 2.8; Result: Unexpected Interrupt – Halting message, locks up. RAM no likey FSB 220 even with loose timings and more juice.

 

Test 8: Reduce FSB=215, DRAM volts = 2.6; Result: Memtest 5 362 errors, Memtest 8 313,263 errors after 1 pass.

 

Conclusion: At this point I gave up trying to OC my memory. My RAM clearly didn’t like FSB in the range of 220 even with higher voltage. I had a successful result at 210 with somewhat loose timings and default voltage, so for now, I concluded max RAM OC = 210 = only a 5 percent OC for RAM that is not value RAM!!

 

Step 5: Consolidate max HTT, CPU speed, and RAM for OC approx. 2.2 GHz.

 

Test Purpose: Test OC of approx. 2.2 GHz. My max HTT = 310; max CPU speed = 250*9=2250MHz; max memory OC = 210 < max CPU FSB of 250 meant my memory was my limiting factor and I need to use a divider to get memory to work with 2.25 GHz OC. Using a FSB of 250 and DRAM Frequency of 166 5:6 would yield a RAM OC of 208, which is just under my max OC of 210.

 

Assumptions/Changes to Settings: (1) FSB =250; (2) DRAM Frequency 166 5:6; (3) memory timings at 2.5-4-4-10; (4) set LDT/FSB Freq to x4; (2) set CPU/FSB Freq Ratio to x9; (3) set CPU VID Control (vcore) to 1.4; (4) set CPU VID Special Control to x110%.

 

Result: OC at 2.25 GHz (a 25% OC above rated CPU speed of 1.8 GHz), memory at 208 (CPUZ reads memory at 205 rather than 208, don’t know why).

 

Stability: Memtest no errors after 1 pass; Prime95 stable 8+ hours. So far so good.

 

Benchmarks:

 

Memtest bandwidth: decrease 8%

 

SuperPi 2M: decrease time by 9%

 

AquaMark3

GFX: decrease 1%

CPU: increase 8%

Total: decrease .1%

 

Sandra

Memory Bandwidth Int.: increase 11%

Memory Bandwidth Float: increase 12%

 

CPU Arith Dhry: increase 23%

CPU Arith Whet FPU: increase 23%

CPU Arith Whet iSSE2: increase 23%

 

CPU Multi-Media Int: increase 23%

CPU Multi-Media Float: increase 24%

Cache & Memory combined: increase 5%

 

Everest

Memory Read: decrease 4 percent

Memory Write: decrease 31%

Memory Latency: increase 2%

 

Overall, I think this has been a successful first attempt at OCing. I’m not breaking any records, but that’s not what I set out to do for right now. Rather that trying to tweak multiple BIOS settings at a time, I limited myself to just a handful that I understood, while leaving the default settings for the rest. I know this may have limited my results somewhat, and starting with some slightly different baseline assumptions might have helped my results, but I thought this was better than taking wild butt guesses at multiple variables without knowing exactly how they were affecting my results. I’d appreciate any feedback and guidance on what you think I did right or wrong.

 

I’d especially like some feedback on how I can try to OC my OCZ RAM more than FSB=210. I feel like I’m probably taking a hit by using a RAM to FSB ratio that is less than 1:1 as well as the loose timings that is probably affecting my results. I can certainly tweak the timings from here, but a 5 percent OC on this RAM seems very low…I know it is rated at PC3200 DDR400 which means a FSB of 200, but what is a typical OC for higher quality DDR400 RAM? Am I in the ballpark, or not? Giving it more juice seemed to help in that Memtest errors were reduced at FSB 210, but loosening the timings helped more than giving it more juice, and it didn’t appreciate FSB 220 at all even with the looser timings. Any ideas on BIOS settings to get better results would be appreciated.

 

If you’ve taken the time to read through this, thank you! I know there are lots of questions sprinkled throughout, but any help or advice on any aspect of this endeavor is greatly appreciated.

[ooztuncer]

Ed, you did it again! Lets add correct and usable information to this thread and come up with an updated "Overclocking Guide"...

 

Now, lets start with ur ram, which one do you have:

 

1) EL DDR PC3200 Platinum Limited Edition (2-2-2-7) --chip--> Winbond BH-6

 

2) EL DDR PC3200 Platinum Edition (2-3-2-5) --chip--> Samsung TCB3, ProMOS 5ns (relabeled), Mosel Vitelic 5ns (relabeled), Winbond CH-5

 

3) EL DDR PC3200 Platinum Edition rev2 (2-2-2-5) --chip--> Samsung 4ns TCCD ~ Brainpower PCB

 

4) EL DDR PC3200 DC Value VX (OCZ4001024WV3DC-K) (3-4-4-8 @2,6V) --chip--> Windond UTT (relabeled as OCZ VX) ~ Brainpower PCB 808

 

5) EL DDR PC3200 DC Gold VX (2-3-3-8 @2.6V | 2-2-2-8 @3.2V) --chip--> Windond UTT (relabeled as OCZ Voltage eXtreme) ~ Brainpower PCB 808

 

 

I'm confused because your sig says you have 2-3-2-7 by default and I couldn't find a match.

 

According to chip on the RAM, I think people will give different advice for overclocking. For instance, UTT(CH-5) is a completely different animal than TCDD. :cool:

[ed_jacobson]

This is my stuff:

 

http://www.ocztechnology.com/products/memo..._el_ddr_pc_3200

 

http://www.newegg.com/app/ViewProductDesc....-146-883&depa=1

 

The stock rated speed is 2-3-2-6, but I run it at 2-3-2-7 at stock speeds because it gives me slightly higher bandwidth with tRAS at 7. This is not TCCD, but I don't which chips they are.

[Guest nitro]

Looks like you did a thorough job testing. The litttle that I have read about your ram doesn't look like it overclocks well. It may take 2.8 or 2.9 volts for you to get anywhere with it. I would try to keep the timmings as tight as possible if you can't get much above stock. I doubt if the little increase you were able to get will offset going with tighter timings. Try A64 tweaker to play with your settings. As for the cpu, up the volts to 1.475x110% to give you 1.62v. This should be safe even for stock cooling. Just keep an eye on your temps. Then try LDT 3x, FSB 270, CPU at 9x, memory on 150 divider. This would give you a little more than 2.4Ghz and you memory will be running close to 200. You may want to try lower than that, say 9x260, and work your way up. Then bench again and compare with what you have. Good Luck!

[nashville1971]

Ed,

 

I'm surprised that you weren't able to find a higher max memory value. I'm using the Rev.2 of the OCZ PC3200 EL sticks--meaning that one important difference between my setup and yours is that I'm using TCCD--and I'm able to crank my memory bandwith to about 300Mhz. My memory is also rated to 2.9v on the vdimm. Set your basic memory timings to 2.5-3-3-7(or 8) and 1T initially on a 1:1. A good initial test for stability at a given memory bandwith is a run of Memtest #5 and #6.

 

Try these additional settings to see if they help with your mem overclock...

 

1.) set DRAM Drive Strength to 8

2.) set tREF to 4078

3.) increase vdimm to 2.9v (but not higher)

4.) disable bank interleave

5.) if all else fails, try the 3/10 BIOS

 

Some have also said that their OCZ PC3200 Platinum sticks function better in a higher bandwith range, like 220-240 Mhz or 240-270Mhz. While this may be true, I have no explanation for it.

 

I hope this helps!

[AndyOCZ]

Hi, the PC3200 Platinum (not Rev 2) has not yet been fully tested on the DFI NF4 board. We received sample boards AFTER they were available retail, so the market had a huge jump on us. All RAM manufacturers are in the same boat here.

 

I also doubt that DFI is testing all availble memory types for BIOS compatabilty, so we may be on our own.

 

I just requested some of this memory for testing on the DFI. I hope to have it in a few days at the most. This memory does pretty well for overclocking on other platforms; of course getting the timings correct in the DFI BIOS is the real challenge.

 

When my samples come I will be of more assitance.

 

Hang in there.

[ed_jacobson]

Hi Andy, thanks for responding. I really appreciate it. FYI, my RAM isn't the platinum version, it's the performance dual channel enhanced latency 2-3-2-6 shown here

 

http://www.ocztechnology.com/products/memo...00_dual_channel

 

Maybe it's the same chips as the platinum variety, I don't know. It was on the DFI list of compatible memory types when I ordered it. And it works great at stock speeds, just having trouble getting a FSB higher than 210.

 

I tinkered around with the bios settings last night, especially with tREF as recommended over on the bleedinedge OCZ forum. Mostly working with slightly loose settings, in the range of 2.5-3-3-7 and 2.5-3-3-10, I tried about 10 variations on tREF, and then with vdimm from 2.6 to 2.9, and the cpu special at 123%, and still couldn't get above 210. tREF made a difference...sometimes only getting errors in Tests 5 and 8, other times Tests 2, 4, 5, 6, and 8, and sometimes Test 1 also. At a FSB of 220 or higher, usually memtest wouldn't boot or would freeze up.

 

Let me know if get a chance to play with it a bit.

[RyderOCZ]

Ed, try using 2.6V on the ram and timings of 3-3-7-3 or 3-4-7-3 (top to bottom in the bios) Trc at 12, Trfc at 24, everything auto down to Tref at 3072, Drive strength to 5 or 7, DATA strength to 2, Max Async at 7ns (try 8ns also), Idle Cycle counter to enable, Idle Limit to 256 clks.

[ooztuncer]

Hi Ed,

 

Actually I asked about your RAM 2 days ago in this thread (look at page 9 reply #221 and #222): http://www.xtremesystems.org/forums/showthread.php?t=50010

 

And also in this one: http://www.xtremesystems.org/forums/showthread.php?t=56898 (check it out page 5, #113-116(this is from the CEO of the OCZ )-119-121-124)

 

According to jjcom, these were(are) Hynix based chips and he did 218mhz 2.5-3-3-10 with 2.8v.

 

And I'm sure that Andy and/or Ryder will help you with the optimal solution...

 

All for u... and for my curiosity also

 

PS: If they are still Hynix based, I believe you shouldn't expect a lot in DFI mobo (according to AG but you did not heard from me :spam: )

[ed_jacobson]

Here are the latest tests I've run on this RAM:

 

Test 1.

 

FSB=215

DRAM - 2.6v

3

3

7

3

Trc = 12

Trfc=24

Auto

.

.

.

Tref 3072

Drive Strength 5

Data Strength 2

Max Async 7

Idle Limit 256

 

Result: 2156 Test 5 errors, 100000+ Test 8 errors.

 

Test 2.

 

As above, except Drive Strength 7

 

Result: 134 Test 5 errors, 100000+ Test 8 errors.

 

Test 3.

 

As in Test 1 above, except Max Async 8

 

Result: 2402 Test 5 errors, 100000+ Test 8 errors

 

Test 4.

 

As in Test 2 above, except Max Async 8

 

Result: 316 Test 5 errors, 100000+ Test 8 errors.

 

 

These settings seemed to have some affect on Test 5, but Test 8 is still a problem. If I increased the FSB to say 230, it wouldn't boot into memtest.

[AndyOCZ]

Hi Ed,

 

Actually I asked about your RAM 2 days ago in this thread (look at page 9 reply #221 and #222): http://www.xtremesystems.org/forums/showthread.php?t=50010

 

And also in this one: http://www.xtremesystems.org/forums/showthread.php?t=56898 (check it out page 5, #113-116(this is from the CEO of the OCZ )-119-121-124)

 

According to jjcom, these were(are) Hynix based chips and he did 218mhz 2.5-3-3-10 with 2.8v.

 

And I'm sure that Andy and/or Ryder will help you with the optimal solution...

 

All for u... and for my curiosity also

 

PS: If they are still Hynix based, I believe you shouldn't expect a lot in DFI mobo (according to AG but you did not heard from me :spam: )

 

Hi, I'm sure this is bad info. I believe the chips are Infineon, not Hynix. I have requested some of these to test on the DFI NF4 board. Hang tight and I will have some recommendations for you soon. I expect to have the memory by mid-week.

 

Thanks

[Smoken Joe]

I think I had the same ram I could only get to 2.2 if memory serves and that was with a divider. Everything was diffrent with some good TCCD. I hope he comes up with some good settigs for you otherwiase you are SOL. I have a feeling you are bot going to get far with that memory even with help so dont get hard on yourself if you dont. You are going at it the right way.