2000MHz Memory On AMD W/ Phenom II X6 1090T & 890FXA-UD5 *Part 3*


Hey guys.

I managed to hit another 32m stable result, and was able to lower the tRAS back to 15. 32m stable means that this ram speed is stable for almost any benchmark and most daily computing. However it is not 100% until it has passed at least 3 hours of blend on Prime 95. This is to ensure the ram and NB are fully stressed and simulate the maximum stress these subsystems will ever recieve.

My next step is to conquer Prime 95 at these speeds. So far I’ve been out of luck. I’ve tried loosening the sub timings, tRCD, tRAS, loosening to 7-9-7. Nothing has worked so far. Most of the attempts end up failing within 5-10 minutes. I might try fooling around with overvolting the ram a bit more, and pointing a fan at them tomorrow and see what I can do.

Stay tuned! And thanks everyone that has viewed my blog so far!

About these ads

Phenom II IMC & Ram Overclocking Guide


Introduction

Over the recent months I have watched and observed as the Phenom II CPU found its niche within various enthusiast communities and online forums. Users have been provided with an excellent, all around, 45nm AMD quadcore.

However, there is one factor that people still find confusing and unclear. Overclocking the Phenom II. Why are people so confused?  One reason may be because they add a lot of new variables and overclock differently than the competitors chip. Some may choose to see Phenom II overclocking as a headache, or a fresh challenge in their hands. I choose to see it as a fun challenge.

Now, one area that people have mass amounts of problems is Ram/Memory stability and overclocking, as well as what role the IMC (NB) plays in affecting ram and performance.

The IMC

The IMC links the CPU to the memory in a system. In the case of the Phenom II AM3 series, DDR2 or DDR3 memory is supported. One important performance factor that people neglect when applying daily overclocks to their Phenom II systems is that the faster the IMC speed is, the better your memory will perform. Additionally, you have to take into account that overclocking the IMC may require voltage increases on the CPU/NB. This may add unwanted heat which happens to be the Phenom II’s arch-enemy. While overclocking though, always keep in mind that the IMC is a very touchy factor in an overclock. If it’s not working out for you, you perhaps need to try different combinations/ratios. When stress-testing the IMC, the most preferable method is Prime 95 blend test for at least 2-3 hours.

The Ram

The memory within an AM2+ or AM3 configuration acts and requires different tuning than your average Intel setup. The RAM ties strongly in with the IMC. If the ram is unstable it can cause instabilities in the IMC and vice versa. Ram on this platform tends to like lower memory frequencies and timings. When tuning ram, take into account every setting you can find in your BIOS. I mean it. Experiment. It may seem like a lot to take in, but hey, grab a coffee, a pad of paper, and sit down for 2 hours and do some testing. The second important portion of RAM clocking is that you must know what type of IC’s your RAM uses. This already will put you in a great position to fine tune the ram. Plenty of online resources can help you identify your IC’s if you do not already know. Once again, after clocking your ram, 100% stability can be found using Prime 95 linked above. Stability testing for benchmarking overclocks can be done in SuperPi 32M.

The CPU Cores

Since the focus of this article is geared towards Ram and IMC clocking, I will be quick on my CPU Cores description.

Phenom II CPUs are usually very intolerant to high voltages when using air setups. Be aware that more voltage may decrease stability going past 1.5V on quad core versions of Phenom II. Also remember that these cores love cold temperatures and scale brilliantly with it. This can be seen in my Canadian Winter blog series, as I will soon start to compare the scaling from normal air temperatures to very cold air temperatures. Even without winter air, every degree that you can lower your temperatures in an air setup is worth it towards overclocking.

Pulling It Together

This creates a 3 component chain wherein lies the secret to stability and performance. You have to realize that if any component in this chain (CPU, NB, Ram) happens to be unstable, the chain will not be complete and you will never see stability. In order to maintain stability, you have to find a sweet spot for each variable according to your setup. There are a plethora of ways you can approach this. However, I discovered that some specific methods are more efficient than others.

The Overclocking – Guide

Starting With The RAM

It is unarguable that even out of the box, with bare bone stock settings in a BIOS, memory will cause the most headaches for the AM3 setup. For example, prior to Phenom II X6 chips which include a brand new IMC revision, you could not take a standard 1800MHz 9-9-9-24 set of DDR3 and get it to run at it’s rated speeds without doing some tweaking to settings in the BIOS . Even sticks that can in fact run at rated speeds out of the box on an AM3 platform can be a pita for stability. This is in contrast to the CPU and IMC where on stock settings, they are supposed to operate flawlessly without any rifts.

1) To begin your overclock start with a default ram clock and timings. For example: 1333MHz 6-8-6-24. This can also depend on the stock specs of your ram to begin with. That is always a good starting point (excluding stock 1800MHz and 2000MHz sticks, as it is unlikely you will get them to run properly at first starting at such a high frequency).

2) You can start speeding up your ram in a variety of ways. By increasing clock speed (Through dividers and HT Ref Clock), tightening timings (adjusting latency timings, and sub-timings in the BIOS), or combining both for the ultimate overclock. Start in small increments on the frequency and timings. Remember that the 4 major ram timings (CAS, tRCD,tRP, tRAS) are very dependent on what IC’s your ram has, and changing these more then one value at a time can cause instant instabilities requiring a CMOS reset. Also remember that it is contradictory to loosen all timings significantly to increase frequency and vice versa. What increase in speed will there be if you are just trading off one thing for the other? However fiddling around with some sub-timings and loosening can help stabilize higher frequency overclocks, while minimally affecting performance. Generally, the most important part of ram clocking is getting familiar with your ram, and what settings greatly affect performance and stability. This is a lengthy process and there is a lot of rebooting involved. General ranges to shoot for would be: 1333MHz (CL 5 or 6), 1600MHz (CL 6 or 7), 1800MHz (CL 6-8), and 2000MHz (CL 7+). What range you want to fall into also depends on your ram, the quality of the IC’s and your ability to tune the timings. Remember to always experiment as much as you can and play around with everything available!

Here are a few sources that list different DDR3 IC’s:

http://www.jmax-hardware.com/bdds/ddr3.html

http://ramlist.i4memory.com/

http://www.hardwareluxx.de/community/f13/hardwareluxx-ram-datenbank-ddr3-404293.html#post7147707

3) With each small change you make it is important to know if you are stable or not. In windows you can use Super Pi 32m as a preliminary stability test. This will not gauge 100% stability, but it will help you figure out what range you will be able to overclock within. Passing 32m also means that you are stable for most benchmarks. If you cannot pass 32m, then your ram is not remotely stable, and you can then adjust your settings accordingly. For full stability testing for 24/7 overclocks, utilize Prime 95 for at least 3 hours. During any stability test, it is important to eliminate the risk of cpu cores and IMC (NB Frequency) causing any stability problems. This is to ensure that if there are instabilities, they are occurring within your ram, not within your processor cores or memory controller. So make sure that the NB and CPU are as close as possible to stock settings prior to testing. Another thing to note: the more ram slots you have occupied with modules, the more stress it places on the IMC. This can make it harder to overclock/stabilize.

4) If you do find instabilities in your overclock there are a few things you can do to help stabilize them. First things first. Do not go crazy with the ram voltage. In moderation though, it can help. Recent DDR3 modules cannot take too much voltage on standard cooling. It is common for a lot of ram to come stock at 1.65v. Usually this ram should not exceed 1.75v for risk of degradation or damage to the ram. However some IC’s are known to like voltage more then others. Micron D9’s come stock at 1.8v usually and love voltage. Now that that’s out of the way, here are a few tips for stabilization. Increasing the CPU-NB voltage slightly is always helpful, regardless of whether the NB Frequency is at stock. NB Voltage is a setting in the BIOS that refers to the physical northbridge on the motherboard. Do not confuse this with the IMC. In some cases, increasing this voltage in moderation can help stabilize high ram overclocks. Another tip is to ensure that both of your ram sticks are in slots 3&4. On some Gigabyte 890FX boards these slots offer higher headroom for ram speeds. Keep the ram cool, pointing an extra fan toward it can always help. The final tip is to adjust drive strengths, which I will talk about in depth in the next step.

5) In the BIOS, there is an advanced set of ram settings. These are called drive strengths. Essentially you can use these to stabilize the ram further. It is very hard to understand which changes in drive strength values are useful unless you utilize memtest. You need to boot from this program off a CD or USB before going into windows. When changing the value of a drive strength always remember to do it for both sticks of ram (sometimes labeled DCT0&DCT1, or A&B). Only changing one value at a time, run a round of tests in memtest. You want to change drive strengths until memtest gives you as little errors as possible on the test suite. Once done so, you have essentially helped stabilize the ram or have created more headroom for overclocking. Remember that you cannot do this if you are getting preliminary tests of hundreds or thousands of errors in memtest. You need to begin with a handful of errors, and this is what you must use as your baseline value that you will start tuning from.

Adding In The Northbridge Overclock (IMC)

Once you have a proven ram overclock that is stable, it is time to work on the IMC. The IMC can vastly increase performance of the memory subsystem in all aspects. It allows a memory overclock to work to its full potential. Overclocking the memory controller is very similar to overclocking the CPU cores. It is much easier to overclock and understand then ram overclocking, as there are less variables in play.

1) When overclocking the IMC you need to know what will affect it’s  headroom. On the Phenom II X4, X3 and X2 chips, the NB Frequency has a max stable range of 2600-3000MHz depending on your chip. For 24/7 overclocks this frequency usually falls around ~2700MHz. This will require voltage increases in the CPU-NB option in the BIOS.  However, Phenom II X6 chips are slightly different. Their max 24/7 range is around 2800MHz-3000MHz. They also require less voltage then previous chips.

2) Start by increasing the NB multiplier till you hit around 2300MHz-2500MHz. Try stability testing this in Prime 95 combined with your ram overclock. Do not adjust the voltage to the IMC just yet. If this passes, then move up a notch. Go until the IMC causes Prime 95 to crash, and then start increasing voltage in moderate increments on the CPU-NB option. When overclocking take into account that the HT Ref Clock (which also controls your ram and cpu frequency) will become the base that is multiplied by the NB multiplier to get your NB frequency. This means that each multi increase will have a bigger increment as opposed to if the HT Ref Clock was left at 200.

3) For CPU-NB voltages, do not exceed 1.5v. After that the IMC won’t really scale well. Also note that each time you add voltage to the CPU-NB, you are adding to the overall heat output of the chip. So this is something to watch under air cooling, as it could hinder your CPU core overclock that you will later add in. The rule of keeping the chip cool also applies to the NB. A cooler chip could equal a higher overclock headroom.

4) When stability testing your final NB overclock, run Prime 95 for at least 3 hours on the blend setting. This is to ensure that the ram and NB are both stressed to the maximum during the 512K FFT iterations. You may wish to run this longer then 3 hours if you like, but that is more stress then any real-world application can offer. After this is stable, then you have yourself a final overclock of the IMC and Ram.

The Final Step

For the final step, you would add in your CPU core overclock. A great guide to check out for overclocking the CPU is my previous guide here. All this information should help in completing your final stable, 24/7 overclock. Here’s an example (it was only 2 hours because of time constraints):

I hope you guys enjoyed my second guide, stay tuned for more in the future.

2000MHz Memory On AMD W/ Phenom II X6 1090T & 890FXA-UD5 *Continued*


Hey guys,

I did some further testing at 2000MHz ram speeds. I discovered that my settings for my 1m Super Pi run were not entirely stable for 32m testing. So I revised a few sub-timings and was successfully able to run 32m. This involved loosening the tRC and tRAS. Keep in mind that this run was done for stability, not to break 32m time records.

My Setup:
AMD Phenom II X6 1090T
Mushkin Ridgeback 2X2GB DDR3 Stock @ 1600MHz 6-8-6-24-1T
Gigabyte 890FXA-UD5 W/ F3 BIOS
Ram in slots 3&4 (specified to support the higher ram speeds rather then slots 1&2)

As you can see, I ran the 1090T at a comfortable 4GHz. This was to avoid any processor instabilities getting confused with potential ram instabilities.  The IMC was pushed to 3000MHz. This was to reflect how this 2000MHz ram overclock could work together in a stable fashion alongside a stressed IMC.

I will continue to try and improve the timings and speeds of this ram whilst maintaining stability in Super Pi 32m. I will try and find a way to bring the tRC timing back down for increased performance.

Stay Tuned

Canadian Winter – Winter Air Overclocking W/ Phenom II X4 955 C2


My two part series about overclocking with subzero winter air. I will be adding on to this next winter.

Part 1

Back for my most recent blog post after a prolonged break from the OC scene, I bring to you some fun winter air testing. I used the same setup as previously. This mainly consists of my trusty Phenom II 955 X4 BE, Gigabyte MA790FXT-UD5P, and Ballistix D9GTS 2GB DDR3 set.

Ahh, the familiar feeling of Canadian winter. A dreaded feeling for most, but hey, I’m an overclocker……and I like cold weather for one purpose….

As some of you may know, I had been struggling to break the SuperPi 1M 16 second barrier using the Gigabyte MA790FXT-UD5P (Least efficient board in 1m testing)  configured in an air setup. However, this is no typical air setup…..but it finally got the job done.  The air ambient was around -8C to -10C.

There are chips that are capable of doing this without such cold ambient air temperatures. Mine is obviously not one of them as seen in this test. The Gigabyte board requires at least around 4.3GHz core speed to break 16 seconds whereas many other board are capable of this at 4.26-4.28GHz.

After that, I decided to do a little CineBench run. This was an untuned run. Also note it is in 32-bit windows.

Last but not least, I saved a validation of a 4.452GHz suicide. However, being away from all the updates, I neglected the fact that my CPU-Z was not up to date. This resulted in my beloved 4.452GHz dump to go to waste. Ah well.

Part 2

After enjoying opening gifts and the atmosphere of the holidays I had almost forgotten about publishing a second blog. Back for the second part, here is Canadian Winter. To see the first part, click here.

So at last, here is my second winter air blog. This time, I focused more on clocking my ram up to see how cold affects that. I used the same setup as previously. This mainly consists of my trusty Phenom II 955 X4 BE, Gigabyte MA790FXT-UD5P, and Ballistix D9GTS 2GB DDR3 set.

Real quick I just want to share what it looks like right outside my house on these chilly winter days, and a few images of my setup and how I utilize the cold air.

As can be seen, I am simply placing my rig in my window sill, and letting it pull the freezing air through the heatsink.

Last blog I displayed how the cold allowed me to break the SuperPi 1M barrier of 16 seconds through a large gain in cpu speed. This time, I decided to see how I could keep the CPU speed slightly lower, while taking advantage of gains in memory, and northbridge clocks. This time my ambient temperature was only just around -5C to -8C.

First, here is the memory speed I reached through testing. I kept timings the same, as I have them set perfectly to my liking. So I ramped up the frequency. I ended with a bootable NB of 3080MHz, and Ram frequency of 1760MHz.

I got lucky and it just so happened that this Ram/NB clock was stable enough for 1M runs. Note that I have lowered the cpu speed since last blog’s run and still attained below 16 seconds.


I also attempted for 4.452GHz again as I failed last time. I didn’t get a validation, but I did get a snap shot.

Shortly after this was the fate of attempting a validation.

Another thing I played around with, on the past knowledge from Chew* was the voltage tolerance of my chip. He claimed that there has never been a quad to boot past 1.6-1.65V. It seems he is right. Even with freezing temperatures, I could not boot at above that threshold.

I also tried attempting some overclocking using the same method on a day that struck with -20C weather. However, I was having issues. The colder the system got, the higher my CPU temperature went. I couldn’t exactly figure it out. With some advice from Chew* and Aaron Schradin I have found it may be due to either freezing of fluid/material within the heatpipes on my TRUE 120 which stops heat dissipation dead in it’s tracks, or attributed to VRM’s getting too cold due to the ambient temps which causes fluctuation in voltages. To combat this issue I tried even insulating my board with putty! (Yes, a first for an air system, I will have pictures of this for my next blog). However this even failed to curb the issue. I finally attempted to build a cardboard “ventilation system”, which would be used to vent cold air only to the CPU. However once again failing to work. This leads me to believe that the problem lies in fact in the heatpipes of my cooler.

This concludes my second winter air blog. I hope you guys enjoyed reading it. I will be back to post more in the future.

|Slappa’s Slushbox| – |Phenom II X6 1090T Overclocked In Ice|


|Slappa’s Slushbox| @ overclock.net
|Slappa’s Slushbox| @ xtremesystems.org
|Slappa’s Slushbox| @ techreaction.net , |Slappa’s Slushbox #2| @ techreaction.net

Video

Part 1

Long story short. Today I got pretty bored. I’ve been wanting to put my 1090T under dry ice for quite some time now. However, my bench partner has been busy with work so I could not obtain the dry ice pot.

So I was thinking of ways I could cool down my processor, in order to play around with my new 6 core. Since I own a Corsair H50 AIO Watercooling kit, I decided to have some fun.

I hung the radiator in an old lunch cooler and filled it to the brim with ice. I also threw a bit of water in for more cooling surface area. I figured this could get my processor 5°C or below. I was right. Yes, the slushbox idea is not a brand new innovative one, but I wanted to try my chip under something else then regular cooling for now.

So I pushed my processor to frequencies I only previously thought could be attained on dry ice (at least with my deneb’s). It seems these new Thuban chips pack quite the punch when it comes to running up the frequency.

Anyways. Enough with the chit chat. I threw together another youtube short just like the good old days when I played with my denebs. So here you are: http://www.youtube.com/watch?v=Wwb1K3_G7Mg

Here are all the screenshot results and validations:

4250MHz Prime 95 Stable 1 Hour

I wanted to see how the slushbox worked under a real stress and if the chip would stay stable. The slushbox doesn’t deal with loads too well, but is still kept below 32C at all times. I believe if I had a good enough water setup, I could attain this stable speed no problem without any slushbox at all.

SuperPi 1m @ 4662MHz

I ramped up SuperPi to see what this chip can do. On my previous deneb chips, SuperPi 1m was my bread and butter, my favourite benchmark to run. Since SuperPi is a single threaded application, I imagine if I disabled some cores or ramped up only one core I could get better results.

SuperPi 32m @ 4500MHz                                                        SuperPi 32m @ 4571MHz

My first run was kind of on the safe side. To avoid a crash. However for second run I ramped up the processor a bit. This was the max I could go while still using all 6 cores. Once again, being single threaded, disabling cores might be able to help my clocking.

wPrime 32m @ 4500MHz

wPrime 1024m @ 4250MHz                                    wPrime  1024M @ 4409MHz

PiFast @ 4500MHz                                                             PiFast @ 4553MHz

Cinebench 11.5 @ 4250MHz                                  Cinebench 11.5 @ 4427MHz

3DMark Vantage @ 4427MHz

SiSoft Sandra @ 4427MHz

CPUZ SS @ 4730MHz – Validation

In the video I attained a speed of 4.75GHz, however it was not stable for validation nor screenshot. It was only a taken as a video clip. The cpu did not last very long at 4.75GHz and crashed soon after.

Part 2

Hey guys. I did some more tuning, got more results. So I am throwing together my second slushbox blog. To see the first one, click here.

I tried to place my rad in the cooler this time so that the ice covered more surface area. This resulted in idle temps of 0-1C. So i pushed my Phenom II X6 1090T even further.  Here goes my second set of results.

4.806GHz All 6 Cores @ 1.648Vcore 1C idle – Validation

It took me an hour to get this validation. I could only previously validate 4.73GHz. I needed to reposition my rad for better surface area of ice. Then I had to wait for my temperature to drop to 0-1C before I made the final push. I couldn’t get any further without going above 1.648Vcore. Here we have it, 4.806GHz on the slushbox. Can’t wait to see what this can do with DICE.

4.902GHz Core #1 @1.648Vcore 1C idle

I chose 3 of my chips cores for this result, meaning I had to disable 3 in the bios. I had previously tested all 6 individually, and the best clockers were cores #0,#1, and #4. The best of the three was core #1, the one which I decided to push for this screeny.

SuperPi 1M – 4.7GHz @ 1.616Vcore – 14.555 seconds

This is a very fast result. The upper limit for my cpu in 1m. Again, this result only uses three cores. I believe I could improve on this score by tuning my ram a bit more.


SuperPi 32m – 4.608GHz @ 1.616Vcore – 14 minutes, 53.148 seconds

This is the first time I have ever broken 15 minutes in 32m without using dry ice. This score could also be improved through a little ram tuning.

UPDATE: Low Volt Overclocking

I wanted to see how far this thing would go without using insanely high volts. It’s pretty amazing what a little bit of cold can do.

This was my boot speed. 4GHz using only 1.296Vcore. That is stock for these thuban chips.

4.5GHz only requiring 1.44Vcore. These things LOVE cold!

As always, these 1090T’s follow suit with their x4 brethren and respond well to cold. I can’t wait to throw some DICE on these and see how they scale. I’m looking for 5200MHz+.

Thanks for checking out my second session guys. I hope this was interesting. And I also hope to inspire many others to put their Phenom II’s under cold. You will not be disappointed with the results.

My Return: The Phenom II X6 1090T


After months of focusing on school, family,and  a job, I thought I had lost my aptitude for computers and overclocking.

I hope I wasn’t forgotten :p

However, now I’m back. And I’m back with a new toy:

I’m just playing around with it for the first time. This is done on my system with an H50 cooler. It’s not completely stable as of yet, but I’m working towards a max.

Update

So I did some more tuning. Here is a little mini performance review for you guys.

My test setup was my sig rig. I’m using a Corsair H50 cooler. This is a retail Phenom II X6 1090T.

I overclocked this 6 core sucker to 4GHz @ 1.472Vcore w/ 2800MHz NB, 1600MHz DDR3 @ 6-8-6-24. I will experiment on lowering the Vcore while maintaining the overclock, I was just playing on the safe side of stability. These chips should also not be run any higher than 1.475Vcore if on air cooling as the change in electrical specs of the chip has made them more fragile. (knowledge courtesy of Chew*)

I was too eager to post with results, so I only ran Prime 95 for 30 minutes, but I used a custom 512K FFT test to stress the NB and Ram mostly.

Then I ramped up a whole slew of benchmarks. So here are the numbers:

Now time for some suicide ;)

http://valid.canardpc.com/show_oc.php?id=1283155

And heres the SS

All I can say is this chip is a beast. I will be doing more stable overclocking, maybe shoot for 4.1GHz.

Update

And higher…

http://valid.canardpc.com/show_oc.php?id=1283224

SS:

AMD Phenom II Shatters 7GHz || Light Speed


Hey guys, I’d like to share with you an excellent video summarizing the past few months of achievements that AMD has had with Phenom II. As many of you may know, I was involved in the Austin, Texas event.

Check it out @ Youtube.

Watch it in HD!

Memory Speed – In The Sky


 

This is the ram I discussed way back in this post.

:D

Real Value: G.Skill Pi DDR3 F3-10600 2X1GB Overclocked


Many of you got into overclocking to save value and money on your PC. This is exactly how I got started. Now I am here to present a part of value. The G.Skill Pi DDR3 F3-10600 2X1GB Set.

This set of ram comes with a particularly low price tag. Not only that, it’s speeds are mediocre at stock. Of course this is only a 2GB set of which most enthusiasts would steer away from, but I’m putting this up to show people the overclock ability of this ram and perhaps to present budget-minded gamers a possible future purchase.


Stock Specs:

1333MHz
2X1GB
7-7-7-18
1.6V

I only bought this set because I was looking for a quick DDR3 solution that was cheap and fast. I was not in for what I expected.

Now lets get to the point, keep in mind the stock speeds of this ram:

1704MHz 7-7-7-15-11-1T Stable 32m SuperPi

 

Next up, SuperPi 1m

 

 

Blazing fast speeds for cheapo ram, can’t get much better than that!