CPU temps of the Mid 2011 MacBook Air

I know my MacBook Air can get pretty warm, but I never measured it quantitatively. I don't know what people use to measure those in OSX; but luckily I have Windows 10 installed via BootCamp (i.e. natively) and I am familiar with the tools in Windows.

I downloaded the trusty duo: CPU-Z and Realtemp. This is the Mid-2011 MacBook Air from the SandyBridge era. The CPU is a Core i5 2557M, which runs at 1.7GHz with a max boost of 2.7GHz.

I didn't want to run any stress tests straight away. But I ran Windows Update and updated the other software installed in there while monitoring the temps. It was astonishing to see the temps hover around 85-90C with a CPU utilisation of around 50% according to task manager, while hitting a max of 95C. I have no doubt it will throttle if the CPU runs at 100%, and probably burn a hole in the bottom of the aluminum unibody chassis if I attempted to run Prime95. So I didn't. This was good enough proof that the temps were terrible in such a thin chassis. I wonder if dust has something to do with it, because it is a 5.5 year old laptop. It cannot be free of dust.

I wonder if it is worth getting it serviced at the Apple store along with the battery. It won't be cheap @ AUD 189. Honestly, I don't want to spend any money on such an old laptop.

[Article] Intel releases IvyBridge-E CPUs. Better late than never?

In English language, there is an idiom called "better late than never", but I'm not sure if you can call it relevant when it comes to the latest CPU release from Intel - the IvyBridge-E. There are few issues with this release which make it very unexciting.

Firstly, the IvyBridge-E is based on the IvyBridge architecture, which is not what the current generation of mainstream CPUs are based on. The current generation is Haswell and Intel's enthusiast CPU is one generation older than it. When Intel released the SandyBridge-E CPUs, the current generation was SandyBridge. Thus, this enthusiast CPU did not fall behind the mainstream CPUs in any test. It at least performed as fast as the top of mainstream CPUs. (Core i7 2600K back them). But now, the top of the line mainstream CPU is the Haswell Core i7 4770K which is faster than the fastest Enthusiast CPU in certain test - tests that don't make use of all 6 cores. That should not be the case.

Secondly, this is not a new platform. This is just a new CPU, which makes use of the previous X79 platform. That platform was already aging at its launch, and after two years, it is completely outdated. You only get two native SATA III ports, no native USB3.0 ports and the chipset is very power hungry. Of course the motherboard manufacturers are putting out refreshed boards, but they have to make use of the 3rd party controllers to at come close to what the current generation mainstream boards are offering feature wise. How hard was it for them to simply release an updated chipset for IB-E? One might be thankful that they don't have to change the motherboard, but the platform is missing a lot of oomph that came with Haswell, that number (of people) will be very low.

Thirdly, the CPUs don't overclock as well as we expected. IvyBridge didn't overclock well as well, but that because the die wasn't soldered to heatspreader. When delidded, they overclocked as well as SandyBridge. But IB-E's die is soldered. So it doesn't face the heat problem. But for some weird reason, it doesn't overclock as well as we expected. It's definitely not a temperature issue. That 22nm tech is giving any advantage over SandyBridge-E CPUs. However, the power consumption is less than the predecessor. Who cares, right? The second you say that you are going to overclock the CPU, you are giving up the privilege to whine about the power consumption.

Fourthly, you are not getting noticeable performance increase from upgrading to IB-E from SB-E. IB-E's advantage is only about 2-5%, which is even less than what Haswell has yielded. But if you are upgrading from a Nehalem 6-core CPU (such as the Core i7 970, 980x or 990x), then the performance increase is worth it, IMO. It's definitely not worth it for SB-E users, especially without any improvement in the platform. Haswell at least gave a new, better platform.

If you want to read an unbiased review, read the PC Perspective's review over here.

IMHO, Intel should have skipped IvyBridge-E and went for Haswell-E straight away. Or, they should have released it at least at the beginning of this year. It looks like Intel doesn't really care. I'm sure their biggest profits (total profits) come from the mainstream CPUs. While these Extreme Edition CPUs are quick lucrative, the number of enthusiasts living in this world is too few.

[Article] Checking if your CPU overclock is stable

The most fun part of upgrading the CPU is overclocking. Of course, it is purely based on luck. You might get a CPU that can overclock wonderfully with low volts and low temps. And you might get a total piece of junk that needs a lot of volts and doesn’t overclock well. (I usually get that 2nd type of CPUs). They say it is silicon lottery. It’s a gambling man!

Temperature is your biggest enemy. In fact, when it comes to Haswell, temperature is your only enemy. To push the clocks higher, you need to up the volts up and that causes temperature to rise exponentially. Most people will have to max out at 4.2GHz on Haswell because of the heat. It’s mostly because of Intel’s crappy work using thermal paste in between the die and the heat spreader but there are other reasons too. Looks like they don’t want the enthusiasts to play with their silicon. Intel’s being naughty.

Say you have a good cooling solution and a good chip. Seeing those scores go up is nice and all, but making sure that your CPU is stable is very time consuming and pain in the butt.

But it depends. You can cheat a bit out of it if you want. You can either look for absolute stability or real-world stability. Failing in Prime95 doesn’t necessarily mean that you’ll ever get a crash (i.e. BSOD) with anything that you would normally do. I am more towards the absolute stability, but I don’t want to stress test for 24hrs in Prime95 with these summer heat.

Ironically, people have found out that sometimes even though Prime95 or any other popular stress testing app suggests that you have rock solid stability, you get crashes in games. So passing these stress tests doesn’t mean your are absolutely stable either.

So stability in my book means a combination of synthetic and real world tests as follows.

1. Prime95 v27.9 Blend – no errors or BSODs for 6hrs
2. Crysis 3 and/or Battlefield 3 – no crashes for 2hrs
3. Handbrake H.264 encoding (one pass) – no crashes for 2hrs
4. Asus Realbench – no errors or BSODs for 6hrs
5. AIDA64 – no errors or BSODs for 6hrs
6. Intel Burn Test: very high stress level – no errors or BSODs for 10 passes

If you can pass all these, then your overclock is stable. Out of all those, Prime95 is the most stressful. All others can pass, but Prime95 can fail with a higher chance. Most people will just run Intel Burn Test and say that their CPU is stable. But that alone is not good enough. Sure, you can say it is almost stable. I’ve been able to pass Intel Burn Test at 4.3GHz/1.22V but Prime95 would fail within seconds. Prime95 needed 1.24V to go on for 6hrs without a crash.

Ultimately, it is your choice. It’s your CPU, it’s your money paid, and it is you who use it. What if you can clock your CPU to 4.5GHz and get no BSOD and crash with the stuff you do with your PC and it fails all those synthetic stress tests? There is no harm clocking it up to 4.5GHz in that case.

Upgrade SandyBridge to IvyBridge-E this September?

Haswell for desktops was a flop. Now it is time to leave that sad incident behind and move on. Moving on means, looking forward to the next launch. The next launch from Intel is IvyBridge-E. IvyBridge-E CPUs are coming out this summer - September to be precise. That's only couple of months away from now. Fun times!

 

The IvyBridge-E CPUs

 

There are 3 models announced. The Core i7 4820K, 4930K and 4960X. Contrary to SandyBridge-E, all the CPUs are fully unlocked this time around. If you remember, Core i7 3820 was only partially unlocked.

The following is a detailed list of specifications of the upcoming IB-E CPUs. (image courtesy of Xbitlabs article)

 

 

Some more facts

The heat spreader is soldered to the die:

IB-E CPUs will have the die soldered to the heat-spreader. (Well, they have only confirmed it for the 4960X) IB didn't. IB simply had a thermal paste interface between the die and the heat-spreader. This caused IB to become hotter than SB when overclocked. Hence IB-E will have more thermal headroom for overclocking, which means it would be possible to overclock it higher than IB. Having a larger die (i.e. more surface area to dissipate heat) and having no iGPU would also help.

No x89 chipset:

It looks like there is no X89 core-logic coming out with IB-E. Intel's chipset roadmaps show X99 after X79 and X99 is for Haswell-E. The issue with X79 is that it lacks a lot of newer tech, such as native USB3.0 and more than two SATA-III ports. And a die Shrink would have improved thermals as well. This is good news for people who are running a Core i7 3820 because they can upgrade to a 4930K and enjoy a massive performance increase without spending a lot of money. You'll only need a bios upgrade at worst case. People running 3930K's or 3960X's won't see much of an improvement though.

 

So, upgrading to IB-E would be worth it?

 

The question remains. Is IvyBridge-E going to be a worthwhile upgrade from SandyBridge? You can receive a 50% increase in core count, unless you are stupid enough to get the 4820K. And there is 10% or so IPC boost. That would translate to about 60% more performance boost.

You will lose QuickSync support, which is not a big deal because Handbrake doesn't support QuickSync yet.

There will definitely be newer motherboards but sadly with the same old X79 chipset. I prefer having the controllers integrated with the chipset. I always disable the 3rd party controllers because they are not fast enough and I don't need that many ports in the first place.

But for most people, other than for bragging rights, this won't be a necessary upgrade. Games wouldn't benefit much for having two extra cores. But it would, if you are using quad GPUs, because the x79 platform provides 40 PCI-E lanes from the chipset (but this would be CPU in the case of IB-E), which allows you to use Quad GPU setups with hardly any performance drop or lag. Generally speaking, only the video editors will benefit from the added horsepower.

I guess I will wait and see how badly these Japanese shops will price the new CPUs. If I've learned something from the history, the pricing would be really bad. A 3930K costs ¥58,000 these days. That's about $580. Even if Intel releases 4930K at the same price, these bastards would sell them for like ¥70,000. Spending that much money only on the CPU is not going to happen in my book. When you think of it, I'd probably be stuck with my 2600K for eternity. The ugly truth!

Wait! I will have to renew my Internet connection in January. That means, I will get like ¥40,000 or so discount on select items when purchased with the internet connection. I hope the 4930K will be in that group of select items. :)

Want to overclock your CPU? Here are some tips to make it stable.

Remember those days when overclocking the CPU was considered dangerous and only a few people would even attempt it, and those people were treated like gods? Not anymore. Anyone can overclock their CPU if the CPU is overclockable and necessary hardware is available. Getting a modest overclock is easy, but if you are trying to squeeze all that extra juice from your CPU, you might need more than just luck.

Overclocking is not dangerous if properly done. Know your limits and obey by those limits. For example, don't insert a lot of volts to the CPU because it will damage the CPU. Keep it within the manufacturer recommended/commonly accepted volts for your particular CPU. Also be wary of the temps of the CPU as well as other components like VRM heat sinks of the motherboard.

Here are some tips.

1) Prepare a separate environment for overclocking if possible. If you have a spare drive lying around, install a free copy of Windows in it and don't install anything else - not even Windows updates or driver updates. Why this is important is because when your OS crashes while stress testing because your overclock is not stable, it can corrupt the files and settings and that alone can give a BSOD when you stress test with different settings the net time. You might think the CPU is unstable, but in reality it could be a corrupted driver or corrupted software that is running in the background.

2) Stress test for a couple of hours at stock settings. This will rule out any issues that might be present in the original hardware. No point trying to overclocking broken hardware. You'll only waste your "precious" time. Running Prime95's blend test should do the trick.

3) Don't overclock your RAM at the same time. Overclock only one component at a time, otherwise you won't know the culprit of the instability if you discover any. Overclocking RAM doesn't give any tangible benefits anyways. You will get better performance by overclocking the CPU. So, if you have high speed RAM (i.e. faster than 1600MHz), I would suggest that you drop the speed to 1600MHz CL9 timings. That would rule out the memory being the culprit of a crash. This is not needed, but just to be sure

4) Make a target. The target should not be in terms of clock speed, but in terms of max CPU temperature or max Vcore, whichever is reached first. For example, max Vcore = 1.4V and max CPU Temp = 85C. First, try Vcore = 1.4V at stock clock speed and run a stress test while monitoring CPU temperature. If you are below 85C, then you can simply try increasing the CPU clock speed. If you hit 85C, then stop stress testing and lower the CPU Vcore until your max temps are below 85C.

However, please note that when you increase the clock speed, the temps would increase a bit more even without increasing the Vcore. So try to stop about 5C below your suggested max CPU temps (i.e. 80C) to give some room for the temps to increase due to clock increase. Now that you have found your max voltage, it is time to increase the clock speed.

5) Be patient. Don't try to rush out thing. If you know that the CPU model you have usually overclocks to, say, 4GHz undoubtedly (all the current generation CPUs do), that would be a good place to start. You don't have to start from the stock speeds. That's not necessary. However, only increase clock speeds one step at a time.

For example, when it comes to Intel, the only way to effectively increase the clock speed is by raising the multiplier. Increase the multiplier one step at a time. That means, if you were starting from 4GHz, which means the multiplier was at 40x, increase to 41x, then 42x and likewise. At each step, stress test for like 30 minutes of Prime95's blend test. That should be enough for the intermediate stress tests. Once you find the maximum overclock, stress test for 24hrs (at least 12hrs) to make sure that is completely stable. If the stress test fails while doing the extensive stress tests, you'll have to drop the multiplier by one level. Then it is 24hrs of stress testing again. (Overclocking takes times. Don't rush it.)

6) Now that you (hopefully) found out your max CPU clock, you can revert to the original RAM settings or even try to overclock it. Remember that when you overclock your RAM, you might have to increase the VTT (memory controller voltage) to make it stable.

7) If you want to overclock further and you were limited by temperature, it is time to invest in a better cooler.

8) If you are having some particular issues, such as crashes at idle, take a look at this thread.

Those were some common guidelines and all that they do is make your life easier, even though at first it might seem like time wasters (preparing a separate environment just for overclocking, for example).

There are more extensive instructions on the forums such as Overclock.net, Techpowerup.com and XtremeSystems.org so read the guides posted there, ask questions and participate. You will learn a lot by doing that.

Are you upgrading to Haswell? Improved iGPU will be the killer feature.

Would you upgrade especially if you are already running a 2nd generation or 3rd generation Core i-series CPU? The leaked benchmark scores aren't that great. They maybe for early engineering samples but I don't think the final chip will be much faster than the chips used in the leaked results. Intel won't be making big last minute changes.

I'm running a 2nd generation Core i7 2600K CPU and it's overclocked to 4.5GHz. We don't know how high Haswell CPUs will clock. You would expect an increase in overclock potential because of process shrink but IvyBridge didn't work that way. So you won't know until the full reviews are out. But if these CPUs can hit 5GHz easily, the increase on IPC will carry the performance delta past 30% mark. That's a healthy increase in performance these days.

(Image courtesy of Anandtech) 

One more issue I have with the leaked specs is that the TDP has increased from IvyBridge (84W vs. 77W) contrary to the popular belief. Haswell was supposed to bring significant power savings to the table. Why has the TDP gone up by 7W? I wonder if most of that power is resident in the iGPU. The biggest performance improvements in Haswell will come from the iGPU which doesn't mean much to most enthusiasts. The only thing you might want to do with the iGPU is video transcoding using Intel's QuickSync technology. Even then, the options are pretty limited. The application support for QuickSync sucks at best.

BUT...but...it looks like the drought is going away. You know what Handbrake is, right? Anandtech just reported that Handbrake is about to get QuickSync support. That would be awesome!!! That would be the reason to upgrade to Haswell, especially I'm only using a 2nd generation Core i-series CPU.

(Image courtesy of  wccftech.com)