Doing it right.

For an index of all my Mac Pro articles, click here.

If you examine the three model years and six versions of the current Mac Pro – 2009, 2010 and 2012 – each of which came in single or twin CPU models – only one of these uses a non-standard Intel Xeon CPU design. The odd duck is the 2009 dual CPU version which uses CPUs without the Integrated Heat Spreader (IHS), an alloy top plate which is both glued and soldered to the CPU and the surrounding base itself. I have read no definitive reasons for this approach and would guess that Apple must have been concerned about thermal performance, the no-IHS design likely being easier to cool with the massive heat sinks used in the Mac Pro. What is mystifying is that the 2010 and 2012 dual Mac Pros reverted to stock IHS CPUs with little or no other apparent changes, other than to the height of the heat sinks to accommodate the thicker CPUs and reversion to stock sprung CPU retainers.

You can see the difference in Apple’s own drawings in their service manuals:

The practical upshot of the dual CPU 2009 design difference is that upgrading the CPU means that the heat sinks must be raised 2mm, a process fraught with risk of damage to the relatively fragile CPU socket, if not to the CPU itself. A damaged socket means a new processor board, some $400, and loss of use while you wait for the replacement. That and a bruised ego, not to mention time wasted.

Additionally, revisions have to be made to the heat sink fan power connectors and additional thermal pad material must be installed on the voltage regulators to maintain proper thermal contact with the heat sinks. Finally, fresh threadlocker should be applied to the retaining bolts for the heat sinks, as I illustrate.

In mechanical tasks proper instructions never use the word ‘judgment’. Judgment is subjective. ‘Subjective’ translates, as often as not, to catastrophe. Thus, in determining a measured approach to CPU upgrades in the dual 2009 machine, I set forth a low risk, objective way of securing the heat sinks with the upgraded, thicker CPU in place which will protect those fragile CPU sockets from damage.

Yet, clearly, the most elegant way of doing the 2009 dual upgrade is to use upgraded CPUs without the IHS fitted. The snag is that Intel appears never to have marketed these. This has caused determined upgraders to use various methods to remove the IHS, including razor blades, unmanaged heat, force and abrasive means to emulate the stock Xeon E5520 design.

The advantages of this approach are that the upgraded CPU is simply dropped into the socket, the loose alloy frame is placed over the CPU and the heat sinks are reattached with exactly the same number of bolt turns as was required to remove them. No changes are needed to the temperature sensor sockets in the heat sink and no additional thermal pad material need be installed. Nice and simple.

The disadvantage is that the IHS removal process is high risk. The risk includes mechanical damage to the CPU and financial damage to your pocket-book if you get it wrong. And it is not easy.

Reader Paul Opsahl had purchased a 2009 dual CPU Mac Pro after seeing my series of articles on this exceptional bargain of a computer. When Paul mentioned that he had managed to remove the IHS plates in the two Xeon W5590s he used in his 2009 dual upgrade, I was all ears. I asked him to detail his method and I think you will find it makes for fascinating reading. As the IHS is both soldered and glued to the CPU and the printed circuit board, some razor blade work remains but Paul’s technique to remove the solder bond is obviously the right way to do this, and I present his work below.

You will find Paul’s approach rigorous and well researched – in marked contrast to the amateur hour methods found on the many chat boards out there. You will also be amazed at just how shoddy Intel’s soldering technique is – maybe reason enough for Apple to adopt a non-IHS design in the first place!

If you want to install upgraded non-IHS CPUs of your choice in a dual 2009 Mac Pro, and want do do it right with non-IHS CPUs, contact Paul and see if he can fit you in while doing his day job and keeping bread on the table. You will find his rates for the work of removing the IHS reasonable. Regard the payment you make to him as the cost of insurance against damaging the CPU sockets in the processor board of your precious Mac Pro. Paul is an electronics engineer, not a backyard hacker, and has the right tools for what is a difficult job.

Paul takes it from here.

Preliminary discovery:

I started with a visit to the local computer recycler (Midwest Computer Brokers) in Walford, IA. I asked if they had any old Xeon processors available – any speed. I explained that I wanted to learn how to successfully remove the IHS. They gave me a Pentium 4 processor that already had the IHS removed with the aid of a blow-torch and a flat blade screwdriver. I decided that I could use that in a first experiment to determine what temperature the solder melts at ( more on that later ).

During a second trip to MCBIA, I specifically asked for a Xeon processor with the IHS intact. They offered to sell me a known good Dual Core 3.03GHz Xeon for $40. The goal with that device was to learn how to cut through the adhesive without damaging the printed circuit board and remove the IHS using a temperature controlled heat plate. I had seen a post where someone had used a double-edge razor for this purpose. I opted for a single-edge razor thinking there would be less chance for me to see blood. The single-edge blade worked, but I later learned that the single-edge blades are almost twice as thick as the double-edge blades ( the double-edge blades I used later on the Quad Core Xeon processors are approximately 0.004 inches thick ). The adhesive is very soft, but tightly compressed between the IHS and the rigid printed circuit board in a gap of about 0.008 inches. That narrow gap means that it’s tough to get even the thinnest razor blade between the IHS and circuit board. It took several passes to get to the point where I sensed that the blade was no longer cutting adhesive. Having access to the dimensions of the adhesive spread would have helped here. Based on a visual inspection of the adhesive using a microscope, I concluded that the IHS was cut free ( except for the solder ). There is a place on one side where the adhesive is absent – mostly likely to allow cleaning solution to escape from under the IHS during assembly.

Temperature controlled heat plate with vacuum:

At that point, I was ready for the temperature controlled heat plate. During my experiment with the Pentium 4 processor, I placed the already removed lid upside down on the heat plate and slowly increased the temperature until the solder melted. The heat plate is controlled with a digital thermostat that makes this measurement quite easy.

The hot plate and microscope at work.

That experiment indicated that 165 degrees C was the temperature I should use during this second experiment. Confident that Intel had used the same type of solder to attach the IHS to a Xeon processor, I set the temperature to 165 degrees C and waited until the temperature controller indicated that the heat plate had reached 160 degrees C. At that temperature, I turned on the vacuum pump that holds the device in place via a small hole in the center of the heat plate and placed the processor on the heat plate with the IHS side down. This technique limits the amount of heat absorbed by the assembly. I started applying a very light upward force on the printed circuit board using a tweezer – just enough to lift it away from the CPU when the solder had melted. The temperature of the heat plate continued to slowly increase – after about 20-30 seconds, the solder flowed and the printed circuit board lifted away from the IHS. I lifted the printed circuit board away and placed it to the side to cool.

Post-removal inspection:

During inspection of the remaining solder on the CPU, I noticed several voids or air pockets where solder was missing. In addition, there appeared to be some areas where there the solder flowed to the IHS, but had not completely bonded with it – referred to as a cold solder joint. These are process problems for Intel, but nothing to worry about going forward in this application unless they are severe. Localized over-heating in the CPU could be a problem if the voids were large, but these were small.

Poor soldering by Intel.

Removal of excess solder and adhesive remains:

The next task was to learn how to remove the excess solder on the CPU and the remaining adhesive from the printed circuit board. I did not want to heat the solder on the CPU with a solder iron as there is no way to know what temperature the CPU might be exposed to. Given that solder is very malleable, I decided to try using the razor blade to cut it away. This approach worked very well – only being careful to keep the blade flat against the CPU. You should not expect to see a clean gold-plated CPU at this point – a very thin layer of solder will not cause a problem in the final Mac Pro installation.

Cleaning the IHS adhesive off the printed circuit board was much more difficult. After another evening with Google, I found a video on YouTube where a Dremel tool had been used along with isopropyl alcohol. Given that the Dual Core Xeon was my test case, I elected to try this approach. I’m not convinced the alcohol is necessary as it does not soften the adhesive in the least – I elected not to use it on the W5590s. My only concern here was that static electricity could damage the CPU – to reduce that risk, I made certain that I was properly grounded with an electrostatic wrist strap. I would have no way of knowing if the processors were damaged until I had used this approach on the W5590s and installed them into the Mac Pro. I also changed the tool-head often in order to efficiently remove the material from all sides.

Post-removal inspection:

After removing the excess solder and the remaining IHS adhesive, I inspected the printed circuit board and CPU under a microscope. The inspection gave me confidence to proceed with this approach on the W5590s. The Dual Core Xeon board has only the CPU under the IHS – there are no other components. This is not the case with the Quad Core Xeon processors. As can be seen on the 2.26GHz CPU’s removed from the Early 2009 Mac Pro, there are 10 power supply bypass capacitors adjacent to the CPU. This made removing the IHS more difficult on the W5590s.

CPU with IHS, adhesive and solder removed.

Process with the Xeon W5590 CPU:

Using the approach documented here, I started work on the 3.33GHz Xeon CPU’s. All went well until the final inspection. At that point I noticed that one of the CPU’s had two damaged capacitors – I had cut into each of them while cutting the IHS adhesive. This problem was solved by purchasing a 2.26GHz Quad Core Xeon processor ( no IHS ) on eBay for $25 and using that as a donor to replace the damaged capacitors. There are 10 capacitors – each apparently 2.2 microfarads. I could have used the CPU’s that were in the Mac Pro, but I elected to keep those in case some issue came up with the W5590s when they were installed. A close visual inspection of the capacitors is important as damage there could result in the metallization layers shorting together.

Damaged capacitors.

For future IHS removal projects, I have made a tailored jig to keep the razor blade flat with the printed circuit board. The design of the jig is such that a maximum depth of cut is maintained on all four sides, precluding damage to the surface mounted capacitors. The razor is held at an angle of 30 degrees between the two metal components with 0.260 inches of the blade extending into each channel – enough to cut the adhesive without damaging the capacitors.The fixture holds the razor blade rigid while the processor is moved, in the channel with the IHS side down, past the cutting edge. The fixture design allows the printed circuit board of the CPU to flex upward as the razor moves into the gap between the IHS and printed circuit board. (I have seen the engineering designs for Paul’s jig, and can confirm that it is exceptionally elegant. The accuracy of the tolerances suggests that there is no risk of damage to the CPU. – Ed).

If you want Paul to remove the IHS from your CPUs, you can email him in Cedar Rapids, Iowa, for more details, by clicking the icon below:

Click the icon to mail Paul Opsahl.

If you use his services, my benefit is precisely zero.

Thank you, Paul.

A note on temperatures. Intel’s web site states that TCase should be limited to 67C (153F to us Luddites); Paul explains that this refers to continuous use. TCase it the temperature on the outside of the IHS as Intel illustrates here:

TCase for the Intel Xeon W55xx CPU family.

How then can the CPU withstand the 165C (329F) required to melt the IHS solder? Well, Intel applies that level of heat when first attaching (however ineptly) the IHS, as that’s what is needed to melt the solder in the first place. Thus application of like heat when removing the IHS should not damage the CPU provided that the period is limited. Intel’s data sheets (doubtless much beloved of Chinese intellectual property thieves, though they still have to pay Novellus and Applied Materials for the costly fabrication machinery when copying/stealing Intel’s CPU designs) have data on tolerances which lead to this conclusion. Bottom line? Don’t overheat and don’t heat for too long and you will be OK. Just enough to melt the solder and permit removal of the IHS.

Best of all, have an expert like Paul do the work for you. His 2009 Dual Core Mac Pro uses de-lidded non-IHS W5590 3.33GHz CPUs, modified using his technique above, so you can be comfortable that he eats his own cooking.

Adding SATA III drive capability.

For an index of all my Mac Pro articles, click here.

SATA III:

One of the signs of the benign neglect afforded its Mac Pro line by Apple is the absence of SATA III compatibility. Insert a SATA III 6gb/s drive in a Mac Pro of any vintage and it will run at SATA II 3gb/s speed. Meaning at half of its design speed.

However, Apple’s sloth and refusal to add a feature found on just about every competing computer on the planet can be countered at modest cost using a PCIe card adapter.

Before making this conversion, my Mac Pro was using two 120GB SSDs, the one containing OS X, Applications and the Lightroom catalog. The other is a backup clone. Placing the LR catalog on the speedy SSD greatly improves LR’s speed, at the penalty of using up space on the disk drive. My 11,000 image catalog is some 40GB and, as a result, my SSD startup drive is over 90% full. Not good.

RAID 0:

Adding a 250GB SSD drive internally would not solve my problem – I would have to replace both the existing 120GB drives to maintain a redundant backup clone. Or I could simply make a 250GB partition on one of the internal HDDs as a backup location. However, it struck me that the two 120GB SSDs could be made into one drive using Apple’s RAID 0 capability built into Disk Utility. Then adding just one 250GB SSD would give me a new startup drive with space to clone it on the RAID 0 SSD pair. RAID 0 makes the two drives appear as one in Finder, and there is no redundancy. On the other hand RAID 0 has the nice side benefit of doubling the effective speed of the two drives as OS X can write data far faster than the SSDs can receive it. Double the number of SSDs and you double the write speed, so older SATA II paired RAID 0 drives in SATA II slots now run at SATA III speeds. Twice as fast. Magic.

Apple’s document on software RAID configuration using Disk Utility is here and makes for interesting reading for those contemplating this approach.

PCIe card:

Now a related snag is that I have no more internal connectivity for additional disk drives. Here is my disk layout before adding the 250GB SSD:

Mac Pro drives before the upgrade. SSD Bak resides in the optical drive area.

The four drive bays are taken up with the SSD boot drive, two HDD data drives and a TimeMachine drive. The backup for the SSD is in the optical drive area.

Shopping around I learned that plug-in PCIe cards are available which will accept one or two SSDs. As these use one PCIe slot, SATA III speeds are available. Apricorn makes a well reviewed PCIe SSD card in both single and dual SSD form factors, but the dual version blocks Slot 3 – not good. Accordingly, I settled on the single SSD version for all of $49, leaving Slot 3 available for later use:

Apricorn Velocity Solo PCIE/SSD card.

The SSD slots into the card and the card into the Mac Pro with no cables involved.

SSD inserted in the Apricorn PCIe card. The additional SATA III data port is circled.

Three of the four retaining screws have been inserted and, yes,
John White does outstanding Nikon Ai conversions!

It gets better. The Apricorn has a second SATA III port for an additional drive (SSD or HDD – any size you can accommodate) which can be placed anywhere inside the cavernous interior of the Mac Pro, without interference with PCIe Slot 3:

SATA III port on Apricorn Velocity.

It makes more sense to use two single-SSD Apricorn cards rather than one dual one if you need to use two SATA III SSDs. Both take two slots (the dual overlaps a slot, making it unavailable), but by using two cards you get two SATA III data expansion sockets. Using one dual card you only get one SATA III expansion socket. So when you come to add additional drives, as I illustrate here using the expansion socket, you will be able to double the number of HDDs or SSDs thus added. Both the 16x (Slot #2) and 4x (Slots #3 and #4) will support full SATA III speeds.

Installed in PCIe Slot #2. #1 is for the GPU, top, #3 is for the USB3 card, #4 is currently vacant.

This means that a second drive, which will have to be powered from a cable from the optical drive enclosure, can be connected to the Apricorn Velocity card, and will be seen by OS X as a discrete drive and will run at SATA III speed. Good to know for future expansion needs, when they arise. For example, I could add a 2TB HDD connected to this port, and make my two existing 1TB HDDs into a RAID 0 array, behaving like a 2TB HDD. In this way I get a fast 2TB data HDD with the backup free. I ended up doing just that and illustrate the process here.

The Apricorn PCIe card comes with a 3 year warranty.

Choice of SSD and installation:

Unlike Apple, today’s Samsung is a major innovator, and its latest 250GB EVO 840 SSD is highly regarded as a state-of-the-art SSD, so I paid an additional $183 for one, making the package total around $250 with tax. It comes with a 3 year warranty.

Installation requires only a small Phillips screwdriver for the four screws which retain the SSD drive to the PCIe card (it takes less time to install than to unwrap the parts) and after using CarbonCopyCloner to clone the existing 120GB startup drive to the new Samsung SSD, I told System Preferences->Startup Drive to use the Sammy as the boot drive. Further, I did a PRAM reset on restarting the Mac Pro, some instances of the boot drive not being properly recognized having been reported unless a PRAM reset is done. (Note: If you have added USB3 capability, as I illustrate here, be aware that you cannot boot from an external USB3 drive. This is because the USB3 ports on the add-in PCIe card are not powered until OS X has started).

When you go to clone your current startup drive to the new SSD,
CCC will warn you that no Recovery Partition exists.

Go ahead and create the Recovery Partition using CCC. Now go to
System Preferences->Startup Drive and make the new SSD the startup.

The Recovery Partition has been created.

Drives before reconfiguring ‘SSD Boot_old’ and ‘SSD Bak’ to a RAID o drive.

Once the Mac Pro was restarted from the new PCIe Samsung drive, I set the two old SSD drives up as one striped (not concatenated) 240GB RAID 0 drive, to act as the backup drive for the new Samsung startup SSD.

Disk Utility used to set up two SSDs as one ‘striped’ (meaning RAID 0) disk set.

This is how NOT to do it:

Concatenation simply strings the two disks end to end – when one is full the other takes over.

As concatenation results in only one disk being written to at any one time, the benefits of speed doubling using a RAID 0 disk set are lost, and the speed is no greater when concatenated than that of the individual drives.

I now have a 50% full Samsung startup drive and a 50% full RAID 0 backup comprised of two 120GB SSDs. The RAID 0 drive pair is reported as one drive in OS X’s Finder. Nice. Do not use the concatenation option – that strings the two drives end to end so you do not get the speed doubling that a RAID 0 striped array delivers. See the test results below to confirm this.

Be sure to enable TRIM for the new SSD and all will be well.

Confirming TRIM is in effect in System Profiler.

The new SSD is properly reported in SMARTReporter:

The new SSD in SMARTReporter.

Performance data:

How does it perform?

I set forth below four sets of BlackMagic disk speed test data – for the original 120GB SSD startup drive (a SATA III Sandisk running at SATA II speed), for the SATA II Intel former backup SSD, for the new Samsung SATA III 250GB Apricorn SSD and for the RAID 0 pair of two 120GB SSDs (the Sandisk and an older SATA II Intel) used as a backup:

Original 120GB Sandisk SATA III SSD running at SATA II speed:

Original Intel X25-M SSD, SATA II:

As you can see, there is little difference between the SATA III drive running at SATA II speed and the SATA II drive.

250GB Samsung EVO Apricot PCIe SSD running at SATA III speed – 2 to 3 times as fast:

120GB + 120GB RAID 0 Sandisk + Intel SSDs used as backup – twice as fast as each individual SSD, two SATA II drives running at SATA III speed:

A) Running in RAID 0:

Both discs are written to simultaneously, doubling the effective speed.

B) Running as two disks concatenated into one:

Concatenation is not the right way to do this. One disc is
written to until full then the other one is written to.

The Recovery Partition:

One final note. RAID does not support the Recovery Partition which allows recovery of the OS in Mountain Lion if your OS gets corrupted. Thus it makes more sense to boot from a plain vanilla SSD than from a RAID 0 pair, as the latter will not have the Recovery Partition available.

Use with Boot Camp with Windows:

There are many reports that Windows will not boot from the Apricorn – or other PCIe SSD – card using Boot Camp. As there is no conceivable scenario in which I will waste time with Windows, you are on your own here. I would suggest you use an emulation application like VM Ware, Parallels or VirtualBox (free) and run Windows from that – an approach which makes Windows just another executable application under OS X and requires no restart of the Mac Pro.

Conclusion:

This is a very cost and performance effective solution. You extend the lives of older, smaller drives, double their speeds, get a twice as fast startup drive and add an additional high-speed SATA III expansion option in the process. Power consumption of drives connected to the PCIe card is way below the maximum allowed. What’s not to like?

Don’t forget to recreate the related CCC Scheduled backup task when you are done – use of UUIDs for drives dictates this.

Update June 2015:

While you can squeeze in one or two SSDs using the Apricorn PCIe card above, there is now a more space efficient approach to adding lots of SSD drives, using the newer mini-PCIe mSATA drives which retail for a modest premium over regular 2.5 SSDs and are much smaller. Up to four 1TB mSATA SSDs can be installed on one PCIe card from Addonics (costing all of $55) and no external power supply is needed. These SSDs can be configured as JBOD (separate drives), RAID0 (two or more drives seen as one with no redundancy but increased speed) or RAID1 (mirrored drives with automatic back-up) using Apple’s Disk Utility.

I have sold my traditional SSDs and migrated to the Addonics/mSATA approach for its space efficiency, better cooling and higher capacity, and you can read all about that here.