Back with the green team for the 2nd time: GTX670.

After getting fed up with all the issues with Crossfire, I decided to ditch the 1GB HD6950 cards that I had, and go back to a single GPU setup. But first, here is a list of issues that I experienced with &the Crossfire setup.

 

Crossfire woes

Gaming:

1. Microstuttering: I was getting a respectable 60fps or more most of the times in games, but it wasn't smooth. I only got to see the difference, when I disabled crossfire just to check. 40fps with a single HD6950 was smoother than 60fps.
2. Driver issues: Multiple times I got flickering issues with BF3 and the only way to go back was to reinstall windows and not install 12.4 drivers. When I disabled Crossfire, it went away. I don't know about you, but for me, disabling crossfire does not seem like a solution.
3. Negative scaling: When I tried Skyrim for the first time, there was negative crossfire scaling. The only way to fix it was to wait for ATI to release a proper driver and disable Crossfire in the meantime. There is a possibility that you will not be able to experience the latest games the way they are meant to be, until a patched driver is released. That's no good!

Heat, noise and power consumption:

Two cards mean higher power consumption. The Gigabyte cards that I bought came with a custom cooler; the famous Windforce3 cooler, and it dumps all the heat right back into the case. That is fine, if you did not have a second card, because the top card gets scorched as well as suffocated. While the max temps did not go above 80C the entire time, I felt, from the hot exhaust, that too much heat was piling up inside the case. The CPU also was running pretty hot when gaming: almost about 5C within stress testing temps. The PSU was also putting out some considerable heat. (the PSU in my case is in the front and you can feel the hot air with your feet) All could be due to the hot air inside the case. My electricity bill was also a bit high ever since I got this crossfire setup.

 

Choosing a replacement single GPU card

 

Everything above compelled me to sell off my two cards and buy a single GPU card. But what would I buy?HD7970 or GTX670? That was not an easy question.

 

Traditionally I have owned more ATI cards than nVidia cards. (In fact, I had owned 8 ATI cards including the HD6950’s that I am about to replace, and just one nVidia card!) I haven't had any unforgettable issue with the ATI drivers in a single card configuration. But a lot of people were saying that they had to ditch their HD7970 card due to driver issues. Probably it was because the architecture was new, and the drivers were premature, if we are talking about 3 months or so ago. But some said they were rock solid. However, majority was in the opinion that nVidia drivers were more stable. But remember, nVidia’s drivers are way too premature for the GTX600 series.

 

While the red team was having troubles with their drivers, grass wasn't any greener on the green side either. The new GTX600 series cards were having some weird microstuttering issues with many games when you enable V-Sync. Microstuttering is a phenomenon that would literally make a game unplayable. I've seen it beforehand with my crossfire setup and it can cause many headaches. So if I bought a GTX670, I would be back to square one. But, I don't use (haven’t used) V-Sync at all, so it is not that applicable to me. The reason why people use V-Sync is to stop the phenomenon called “image tearing” when the FPS exceeds the display’s refresh rate (usually 60Hz). It is more visible with FPS just above the refresh rate. Anyways, this microstuttering issue is definitely a driver issue because an older driver (cannot recall the version) magically doesn’t have it. People are expecting nVidia to fix it in June, but I don’t know where they knew about a June fix.

 

All in all, the drivers from both teams can get really ugly. (I don’t really know why, because why should it be dependent on a game, as long as you conform to the DirectX spec.) But generally, the support for SLI is better. nVidia releases patches faster and their multiGPU technology gives smoother gameplay experience at the same fps. But, again, it doesn't really matter to me because I have made up my mind not to buy a multiGPU configuration ever in my life.

 

So it came down to the feature set. nVidia has brought a range of new features with their new GTX600 series cards such as adaptive v-sync, GPU boost, frame rate limit, GPU transcoding with faster than Intel QuickSync (version1.0) and last but not least, lower power consumption. ATI has much less feature set, and the biggest thing they have are higher GPGPU performance (which I don’t care!) and lower idle power consumption. (useful, because your PC is at idle more often than not - but the difference isn’t night and day)

 

When I asked about this question, "HD7970 or GTX670" on techpowerup forums, most people strongly voted for the HD7970 saying that they overclock higher. But when I checked the prices online, the reference HIS HD7970 was almost same price as a factory overclocked Palit GTX670 Jetstream which performs similar to a stock GTX680. Then they suggested that I should rather go with the non-reference GTX670 because the temps and noise was too high when you overclock a HD7970 with stock cooler. Besides, I did not want to try my luck with the silicone lottery. Reference HD7970 is definitely slower than an overclocked GTX670, so HD7970 has to be overclocked if you need to get the full value for your money. But since I have been very unfortunate with overclocking graphics cards (HD5770 850MHz core couldn’t reach 860MHz, HD5870 875MHz core couldn’t reach 900Mhz, HD6950 870MHz core couldn’t reach 900MHz etc.), I was afraid that I would end up with a dud HD7970. Even if I buy the Palit GTX670 Jetstream and cannot overclock a single bit, I would end up with the faster card for the same price.

The card

So in the end, I went ahead with the Palit GTX670 Jetstream. This was also one of the cheapest cards I could find. Even the reference cards were more expensive. (Sure, the reference card from Palit was a bit cheaper, but that price difference was way smaller than the gains. I was literally getting a GXT680!)

I’ll post back telling my experience with the GXT670 in a few weeks. Until then, drool away. :D

ASRock Z68 Extreme4 Virtu issues resolved

ASRock released a new beta UEFI (L2.05) that fixed the broken Virtu support.

It is back to the way it should be. It does not install itself as a trial version.

Now why can’t they bring Virtu support to mainstream apps? At least to Handbrake?

Silencing the PC (Part1)

Motherboards these days come with all sort of fancy fan controls. Asus Z77 series motherboards have a very intelligent and cool design. Others too have some sort of control.

But ASRock definitely has to improve in that regard. The fan control on my Z68 Extreme4 must be the worst among the current mainstream boards. It does have a lot of fan headers (6 of them) buy not all of them can be controlled via bios (my bad, UEFI) or their AXTU software. The best you can do is, set a value manually and if some temperature threshold is reached, it would run at max speed. Though the funny thing is that it is pretty much same as Asus' design (because ASRock is Asus in a way) but Asus' software does the trick.

Recently I have taken an interest in the projects and articles at silentpcreview.com (SPCR). I quite like the idea of having a silent PC without any rattling going on. Noise factor also gives an impression about the build quality of the PC - low noise being better quality.

Noise is mostly caused by two things in a PC
1. Fans (airflow noise as well as motor noise and vibrations)
2. Hard drive noise (operating noise as well as vibrations)

Since I use an SSD, there is not much hard drive noise because SSD (which makes zero noise) is used more frequently and the hard drives (3 of then) are used less frequently. Thanks to the rubber mounts, there are no vibrations caused by the hard drives.

So in my case, it ultimately comes down to the fans. The case that I am using, the Silverstone Raven RV03, is designed for air cooling. So you cannot do without fans. The arrangement of fans is like this.

1. Two 180mm fans mounted on the floor to suck air in. They are rated for 700~1200RPM. 700RPM is tolerable, but 1200RPM is very noisy. It is wind noise, not motor noise.
2. Two front intake Enermax T.B. Silence 120mm 1000RPM fans mounted in the 5.25 inch drive bays, pushing cool air right into the motherboard's VRM area.
3. One rear exhaust fan, a Scythe Kama Flow2 120mm 1400RPM to take out the hot air around the video cards.
4. One intake fan, same Scythe model part, on the back of the motherboard tray for cooling the drives (all my 4 drives are mounted on the back of the motherboard)
5. One top exhaust fan, a Kama Flow 2 120mm 1900RPM fan, which is also used as the CPU fan on the Antec Kuhler 620 radiator.

In addition to these 7 fans, I have 7 other fans in total inside the case. 6 are on the video cards (3 fans on each) and the fan on the PSU.

So, silencing 14 fans cannot be easy! Some fans, like the one on the PSU cannot even be controller manually. Also, it is best to leave the GPU fans automatically controlled via the driver. At idle they are not that loud.

So it comes down to the 7 fans that I can control. The bottom two 180mm fans are quiet at 700RPM but definitely not silent. ~600RPM would be ideal. The two Enermax fans are silent from 1ft away. They don't really need to be controlled. The other three fans are definitely too loud. I cannot run the fan behind the motherboard at full speed because it would make a weird low pitched noise. There isn't a clean path for air, so it can be the source. CPU fan nor the video card exhaust fan don't need to run at full speed when the PC is idle. (Also I am not entirely happy with CPU fan. It does not feel good enough to run on a radiator. I am thinking of buying a Gentle Typhoon 3000RPM fan. True, it is louder than the existing fan, so I would definitely be controlling it. But knowing that you have the power when the need is there is comforting.)

The conclusion? I need a fan controller. Not an expensive one though. But I need a lot of channels.

Overclocking the Intel Core i7 2600K with ASRock Z68 Extreme4

There are few important things to know about overclocking your Sandybridge CPU.

BCLK:

Unlike the previous generation CPUs, with Sandybridge CPUs you don't have two parameters to change  when increasing the clock speed.

Clock speed = BCLK x Multiplier

Sandybridge CPUs have a default BCLK of 100MHz. You cannot increase it more than few MHz before the system becomes unstable. Some chips are capable of 103MHz or more, but most chip + motherboard
combinations don't like going above 100MHz at all. So the easier way to overclock your SB CPU is by increasing the multiplier. For that, you need to buy an unlocked CPU - denoted by the letter K at the end of the model name (e.g. 2600K)

You can overclock a locked CPU too, if it supports Turbo and if your motherboard allows you to do so. For example, the Core i5 2500 (non-K model) which runs at a nominal speed of 3.3GHz has a max turbo multiplier of 37(when only one core is being used) and you can use that multiplier even when all cores are being used. (again, this depends on the motherboard)

So, the easiest path to walk if you are overclocking, is buying an unlocked CPU. Funny thing is though, overclocking came into existence to extract free performance out of cheap chips, but in this case if you want to extract that free performance, you need to buy the most expensive models as the K chips come with a price premium.

Vcore:

Overclocking a SB chip in theory is pretty straight forward. You increase the multiplier and if it is unstable, you increase the Vcore. Most people say 1.35V is the max safe Vcore for 24/7 operation. Others have different views. It is not written in stone. My advise is to use a Vcore and a overclock that is reasonable because there is always diminishing returns. For example, if your chip does 4.5GHz with 1.3V but for 4.6GHz it requires 1.4V, then it is not worth it IMHO.

You can set Vcore two ways.

1. Fixed Vcore: when you set this, the CPU will run at this Vcore at all times; even when it is idle. You might think then it will use a lot of power even when idle. Not so, if you have all the power saving options turned ON. However, at high multipliers (45 and above), you have to turn off power saving features such as C1E, C3, C6 states to make it stable.

2. Offset mode: when you set this, you will get a reference Vcore for your respective multiplier, and you can add a positive or negative offset Vcore to fine tune it. If you set an offset of zero, it means you are using a Vcore assigned from the default VID curve for that particular CPU. Remember, it is a curve. The VID at 1.6GHz, 3GHz, 4.5GHz are not the same. It is a function of multiplier. Intel can only guarantee normal operation within the CPU’s intended clock speed. When you use offset mode, you can have a lower idle Vcore compared to fixed Vcore mode. Also, if you have Intel Speed Step Technology turned ON, it will not just set 16 or overclocked multiplier, but can also use intermediary multipliers – and the Vcore will be adjusted accordingly. All this is good for power saving. But hardcore overclockers (am definitely not one – I aim for 24/7 clocks) might not care.

One important thing when using offset mode is, when you set a value in UEFI, you cannot see the resulting Vcore for the offset value. You have to reboot first. Also, you need to have power saving settings (C1E, C3, C6 states) enabled when using the offset mode.

Note: Every CPU is different. The VID curve is also different from CPU to CPU. Thus you cannot simply compare offset values of two CPUs.

PLL Voltage:

This is another value you might have to change when you overclock. Increasing it doesn't really affect the overclocking headroom or stability per se, but it can have a sweet spot depending on your setup (lower than stock or higher than stock). Usually you don’t have to play with it unless you are going for 4.8 - 5GHz. Otherwise, you can lower it as much as your motherboard allows you to; as long as you can POST, then it is fine.

PLL Overvolting:

No, this is not the same as above. Increasing the PLL Voltage and enabling PLL overvolting is not the same. I do not know what the difference is, but if you cannot boot at high multipliers (usually 48 or higher), you have to enable this setting.

RAM Voltage:

The recommended RAM voltage for Sandybridge CPUs is 1.50V. It is well advised that you do not use  the old 1.65V RAM, although there aren't any reports of killing the CPU with 1.65V RAM. Remember though that if you have 4-sticks of RAM and are running at high speed (1866, 2133), you might have to increase the VTT.

Basically, these are all the voltages you have to play with. Just keep the rest at AUTO.

Load Line Calibration:

This is another important setting. Intel CPUs have something called a VDroop (Voltage Droop) when you load up your CPU. Actually, VDroop is a design feature. Without it, when the Vcore changes (going from idle to load), there can be positive voltage overshoots which might be high enough to damage the CPU. Also noteworthy is that the amount of VDroop is different from motherboard to motherboard mostly depending on the quality of the components.

Most motherboards have a feature called Load Line Calibration (LLC) to counter this *problem*. The problem with VDroop is, when you set a Vcore in UEFI, in reality it will be much lower than that when you run your stress tests. This goes against Intel's spec and brings back the problem of Voltage overshooting. (who cares, right?) There are different levels of LLC you can set. Different levels lower the droop by different amounts. Usually, one notch below the highest LLC level syncs closely with the Vcore you set in UEFI. Highest LLC would take the load Vcore above the Vcore you set in UEFI.

However, LLC also has a problem for overclockers of Sandybridge CPUs. People are well aware of SB CPUs giving BSODs (code 124) when idle, even though they are very stable at full load. This is due to LLC. When you set a high level of LLC, the idle Vcore drops too low, and then it crashes.
For example my ASRock Z68 Extreme4 board has level 1 to level 5 of LLC, level 1 being the highest level.
1. 4.5GHz @ offset of +0.010 with LLC level 2
  idle = 0.970V
  load = 1.312V
2. 4.5GHz @ offset of +0.085 with LLC disabled or level 5
  idle = 1.075V
  load = 1.312V
You can see there is a significant difference between idle Vcore (in case you didn’t know, 0.1V is a significant voltage when it comes to CPUs). Option 1 will give idle BSODs while 2 will not. But both are fully stable at full load.

Spread Spectrum:

This is not a setting that is helpful for overclocking per se, but if you don’t like your CPU using a BCLK of 99.8MHz when you set 100MHz in UEFI, set it to “disable”. The purpose of this setting is to reduce signal interference from the CPU and other components, so especially if you are not using the onboard audio, there is no real benefit keeping it enabled.

Memory timings:

SB’s performance is hardly affected by the RAM speed. I would just use SPD setting if you have 1600MHz or faster RAM. You can play with the settings, but it is usually not worth the effort you put into it. Just buy some 1600MHz or 1866MHz RAM, and run them at SPD timings.

So that’s pretty much the basics. There are other settings too, such as setting per-core level multipliers (some boards do not support this), current limiting, power limiting, tuning of power circuitry but for a conservative overclock to like 4.5GHz, you don’t have to touch them. Just keep them at AUTO.

The settings I use:

Lucid Virtu broken with L2.03 beta UEFI for ASRock Z68 Extreme4 (non Gen3)

Remember that in the previous post I mentioned that ASRock pulled out the *new* L2.02 beta UEFI from their website? Few days later, they released the L2.03 beta UEFI. Of course, I didn't think twice and upgraded. (Who did you think I was?? XD)

Everything is fine after the update, except for one little thing. They somehow broke the Lucid Virtu License with it. Just because you have a Z68 motherboard, you cannot install the Lucid Virtu software. The motherboard has to have a license to use it. I think it is there in the UEFI microcode. So when you install the Lucid Virtu software, it will check if the motherboard has the license to use it, and it will install
fine. If not, it will install as a 30-day trial. That's exactly what is happening with this UEFI update. I even reinstalled Windows to see if it was something else. No, other people also have reported the problem. Oh
well! I wish I didn't have to reinstall Windows.

As usual, the latest version that lists on Lucid's website is newer than the one listed on ASRock's website. So naturally, you would download the one on Lucid's website (currently, version 114 is the latest. ASRock
lists only version 112). When I tried to install it, it mentioned that this is a 30-day trial. When I install the one on ASRock's website, it would not show that message. But when you go into the ABOUT page in Virtu Control Panel, it would list it as a 30-day trial. Bummer!

This is what it is like now

This is what it should be like (see it is not a trial)

I hope they fix it soon because I use QuickSync with Cyberlink MediaEspresso to convert videos for the iPhone QUICKLY. It is few times faster to encode with QuickSync.

When Lucid Virtu gets installed as a trialware, you cannot change the following things in the control panel.
1. Disable displaying of the Lucid Virtu logo when it is being used

2. Modify the list of supported apps (only the 3 apps shown below are supported)

ASRock updated the UEFI on Z68 Extreme4 (and other Z68 boards) for IvyBrigde, then pulled the update?

Here’s a comparison between 1.60G UEFI and L2.02 beta UEFI.

The makeover actually is same as the one on the Z77 mobos. =)

Main screen:

OC Tweaker screen:

Note: The new UEFI lets you set different multis for different core utilization levels, but it did not work for me. I would just stick to the 4-core multi. Also, when this is enabled, CPU speed shown on the POST screen would be a weird value, like 25GHz or something.

Advanced Screen:

New options

1. Intel Rapid Start Technology

2. Intel Smart Connect Technology

Advanced\CPU Configuration Screen:

Advanced\North Bridge Configuration Screen:

Advanced\South Bridge Configuration Screen:

Advanced\Storage Configuration Screen:

Advanced\Super IO Configuration Screen:

Advanced\ACPI Configuration Screen:

Advanced\USB Configuration Screen:

HW Monitor Screen:

New sections specific to the new UEFI:

Advanced\Intel Rapid Start Technology:

Note: So far I have not been able to get this to work. I can enable it in UEFI. I can use ASRock’s own software to make a hidden partition. But I cannot install the Intel driver. It says “Platform not supported”.

Advanced\Intel Smart Connect Technology

I don’t have anything to test feature with.

 

Anyways, I wonder why they pulled this version. Maybe it is not complete. As far as stability goes, this beta BIOS is top notch. Further, the cold boot speed is almost identical, but warm boot speed is super fast.

Unlocking Gigabyte GV-R695OC-1GD. Possible or what?

Few months back, I bought a couple of Gigabyte's HD6950 1GB OC cards to replace my HD5870 card. I was never a fan of multi GPU setups but I wanted to try it out. Besides, I had to buy something that would last long before the wedding, because I knew I will not be able to spend on toys as much as I used to. (It has nothing to do with getting my wife's consent, but more to do with higher cost of living.)

Anyways, it is a well known secret that (most of) the HD6950 2GB reference cards would unlock the full shaders of a HD6970. The success rates of unlocking 1GB cards is almost nil. These cards being non-reference makes it worse.

But I had my reasons for buying these cards.
1. They were the cheapest HD6950 cards from a known manufacturer.
2. They had a non-reference cooler which performance much better than the reference cooler.
3. They came with a ~10% factory overclock
4. I thought 1GB was enough because I didn't see myself buying a 1600p display or multiple 1080p monitors. (But still there are games that use more than 1GB VRAM even at 1080p)

Besides, the performance gains of unlocking the full shaders yielded in a mere 1-3% FPS boost at same clocks. The core overclock always gave tangible gains, so I picked the guaranteed higher clock speed over
something that might not work at all.

I didn't try hard to find out if these cards would unlock at all. I took the facts for granted, until one day I saw on a thread someone asking if we were able to unlock these cards. Not only he was asking the question, he was referring to reviews posted by people on Newegg saying that they (some) managed to unlock the shaders. It kind of got into my head and I searched everywhere if people were able to unlock the card and especially if there were any guides out there. I knew flashing a HD6970 bios won't work. I had to modify the HD6950 bios - the exact one on my card.

Well, it was not that simple. Most people had to short out two pins on the BIOS chip (pint #3 and #8 iirc), to be able to flash the BIOS. Oh, I forgot to mention that these cards don't have the backup BIOS, so the risk of killing the cards was high. I didn't want to mess with the pins either. The worst that seemed to happen if I could not write to BIOS is that it just won't flash. It wouldn't brick the cards. So there was no risk by trying without the pin short mod.

But from the very first moment, I had problems. I couldn't save the BIOS file from GPU-Z. It would always end up in a BSOD. It could have been anything, but I had to unplug the 2nd card and extract the BIOS. After that, when I opened the BIOS file from RBE, I got an error message that pointed to something was wrong with the overdrive settings.

I forgot to mention one weird thing with this specific card. I think the problem is with this specific firmware. HD69xx series cards have Power Control Settings in the Catalyst Control Center, where it puts an upper limit to the TDP to keep the card from burning out when running stress testing tools like FurMark. You can increase it by 20% and that's it.

Normal cards:

My card:

People increase it when overclocking. But my cards don't have it. I reinstalled drivers and did everything possible to get it to show up, but it never showed up. Something was wrong with the overdrive settings. It could be that Gigabyte fiddled with the firmware to get the factory overclock stable. We'll never know.

Anyways, I was adventurous. I ignored the error and proceeded with modifying the BIOS file. With the heart pounding, went with the flash and I was successful. Both cards unlocked fine. The following are the exact steps I followed.

1. Saved the bios file from gpu-z (5.6)
2. Opened up RBE 1.28. got the error about overdrive thingy, ignored. (wth!)
3. Changed shader count to 6970 (didn't touch anything else)
4. Saved bios and opened through winflash 2.01.13 (the one before latest as per today)
5. Flashed, rebooted, checked GPU-Z

(Some might not be required, but I did so to be extra safe.)

Also note that for some people unlocking doesn’t work at all. For some, they need to short out the BIOS write protection pins. But for few other lucky guys like myself, unlocking is a walk in the park.

After unlocking:

What about the results?

A slight gain in 3DMark11 score. But I don’t have the old results to compare. I remember getting  Nothing notable, but I didn't have to drop the overclock either. Something is better than nothing, eh? In any case, the knowledge gained was valuable IMO.