Power consumption, stress tests and diagnosis.

Power consumption:

Using the calculator here I input the components for the modified Mac Pro:

Worst case power consumption data.

This assumes everything is running flat out. The PSU in the Mac Pro is 980 watts maximum continuous power, call it 1,400 watts peak power, more realistic of real world ‘spiky’ power demands, so it’s comfortably able to handle even the thirsty W5590 Xeons. At a more granular level, the two backplane sockets for powered GPUs are limited to 150 watts total, with the PCIe slots capable of delivering an additional 75 watts, for a total of 225 watts. So a GPU like the latest nVidia GTX780, which can draw as much as 250 watts, is demanding unsafe power draws when maxed out. If in doubt, add an external power supply to power extreme GPUs. If you use two of these, external power is essential.

Using the stock E5520 Xeons (2.26GHz) the BOOSTA and BOOSTB fans (inside the two CPU heat sinks) run at 1100rpm. The E5520 have a Thermal Design Power (TDP) of just 80 watts, at which point the CPU is throttled back to prevent meltdown. By contrast the W5590 CPUs have a TDP of 130 watts, so they will run hotter under load. At idle, my CPU fans run at 1,650 rpm, considerably higher than stock. As these fans are buried inside their respective heat sinks and the exceptionally robust case is well sound insulated, the increased fan speed results in no discernible increase in noise, which continues to be approximately 44dB at ear level, two feet distant from the front of the Mac Pro, which stands out in the open to the left of my desk.

Other fan speeds (PSU, processor cage, PCIe space, GPU) remain unaffected by any of the enhancements made.

With the components shown above, my Mac Pro idles at 188 watts. Set to Sleep it consumes but 9 watts.

Stress tests:

Ordinarily I test stability by ripping a DVD then compressing it in Handbrake. While this really exercises the CPU and memory, it’s not the best test out there. Handbrake does not work every CPU core and thread to the maximum.

There is a Stress Test function in Geekbench. This application is ordinarily used to test CPU and RAM performance, but go to Geekbench->Benchmarks->Run Stress Tests and this is what Geekbench will do to your 16 threads, or how ever many you have:

Stress Tests, viewed in Activity Monitor. Geekbench left, Handbrake right.

Well, I ran this Stress Test and after 30 seconds …. the Mac froze! No kernel panic, no messages, just frozen. Incidentally, trying Handbrake it froze also!

As all my memory was properly reported in System Profiler, my first thought was that there was an application causing the problem, yet removing applications one by one made no difference. The Mac Pro continued to fail the Stress Test. So it’s not an application issue.

Suspicion then fell on the memory as Geekbench tests both the CPU and the RAM. Four of my six 4GB sticks migrated from the Hackintosh where they had performed perfectly for 2 years. Two more were added recently to take advantage of the three channel Xeon CPUs which optimally address three sticks of RAM each. Three sticks for each CPU, for a total of 24GB.

So I pulled four sticks and ran the Stress Test, leaving just 8GB installed. Perfect. Stress Test passed.

Added back two more. Still perfect. Stress Test passed again.

Added the last two …. still perfect!

Conclusion? One or more of the memory sticks was not properly seated and was causing failure of the stress test. Quite how a stick can get improperly seated beats me, but it’s not the first time I have encountered the ‘badly seated memory sticks’ issue, although this is the first time I have been the victim.

And the result?

Passed with flying colors.

Temperatures? Benign – the highest I saw was 140F for one of the CPUs, everything else in the 100-110F range. The fans never broke a sweat despite the brutal nature of this test, running at minimum speeds throughout.

I’m now satisfied the Mac Pro can take just about anything I can throw at it.

The Diagnostic process:

With a host of variables in play the only way to diagnose issues is to remove or replace one variable at a time. For example, to pull memory sticks and close applications at the same time would be a pointless process. If the issue was solved as a result, you would not know which was the cause of the problem. It may seem slower to work one variable at a time, but, net, it’s the fastest diagnostic approach.

By the way, all that pulling and replacing of memory sticks is hardly onerous in a 2009 or later Mac Pro. The design makes the removal and replacement of the processor/RAM assembly a speedy affair.

Temperature issues.

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

Long time readers will recall that what drove me to using a Hackintosh five years ago was that I managed to fry the GPUs in three iBooks and two iMacs in quick succession. The last two were just out of warranty, naturally, and my efforts to better ventilate the nVidia 7300 GPUs and then to replace the 7300 in one with a 7600 all failed. Further, Apple denied any cooling issue existed even though their own discussion boards were replete with complaints. This led me to put out the iMac in disgust to a location better suited to its abilities:

The garbage awaiting collection

Thus it comes as no surprise that I have kept a watchful eye on operating temperatures ever since and even though the Mac Pro comes with state-of-the-art cooling, it’s a hard habit to shake.

The purpose of this piece is to address Mac Pro temperatures and suggest one minor enhancement.

The Mac Pro is exceptionally well cooled, with six fans in the 8-core model. (The 4-core has five, as it has one processor, not two). One pair of fans cools the processor cage in a push-pull configuration, each massive processor heat sink contains a fan, the PCIe fan cools the space occupied by the PCIe cards and disk drives and the large 980 watt power supply has its own fan.

The other day I ripped a movie then compressed it using Handbrake. That’s a stressful process which exercises all 8-cores in the two Intel Xeon CPUs used in the 8-core Mac Pro. It was an exceptionally warm day and the Ambient (case vicinity) temperature reported by iStat was 90F. All the operating temperatures of the various monitored components were fine, running in the 120F range, but with one exception. The Northbridge chip was reporting 165F. The fans did not spool up and the job ran fine, a full length uncompressed movie being reduced to a compressed iPad M4V version in 25 minutes. Nice.

Well, I did some reading of chat boards and they were – how to put this politely? – innumerate.

Thus, I went to the source, the excellent Intel web site, to check on the Northridge thermal specifications. Northridge is the chip which controls communications between the CPU and RAM, and Wikipedia has a fine piece on it here. Note that the increasing drive to integration saw the Northbridge chip move within the CPU case by the time Intel produced the later Sandy Bridge CPUs, which is what my Hackintosh used.

Intel’s thermal specs provide the definitive answer in their Northbridge Thermal Guide. Here’s the key extract:

Northbridge temperature limits.

I’m strictly a Farenheit guy and water boils at 212F at sea level in my world, so Intel’s 104C converts to 219F as the case temperature limit for the Northbridge chip. One commentator on a chat board stated he removed the chip and refreshed the thermal paste, but the loss was only 4F, which hardly seems worth the effort.

Nonetheless, as I never like to see high temperatures, I did some measurement. Here’s my iStat reading with the Mac Pro just doing some modest surfing:

Stock, static temperatures.

I then fired up SMCFanControl, which you can download free – click the link, and increased the fans speeds of the processor cage Intake and Exhaust fans from 800rpm to 1050rpm, thus:

Intake and exhaust fan speeds.

The processor assembly is the part-withdrawn assembly in the picture below (4-core above, 8-core below), holding the two CPUs, RAM and, yes, that hot Northbridge chip. There are two CPU fans within the heat sinks on an 8-core machine, named BOOSTA and BOOSTB, the first barely distinguishable in the image above:

Processor assembly partly withdrawn.

This assembly nests within the processor cage which I earlier removed to install an Airport card on the backplane board:

The ever diligent hound guards the processor cage.
Note how the fans have been moved in 1/2″ toward the center of the cage.

Those fans, which work in a push-pull configuration, are the ones we want to speed up if more air flow is to be directed over the Northbridge chip.

The modest speed increase for the minimum fan speeds from 800rpm to 1050rpm resulted in a 1-2dB noise increase at my working location, in practice barely detectable. But the benefit in terms of Northbridge temperature drop was excellent:

Northbridge temperature
drops 24F after modest fan
speed increase.

That’s a good noise:efficiency trade-off in my book and I’m leaving those two fans at that speed.

To satisfy curiosity, I cranked those two fans up to 3600+rpm, resulting in an unnacceptable noise level of 60dB. Not for daily use, but handy if you need quick cooling on a stressful task in excess of that offered by the automatic variable speed fan controllers in the Mac Pro. The Northbridge now dropped from 139F stock to 101F:

The fans cranked up.

Bottom line? For the temperature-obsessed, like me, a slight increase in fan speed to 1050rpm from the stock 800rpm for the intake and exhaust fans on the processor cage provides peace of mind.

It’s hot and I want to leave the cover off:

That is a bad idea. The Mac Pro’s massive alloy side cover plate is part and parcel of the thermal design. Look hard at the inside and you will see insulating rubber gaskets which seal the enclosure when the cover is in place.

The rubber seal inside the Mac Pro’s side cover.

With the cover in place the tunnels formed by the cover are completed, presenting a side-sealed conduit for the push-pull processor cage fans, resulting in optimal cooling. Leaving the cover off actually compromises cooling performance. Do not do it.

The value proposition.

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

The Mac Pro’s interior – engineering elegance redefined.

The interior of the Hackintosh –
made by the same people who assembled Fiats in the 1960s.

Here I take a quick look at the differences in the cost of acquisition between an used 8-core 2009 Mac Pro and a new Core i7 Ivy Bridge Hackintosh.

I am focusing on differences only. Displays, memory, SSDs, hard drives, GPUs, keyboards, mice, cables and so on are identical for like-equipped machines and can cost as much or as little as you desire.

The comparison commences with an 8-core 2 x 2.26GHz Intel Xeon 2009 Mac Pro, at the price I paid. The only additions to the chassis to make things comparable are a used Airport card, with the GT120 stock GPU being sold. Both machines can use a GPU of choice – it’s not a component of difference.

The stock twin Intel Xeon CPUs in the Mac Pro deliver performance identical to an Ivy Bridge i7. They can be readily upgraded for less than $500 for a 50% boost in performance – the Xeon is not easily overclocked. The Ivy Bridge is pretty much maxed out stock, though the unlocked version can be overclocked for maybe a 25% performance gain. At that point the Mac Pro will be the faster machine, and will be considerably superior on multi-threaded operations.

The Mac Pro is stripped of dated components including its memory, hard drive and the GT120 graphics card, the latter still in production and commanding a good used price. That GPU is a dated slowcoach, not up to modern demands.

Now you can see why a Hackintosh no longer makes economic sense. The Hack’s Corsair Obsidian is a nice case, but at 25lbs of pressed steel to 41 lbs of machined alloy it cannot hold a candle for quality, robustness or modularity to the Mac Pro. For that matter, nothing can. The Mac Pro has to add USB3, built-in on the Hack’s Gigabyte motherboard, whereas everything in the Hack – PSU, motherboard, CPU, cooler, fans and so on, is extra.

Elements of difference – Mac Pro vs. Hackintosh.

The cost is near-identical between the two machines but the Mac Pro owner, in addition to pride of ownership, never has to tinker, hack or worry about OS upgrades. He will never know the misery of ‘Boot0’ errors or arcane fixes to get online to the AppStore. He will also enjoy a resale value exceeding 80% of cost whereas the Hack owner will lose 50% over the next 3 years. While a well engineered Hackintosh can be reliability personified – mine was – any major OS change involves much research, reading and delay before implementation, for fear of breaking something. The Mac Pro user, by contrast, simple clicks on the ‘Upgrade’ box. What’s your time worth?

The Mac Pro buyer’s biggest issue is finding a nice, clean 8-core machine – a task made no easier by my series of articles on the value offered! After that, he’s ahead of the game economically, psychically and operationally.

Better graphics.

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

Where we are headed.

The stock card which comes with the 2009 Mac Pro is dated by modern standards. The GT120 is a slow performer and supports only two external displays, one using DVI the other Mini DisplayPort.

My goal was to transplant the Zotac nVidia GTX660 card from my Hackintosh to the Mac Pro and in practice the process proved to be easy.

First, however, as the GTX660 requires a separate power supply cable, I bought one of these from Amazon. You can pay MacSales three times as much for the same cable if you like. Disregard the ‘G5’ designation – this is the correct cable.

Here’s how it looks:

PCIe power cable. The small end goes
in the socket on the backplane board,
the large end plugs into the GTX660.

There are two backplane (motherboard) sockets in the Mac Pro. With the chassis oriented so that the PCIe slots go left-to-right in front of you, with their brackets on the right, these two sockets are found immediately to the left of the rear of the PCIe slot nearest to you and just to the right of the large, grey central fan, abutting the base of the processor cage:

PCIe power sockets.

Undo the two thumb screws atop the PCIe brackets on your right and remove the existing GT120 card. Then remove the adjacent blanking plate:

Ready for GTX660 installation.

The GTX660 is a double width card, so that second blanking plate must be removed.

Depress the slider button atop the central fan enclosure – this is generally referred to as the PCIe fan:

Button on the fan casing depressed.

With the button depressed the outer casing of the fan assembly can now be slid to the left. This moves the PCIe card chromium retaining bar to the left and allows insertion of a full length card:

The fan casing/retaining bar have been moved to the left.

Insert the GTX660 card:

Graphics card inserted. Arrow denotes the retainer bar operated by the outer fan casing.

Check all is properly aligned:

Graphics card sockets properly aligned. Orange = DVI-I; White=DVI-D.
27″ and 30″ displays must be connected to the white or orange DVI sockets
using dual-link DVI cables for maximum definition with displays which have
more than 1920 x 1200 pixels.

Insert the PCIe card retainer and tighten the thumbscrews. Then remember to slide the PCIe fan casing back to its original position, thus locking down the rear of the graphics card.

PCIe card retainer in place.

Insert the other (large) end of the PCIe power cable into the graphics card. The chassis has been flipped through 180 degrees here with the GTX660 sockets on the left, for clarity of illustration. Some later cards require that two power cables are used; the GTX660 needs one. The stock GT120 needs none.

Power cable plugged into the graphics card.

Replace the cover, plug in all your cables and you are ready to go.

I use three Dell displays all with DVI sockets. The GTX660 provides four display sockets – DVI-dual link (for 27″ and 30″ displays as well as all smaller ones), DVI single link (all others except 27/30″), DisplayPort (not Mini DisplayPort) and HDMI. My displays are connected with two DVI cables and one DisplayPort->DVI cable.

How is it in practice?

Anomalies:

With Mountain Lion 10.8.4 (and 10.8.3) Apple started including nVidia drivers for this card as part of the OS. Any OS before the last version of Lion will not cut it with any GTX6xx series card. You must have late Lion or Mountain Lion. My Mountain Lion installation is plain vanilla, downloaded from the AppStore, and the card was recognized out of the box. I did not install any separate nVidia drivers. However, as the ROM in the card is not the officially sanctioned Apple ROM (the ‘Apple tax’ at work), you will lose the Apple-logo splash screen. After a period of blackness on restart, 25 seconds starting from an SSD drive, your display(s) will come to life with the login screen. The other anomaly is that common temperature monitoring tools will not display GPU temperature, another quirk attributable to the ROM.

The way around these anomalies is to fit a ‘Made for Mac’ version of a graphics card – the top end being the nVidia GTX680 ($500) and the Radeon 7950 ($480). Both are overkill for still photographers but great for video processing. The GTX680 is bumping up against the 300 watt limit for power supply from the backplane sockets, using 250 watts, so if you can live with it I would suggest the Radeon 7950 which comes in at 200 watts. You can see the comparisons here.

Another alternative is the Radeon 5770 which costs $250 in the ‘Made for Mac’ version. It is slightly slower than the $200 GTX 660. It can drive up to three displays (one DVI, two MiniDisplayPort), one of which can be 27/30″.

A workaround for the Anomalies noted above:

So the only thing a non-‘Made-for-Mac’ overpriced ($50-200 more for a couple of lines of changed code in the card’s ROM – highway robbery) graphics card has is that the Apple splash screen and spinning gear wheel are not seen at start-up. Often, it has to be added, the performance of ‘Made-for-Mac’ cards is slightly lower than that of the stock items for reasons known only to Apple. Using a non-‘Apple’ card, the system goes from a black display to the login screen in 25 seconds. There is no splash screen or spinning gear wheel. This is no impediment in practice. If you are doing a clean OS X. Stall, nothing is lost or invisible. If you are unfortunate enough to want to boot Windows from a BootCamp partition or separate disc, there is a utility which allows you to specify that partition or disc as the one to start from on reboot. You look it up – I do not use Windows.

Performance of the GTX660:

In a word, a lot better than the GT120 (hardly difficult) but not at the stellar level of the same card in the Hackintosh with an over-clocked Sandy Bridge i7-2600K CPU.

I cannot run the latest Unigine test app on the old GT120 (the card cannot cope) but here is the result with the Mac Pro – click the above link to compare with the Hackintosh:

Unigine graphics test – Mac Pro 2009.

Here, also for comparison, is Novabench:

Novabench for the GTX660.

In summary, performance is 70-80% of that in the Hackintosh and of the order of twice as fast as the GT120.

How is it in practice with Lightroom and Photoshop?

Really no noticeable difference from the very fast Hackintosh. LR5 loads in 7 seconds, file-to-file Loupe review is faster than you can hit the arrow button (I use 1:1 previews). First entry to the Develop module is 6 seconds and 1-2 seconds thereafter. Flipping between images in the Develop module is instantaneous – which is why you install a card of this caliber in the first place.

Export to Photoshop CS5 is 3 seconds or less.

With its twin fans the GTX660 is an exceptionally quiet card, quieter than the stock GT120 and much quieter than the 9800GT which I used for a few days while awaiting the PCIe cable for the GTX660. It makes an already quiet machine near perfect when it comes to noise. 18″ from the front of the case the sound level was recorded at 45dB, compared with 44dB with the MacPro off.

Noise. The 73dB spike is from a birdie tweeting outside my window!

The iPhone/iPad used above is named ‘Decibel 10th’ and is free in the AppStore.

Revised power consumption data:

With the GTX660 card in use with three displays and with the GT120 in the case but not connected to any displays, the static power consumption of the MacPro is 201 watts, excluding the displays. With the Mac Pro in sleep mode the power consumption falls to 10 watts. Moral: Put your Mac to sleep when it is not in use!

Each of my three Dell 2209WA 21.5″ 1680×1050 displays uses 44 watts, so with three displays the Mac Pro is using 333 watts/hour. Not trivial. The displays sleep when not in use as does the Mac Pro. A sleeping display draws no power.

nVidia web drivers:

After I had installed the far faster Intel Xeon 3.33GHz W5590 twin CPUs, it came to my notice that nVidia had made available new drivers on its web site. OS X Mountain Lion had started including nVidia drivers with 10.8.3 so I compared the web drivers with those which come with 10.8.4 which I am currently using. You can download the nVidia web drivers here. Once installed they add a System Preference pane thus:

nVidia web drivers installed.

Running Unigine there is negligible difference:

Native OS X 10.8.4 drivers.

nVidia web drivers OS X 10.8.4.

The only thing learned here is that using a faster CPU does improve Unigine results – compare with the Unigine data above run with the stock Xeon 2.26GHz E5520 CPUs.

So why bother installing these? Well, in the event nVidia should update drivers part way through an OS X release, you will be able to install the improved drivers without having to wait for an update from Apple, so why not install them?

Conclusion:

The GTX660 (nVidia, Zotac, EVGA, MSI, PNY or any other brand) is a great enhancement to the stock GT120 card which shipped with the 2009 Mac Pro, but see ‘Anomalies’ above. The state of the art is to be had in the far costlier nVidia GTX680 or Radeon 7950 which, if bought in the ‘Made for Mac’ versions, will sidestep the anomalies noted.

Repair costs.

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

For a machine with a minimum life expectancy of 5 years, and maybe as much as 7-10 years, repair costs are a significant consideration. Well made computers are like cars – it makes much more economic sense to fix them when they fail after a few years than it does to replace. So many car owners will sell a car after, say, the alternator fails at 100,000 miles. They get the part replaced and the new owner has his depreciation paid for and a new alternator into the bargain. Major component failures can then be expected to occur every few years with the average repair/replacement cost being perhaps $500/year. Far, far cheaper than a new car whose value drops 20% the moment you drive it off the lot. That’s a lot of alternators.

Thus it is with computers. While few would argue that it makes any sense to fix a piece of junk like a Dell or HP – indeed, it makes even less sense to buy one in the first place unless your car of choice is the Yugo – things are significantly different with the MacPro and even with the G5 PPC PowerMac which preceded it, in a case of very similar design. These machines are made to last and their high degree of modularity makes replacement of worn parts exceptionally easy and low risk.

Once again, buy the Service Manual referenced in yesterday’s column. It’s $5 for heaven’s sake.

So what parts, absent accidental damage (you drop the Mac Pro on your foot and you spend 6 months in traction; repair of your computer is the last thing on your mind) are most likely to fail and what do new replacements cost?

First up is the power supply. This 980 watt behemoth is not going to blow out on you unless you do heavy-duty video rendering day in and day out in a poorly ventilated space.

The Mac Pro’s power supply.

This runs $290 – just Google the part number. Don’t buy used or refurbished – that’s a false economy.

The other moving, wear parts are the five fans, two of which are in the CPU heat sinks:

The three other fans in the Mac Pro.

Compared to $10 Hackintosh fans, these are not cheap but, then again, they rarely fail.

The front processor cage fan runs $45, the rear processor cage fan runs $38 and the PCIe (case) fan can be had for $59.

The CPU fans are buried in the heat sinks – CPU A costs $148 for fan and heat sink and CPU B is $258.

That’s all for the moving parts.

What’s the worst that can happen? You somehow manage to fry the backplane board (motherboard) or processor board. Those will set you back $344 and another $400.

Unless you are very unlucky, I would doubt that five year repair costs for parts unique to the Mac Pro will be significant. All other parts like hard drives and so on are no different from those found elsewhere and easily found and replaced.