[Ret Sticky]Overclocking sndbx for A64 939 systems with Winchester, Opteron dual core

Stability testing using Stress Prime 2004 (SP2004)

I define a system including all its components (CPU, memory, chipset, video subsystem, ...) "stable" at certain frequency, voltage and temperature only if it can run ALL "well tested" and "error free" programs with properly installed OS for certain required amount of time.

If some programs, drivers crash the system with properly installed OS (with fixes if necessary), and if it can be established that the abend, crash, exception are not due to the software, I do not consider the system as "stable".

If a system can pass certain programs at lower frequencies but fails at higher frequencies of CPU, memory, system bus, then the system is considered not "stable" at the latter higher frequencies.

IMO, Prime95 is essential for stability testing.


Stress Prime 2004 (SP2004) is based on Prime95 with a useful graphic interface. I tested it and think it is very flexible for setting up Prime95 runs.

In additional to the usual Prime95 options of small FFT/large FFT/blend/custom, in place, priority, run time per FFT size. It allows
- setting up arbitrary virtual memory size for a Prime95 run or in-place Prime95 run
- selecting a custom set of different FFT size for the run (from 8K to 4096K)
- rearranging the ordering of the various FFT size for the run
- showing the actual time passed during the run
These options are very handy for diagnosing Prime95 failure, whether it was CPU or memory or system related (chipset, system bus).
(The tested version reports incorrectly the HTT frequency.)


CPU: Winchester 3000+ CBBHD 0447
memory: G. Skill 4400 LE 2 x 256 MB (Samsung TCCD)
motherboard: DFI LanParty UT NForce4 Ultra-D (rev. A02, bios 03/10/05)
cooling: XP-90, 80 mm Tornado with fan control
OS: Windows XP Professional SP1

- CPU: 2.72 GHz = 302 MHz x 9, 1.52 V
- memory: 302 MHz, 2.5-3-3-7 1T, 2.8 V
- Prime95 tested for 28 hours (user aborted)

This system is able
- to boot OS at 2.95 GHz (328 MHz x 9),
- to complete SuperPI 1M at 2.90 GHz (322 MHz x 9),
- to complete SuperPI 32M, 3dmark 01/03/05 at 2.85 GHz (317 MHz x 9).
So as usual, Prime95 is more difficult to pass, the usual frequency span of 200 - 250 MHz between boot/SuperPI/3dmark/Prime95.


Stability testing using memtest, SuperPI (32M) and Prime95

hitechjb1 said:

It is obtained by reducing memory timing from 2.5-4-4-8 1T to 2.5-4-3-7 1T, memory frequency is lowered to 315 MHz from 317 MHz and
CPU is lowered to 2.83 GHz from 2.85 GHz from that previous result.

This shows a good trading memory frequency and CPU frequency with tighter memory timing.

Click to expand...

I wonder how your jaw will drop once you set RAS-to-CAS-delay to "3" instead of "4"- you should really try that- no matter if you have to lower the frequency of you RAM.

For calculating the right memfrequency I'll attach a tool- simply rename from .jpeg to .exe and enjoy if you don't have it already

edit: woh, seems to have problems...

Some results about comparing memory frequency, memory timing, memory divider

These are some numbers on 3dmark01 on different memory timing at various memory bus frequency. They were not measured at the same day (not well controlled) and 3dmark01 also can fluctuate by 200 points with the same exact setting. (I have been thinking about how to tighten the spread of 3dmark01 and other benchmarking numbers, but do not get a chance to look into it yet.)

DFI LP UT NF4 Ultra-D
Winchester 3000+ CBBHD 0447
BFG 6600 GT 1.19/0.59 GHz

G. Skill 4400 LE TCCD 2x256 MB 2.8 V

CPU_multiplier = x9
memory_HTT_ratio = 1:1 (memory = CPU/9)

CPU 2.77 GHz 308 x 9 MHz memory 308 MHz 2.5-3-3-7 1T 3dmark01 23426
CPU 2.77 GHz 308 x 9 MHz memory 308 MHz 2.5-3-3-7 1T 3dmark01 23508
CPU 2.77 GHz 308 x 9 MHz memory 308 MHz 2.5-3-3-7 1T 3dmark01 23573
CPU 2.79 GHz 310 x 9 MHz memory 310 MHz 2.5-3-3-7 1T 3dmark01 23497
CPU 2.80 GHz 311 x 9 MHz memory 311 MHz 2.5-3-3-7 1T 3dmark01 23576

CPU 2.80 GHz 311 x 9 MHz memory 311 MHz 2.5-4-3-7 1T 3dmark01 23361
CPU 2.84 GHz 315 x 9 MHz memory 315 MHz 2.5-4-3-7 1T 3dmark01 23643
CPU 2.85 GHz 317 x 9 MHz memory 317 MHz 2.5-4-3-7 1T 3dmark01 23597 ?

CPU 2.85 GHz 317 x 9 MHz memory 317 MHz 2.5-4-4-7 1T 3dmark01 23829
CPU 2.85 GHz 317 x 9 MHz memory 317 MHz 2.5-4-4-7 1T 3dmark01 23829
CPU 2.85 GHz 317 x 9 MHz memory 317 MHz 2.5-4-4-8 1T 3dmark01 24365
CPU 2.86 GHz 318 x 9 MHz memory 318 MHz 2.5-4-4-7 1T 3dmark01 23784


CPU_multiplier = x9
memory_HTT_ratio = 5:6 (memory = CPU/11)

CPU 2.86 GHz 318 x 9 MHz memory 260 MHz 2.5-3-3-6 1T 3dmark01 23095

CPU_multiplier = x9
memory_HTT_ratio = 9:10 (memory = CPU/10)

CPU 2.85 GHz 317 x 9 MHz memory 285 MHz 2.5-3-3-7 1T 3dmark01 24407
CPU 2.86 GHz 318 x 9 MHz memory 286 MHz 2.5-3-3-6 1T 3dmark01 24027
These two are interesting, higher than memory 308-311 MHz 2.5-3-3-7 1T (see explanation below)


TwinMOS Speed Premium 3200 UTT (AA4T 44D) 2x256 MB 3.5 V

CPU_multiplier = x9
memory_HTT_ratio = 5:6 (memory = CPU/11)

CPU 2.77 GHz 308 x 9 MHz memory 252 MHz 2.0-2-2-5 1T 3dmark01 23246
CPU 2.77 GHz 308 x 9 MHz memory 252 MHz 2.0-2-2-5 1T 3dmark01 23371
CPU 2.79 GHz 310 x 9 MHz memory 254 MHz 2.0-2-2-5 1T 3dmark01 23409
CPU 2.80 GHz 311 x 9 MHz memory 255 MHz 2.0-2-2-5 1T 3dmark01 23576


Summray:

1. 288 MHz 2.5-3-3-7 1T (CPU/10) may get higher point than 3dmark01 than 308-311 MHz 2.5-3-3-7 1T (CPU/9).
Possible explanation, higher memory_CPU_divider gets higher memory bandwdith efficiency with memory_HTT_ratio < 1 for the same CPU multiplier (better than 1:1). This is different from general belief (advanced topic).
Memory bandwidth and efficiency in terms of CPU frequency, memory frequency, CPU_memory_divider, CPU_multiplier, memory_HTT_ratio

2. Memory at 308-311 MHz 2.5-3-3-7 1T (TCCD) with CPU/9 is about tie with 252-255 MHz 2.0-2-2-5 1T (UTT) with CPU/11 (higher efficiency) in 3dmark01.

3. As 250-260 MHz is about the max for many BH-5/UTT 2-2-2-5 1T. If some TCCD can attain higher than 317 MHz with 2.5-3/4-3/4-x 1T, and even with a smaller memory_divider (9-10 for TCCD vs 11 of UTT, smaller divider gives lower bandwidth efficiency), the 3dmark01 result shows that the TCCD would still come out ahead.
E.g. as illustrated by the 3dmark01 result with a 6600GT at 1.19/0.59 GHz
- upper-23k to 24k+ for TCCD at 317 MHz 2.5-4-4-8 1T CPU/9 vs mid-23k for UTT at 255 MHz 2-2-2-5 1T CPU/11
- 24k+ for TCCD at 285 MHz 2.5-3-3-7 1T CPU/10 vs mid-23k for UTT at 255 MHz 2-2-2-T 1T CPU/11

4. 308-311 MHz 2.5-3-3-7 1T (CPU/9) may be about equal to 312-314 MHz 2.5-4-3-7 1T (CPU/9) in 3dmark01. 1 tRCD cycle averaged about 3-4 MHz in memory frequency at 300 MHz level.

Consistent with:
Testing UTT and TCCD memory modules in Winchester and DFI NF4 Ultra-D setup

hitechjb1 said:

From table 1, 2 and 3,
- 1 MHz of memory frequency at 2-2-2-5 is about 60-80 points in 3Dmark01 (at 252-254 MHz)
- 1 cycle of tRCD is about 200-300 points in 3Dmark01 at 311 MHz

Click to expand...

5. 317-318 MHz 2.5-4-4-8 1T (CPU/9) may be better than 308-311 MHz 2.5-3-3-7 1T (CPU/9) in 3dmark01.


Memory frequency and latency tradeoff

How much frequency increase is needed to break-even with low latency

Testing UTT and TCCD memory modules in Winchester and DFI NF4 Ultra-D setup

SuperPI 32M is a relative "quick" way to test CPU and system speed and stability

Wow, awesome stuff man. Hey how come you show 2 or 3 runs with the same stuff, and then others only one?

Also, I'm glad you pointed this out:

"Memory at 308-311 MHz 2.5-3-3-7 1T (TCCD) with CPU/9 is about tie with 252-255 MHz 2.0-2-2-5 1T (UTT) with CPU/11 (higher efficiency) in 3dmark01."

I'm glad b/c I'm running my rig at 295x9 with the UTT at 242mhz 2-2-2-5, and a lotta ppl tell me "Oh, you'd do better with 1:1 TCCD 2.5-3-3-7 at that 295..." Since they're about the same I'm already selling my PDP Patriot. I was thinking of buying some expensive G.Skill, but I guess it really won't be worth the money. Of course it's great stuff, but the money to performance will be a very unappealing ratio for the upgrade.

HI ALL,

ANyone know if its better to buy the Venice 3000, 3200, 3500. or 3800?

Which do you guys suggest purchasing??

No Venice is out yet, how can we possibly know?!?

I got a Venice 3800... It's running 400x12 right now.

Sli Bridge For Mod Dfi_nf4 Ultra D

HI All,

I have sli bridge for sale, if anyone is interested. I have gigabyte sli bridge and MSI sli bridge. I will sell for a reasonable price. I don't want to jack anyone.

No selling outside the classifieds.

Let's keep this thread mainly for technical questions, discussions, findings and results.


Update on Cool and Quiet (CnQ) Testing

Latest:
In bios (03/10/05)
- K8 Cool&Quiet Support to "AUTO"
- Cool&Quiet MAX FID to "AUTO" or 9 (9 for Winchester 3000+, 10 for 3200+, etc)
- CPU VID Startup Value to "StartUp"
- CPU VID Control to "AUTO" (CnQ also works if setting to a fixed VID, but voltage won't be able to change from idle/light to heavy load, wasting power and not as cool when idle)
- CPU VID Special Control to "Above VID * 104%" (or "Above VID * 110%", etc)
- use latest AMD processor driver
- XP power management set to "minimal" (in XP Control Panel)

Under these settings, it has been reported that Cool&Quiet works with some CPU overclocked to 300 MHz x 9 = 2700 MHz.
Cool&Quite is still not working to full overclocking of 2700+ MHz in this setup. It worked up to 270 MHz x 9 = 2430 MHz.
(It works now to 293 MHz, see below).
System booted up with x5 multiplier, CPU at 270 MHz x 5 = 1350 MHz, CPU VDD at ~1.1 V
Under load, CPU VDD was increased automatically to 1.40 V * 104% ~ 1.39 V (board undervolted) and FID increased automatically to x9 (270 x 9 = 2430 MHz).
System won't boot if HTT is raised higher than ~275 MHz. Also if changing HTT using ClockGen above 275 MHz, system hanged.

04/10/05:
Fixing LDT multiplier to x3 instead of AUTO.
System boots fine and CnQ works up to 293 MHz HTT (increased from 270 MHz), and CPU at 293 x 9 = 2637 MHz.
Prime95 is stable and CnQ can switch between run and stop of Prime95.

Still, with CnQ, the level of overclocking is up to only 293 MHz, much lower than the highest level of this setup (300 - 320 MHz HTT and 2.8+ GHz CPU). One remaining question is that whether the current PSU (Antec Ture 550 with 20 pin connector) limits CnQ to higher HTT and CPU level (to be resolved).

Between idle/light load to full load,
voltage swing (from 1.1 V to 1.4 V) * 1.04 = 1.4 / 1.1 = 1.27
frequency swing (HTT x 5 to HTT x 9) = 9 / 5 = 1.8
current swing between idle and full load = (1.4 / 1.1) * (9 / 5) = 2.29

For system that cannot be overclocked to such level, maybe CnQ can also reach the highest level of overclocking of those systems.

CPU VDD is switching between 1.1 V to 1.4 V (times 104 or 110%), CPU frequency from 293x5 to 293x9 MHz, a difference of 80%.
Idle/light load CPU and system temperature drop about 5 C as VDD is decreased by about 0.3 V using CnQ.

CnQ is cool .


The original post about CnQ Testing is updated with this information.
Cool and Quiet (CnQ) Testing

Post #242 is awesome. I'm going to make it a Favorite.

Yeah, hitechjb1, reading through your threads is like reading an encyclopedia!!! I just keep clicking on links thinking, "Oh, that looks interesting..." and before I know it, I have sooo many windows open I can't even remember my name anymore!!!
LOL

Keep up the good work. I especially appreciate the comparison between UTT and TCCD

OC NOOBIE said:

HI All,

I have sli bridge for sale, if anyone is interested. I have gigabyte sli bridge and MSI sli bridge. I will sell for a reasonable price. I don't want to jack anyone.

Click to expand...

I'm confused, even though you're not aloud to sell outside classifieds, why did you pick this thread to sell the SLI bridge anyways? It doesn't make sense.

Probably because there are a lot of people subscribed and watching it
Although an SLI bridge is something that some people may want to get a hold of, selling outside the classifieds and trying to hide it in an unrelated thread is kinda lame...

I wish I could find where he's selling them anyways. I modded my Ultra-D and need one sometime or another. Do you know if any stores or any place sells them?

And back on topic: gvblake, I totally agree with you about the encyclopedia thing. Damn, hitechjb1, you are awesome!

http://www.ocforums.com/showthread.php?p=3637644#post3637644
can you please help me hitech!

noktekniq said:

http://www.ocforums.com/showthread.php?p=3637644#post3637644
can you please help me hitech!

Click to expand...

The settings that I have been using, testing, experimenting, evolve over time, ..., there is no single setting to describe all. The basic techniques to achieve such overclock and details are described in the various posts with links listed in the first post.

If the system is run at stock or very slightly above stock, then directly copying the various settings would work generally. But if one would want to push to the last 1% or 0.1% of the maximum potential of various components (CPU, memory, chipset, GPU, ...), even using the components of SAME model, manufacturing date, the various settings and optimization of the entire system would not be exactly the same due to the random distribuation of individual components. Even two chips cut from the same silicon wafer next to each other may behave differently. Memory modules made even from DRAM chips of the same batch may not clock to the same high frequency and low timing.

I would suggest using others settings as reference and guideline only, try to understand the essence of such settings and tune the individual system. Careful observation and systematic searching, patenice are key to obtaining good overclocking results, and developing experience and skill over time, ....

I made some comments in your above thread, but that cannot be replaced by the actual testing and systematic tunning of the actual system.


Some posts that may be useful are relisted here:

Hardware, testing and results
DFI LanParty UT Nforce4 Ultra-D with Winchester
Hardwares and testing
Summary of results
DRAM Bios Setting (for TCCD) bios 02/17/05
DRAM Bios Setting (for TCCD) bios 03/10/05 @ 315 MHz

Overclocking techniques
Overclocking setting for various bus frequencies (with a memory divider table)
Some overclocking scenarios for 939 Winchester/Venice/San Diego
How to determine an upper limit for CPU and memory
Bare minimum voltage at maximal overclocking
Auto-adjusting of the HyperThread system bus frequency using AUTO LDT multiplier

Memory optimization
Memory frequency and latency tradeoff
How much frequency increase is needed to break-even with low latency
Testing UTT and TCCD memory modules in Winchester and DFI NF4 Ultra-D setup
Some results about comparing memory frequency, memory timing, memory divider

Stability testing
SuperPI 32M is a relative "quick" way to test CPU and system speed and stability
Stability testing using memtest, SuperPI (32M) and Prime95
Stability testing using Stress Prime 2004 (SP2004)

A64 CPU related
A64 940, 754, 939 CPU Models, OPN code, rating
Low PR 90 nm 939 Winchester (Sept 2004)

thanks hitech for telling me that. i would appreciate it if you can help me determine the best setup for my bios to reach goood overclocking speeds. if you need addition info on my setup please ask and ill try my best to give you the information. btw i'm wishing to hit over 2.7 which is my goal. currently at 2.55ghz STABLE anything above that and the parts start acting up!

Running the computer Cool and Quiet

Summer is coming (in the northern hemisphere), in general the highest stable overclocking is about 100 – 200 MHz lower in warm summer days as the room temperature and hence the system ambient temperature is usually higher. E.g. an increase of temperature by about 10 C, it would result in a drop of highest frequency at 2.5 GHz level of about 0.4%/C * 10 C * 2500 MHz = 100 MHz.

Using the Winchester 3000+ and the DFI LP UT NF4 Ultra-D as an example, here is shown how to run the system cooler and quieter (in summer) by using Cool and Quiet and by lowering the max overclocking by 100 – 200 MHz using a lower voltage (down to 1.1 V) and low RPM fans (all fans under 2800 RPM). Actually the fan does not even spin when the system is idle or under light load.

DFI LP UT NF4 Ultra-D
Winchester 3000+
G. Skill TCCD 4400 LE
CPU fan: 90 mm Enermax UC-9FAB adjustable fan (1200 - 2500 rpm)
chipset fan (from motherboard stock chipset fan)
system fan: low rpm fan (~2500 rpm)

Highest overclocking (with Cool and Quiet disabled)
2.95 GHz (328 MHz x 9) 1.60 V 3.0-5-5-10 1T Windows XP boot, Sandra CPU
2.90 GHz (322 MHz x 9) 1.60 V 3.0-5-5-10 1T SuperPI 1M stable
2.85 GHz (317 MHz x 9) 1.55 V 2.5-4-3-7 1T SuperPI 32M, 3dmark 01/03/05 stable
2.73 GHz (303 MHz x 9) 1.50 V 2.5-3-3-7 1T Prime95 stable


Using Cool and Quiet (CnQ), highest overclocking = 2.64 GHz (293 MHz x 9)

........................ Idle/Light load ... Load (Prime95 stable)
CPU frequency .......... 1.47 GHz .......... 2.64 GHz
........................ 293 MHz x 5 ....... 293 MHz x 9
CPU VDD ................ 1.10 V ............ 1.38 V
Memory ................. 2.5-3-3-7 1T ...... 2.5-3-3-7 1T

Ambient room temperature = 22 C
CPU temperature ........ 27 - 30 C ......... 39 - 41 C
System temperature ..... 29 - 30 C ......... 36 C
Chipset temperature .... 37 – 39 C ......... 37 – 39 C
CPU fan ................ 0 RPM ............. ~ 2200 RPM
System fan ............. 2600 – 2700 RPM ... 2600 – 2700 RPM
Chipset fan ............ 2700 – 2800 RPM ... 2700 – 2800 RPM

Ambient room temperature = 17 C
CPU temperature ........ 26 - 27 C ......... 37 - 38 C
System temperature ..... 25 - 26 C ......... 31 C
Chipset temperature .... 33 – 34 C ......... 34 – 35 C
CPU fan ................ 0 RPM ............. ~ 2500 RPM (faster fan setting)
System fan ............. 2600 – 2700 RPM ... 2600 – 2700 RPM
Chipset fan ............ 2000 – 2100 RPM ... 2200 – 2300 RPM


Between idle/light load to full load,
voltage swing (from 1.1 V to 1.40 V) * 1.04 = 1.40 / 1.1 = 1.27
frequency swing (HTT x 5 to HTT x 9) = 9 / 5 = 1.8
current swing between idle and full load = (1.40 / 1.1) * (9 / 5) = 2.29

CPU VDD is switching between 1.1 V to 1.38 V, CPU frequency from 293x5 to 293x9 MHz, a difference of 80%.
Idle/light load CPU and system temperature drop about 5 C as VDD is decreased by about 0.3 V using CnQ.


Cool and Quiet bios setting

- K8 Cool&Quiet Support ..... "AUTO"
- Cool&Quiet MAX FID ........ "AUTO" or 9 (9 for 3000+, 10 for 3200+, etc)
- CPU VID Startup Value ..... "StartUp"
- CPU VID Control ........... "AUTO"
- CPU VID Special Control ... "Above VID * 104%" (or "Above VID * 110%", etc)

- CnQ also works if setting to a fixed VID, but voltage won't be able to change from idle/light to heavy load, wasting power and not as cool when idle
- use latest AMD processor driver
- XP power management set to "minimal" (in XP Control Panel)


In the figures,
CPU: temp1, fan1
system: temp2, fan2
chipset: temp3, fan3
Vcore1: CPU VDD
Vcore2: CPU LDT Interface

Ambient room temperature = 22 C






Cool and Quiet (CnQ) Testing

Continuation from last post:

Ambient room temperature = 17 C

The figure shows the system running Prime95 started from idle, and the transient response of temperature and fan speed of CPU, system, chipset. The time scale of the temperature chart is about 12 minutes.




System run into 40 minutes of Prime95. The figure shows how the temperature and fan speed of CPU, system, chipset vary.

CPU temperature within 2 - 3 C.
System temperature about constant.
Chipset temperature within 1C




After 47+ minutes, Prime95 was stopped. The figure shows how the temperature, fan speed of the CPU, system, chipset decrease, as well as the CPU VDD and LDT interface voltage. Note that under Cool and Quiet,
- the CPU fan stopped spinning under light load
- CPU VDD dropped from 1.4 V to 1.1 V
- CPU frequency dropped from 292 MHz x 9 to 292 MHz x 5.