Overclocking Tool for the Mac Pro

[JeffDM]

For the guys saying this will shorten the life of your Mac, tell that to all the guys who bought the 3.0ghz Mac Pros, and the iMac owners who bought the 3.0ghz iMacs. They're the same chip, with a speed increase, or a multiplier increase.

Same masks, but not necessarily the same chip because of small variations in the process, feature sizes can vary. Sometimes chips are marked for a lower speed for marketing reasons to fill demand for cheaper chips, sometimes they are marked lower because the batch doesn't test to be as stable at high speeds as other chips, or they run outside the established thermal boundaries at the higher speeds.

the current Mac Pro is not liquid cool. right? so overclock is going to be hot.

Liquid cooling doesn't magically change the physics. All the liquid is is just a means of moving heat to a large radiator. The current heat sinks appear to use heat pipes anyway, which do a similar job but less prone to leaking. That liquid cooling was necessary for the fastest G5s because they ran very hot. Mac Pro CPUs don't produce anywhere near as much heat.

when is the next mac pro coming? Sept? Oct?

The next chip is slated to ship in small quanities in November, assuming there are no delays. I would expect very late 2008 or very early 2009.

So users think they know what clock speed their MacPros should be running at better than Apple (who designed the computer) do?

It's not that simple. Computer makers need to have assurances that the parts will be reliable at the stated speeds for millions of parts over a very wide temperature range. So the rating may be a bit conservative, but to get a certain reliability rate, some parts may have plenty of head room where others might have narrower margins. Often it's not practical to exhaustively test each part. So there should be some headroom, especially if you're not pushing the high end of the rated operational environment. That, and the Mac Pro machine has quite a bit of head room in terms of its thermal and power design.

I wouldn't do this with computers used to make money with. A few oddities in operation are fine for personal use.

so wouldnt work for mac book pros?

Even if it did, I would suggest not. Mac Pro does have significant excess cooling capacity. Notebooks run hot enough as it is.

 

[christoh]

have been running my 2.8Ghz at 3101MHz for most of the evening with no ECC RAM problems. (8 of my 10GB are OWC aftermarket FB-DIMMS) Running mprime currently...

One problem I have noticed is an occasional audio interruption/cutout with the Presonus Firewire audio interface I use to drive my speakers. That's just while playing music. In a Skype conference I can't use that at all for audio and have to use headphones plugged into the tower.

(tangent: I've also noticed a lot of problems with this Mac Pro and firewire audio interfaces.. there was a nasty kernel panic which happened all the time that Apple fixed in 10.5.3, but even now, using an firewire iSight at the same time as firewire audio is still asking for all sorts trouble.)

So.. I may fire up this tool when I need to do some intense CPU crunching, such as VisualHub video re-encoding, however in general use I will leave it at the factory clock because I need my audio.

These USB/Firewire problems all result from the fact, that Apple uses the bus clock, instead of a realtime clock. All Intel Macs have a HPET timer, which is an excellent source of realtime. These HPETs (High Precision Event Timer) are a system programmers dream. They even feature aperiodic interrupts, which are helpful when you need to play two multimedia streams with different fps, e.g. PAL 25 fps and NTSC 29.97 fps.

Bus clock timing is more precise in the nanosecond resolution. This is needed, when profiling and tuning up your code. However, every game developer knows, that you shouldn't use it as a real time clock source.

Many computers, e.g. most notebooks, underclock themselves when idle to save power. AMD multi-core-processors can slow down each core in a single package seperately. This results in a different "real time" on each core.

So hands off timing sources, that derive from the bus clock as local APIC interrupts or the RDTSC instruction.

Windows, Linux and even most Hackintosh/OSX86 flavors use pmtimer and/or HPET. Hackintosh wouldn't run on AMD machines, if it uses a bus clock source because of different clock speeds in each core.

I don't know, why Apple decided to waive the rule "Get real time only from a source, that actually supplies real time". One idea might be, that it is easier to maintain the same code base for Intel and Power Macs. While Power Macs have real time sources, they are quite different from Intel Macs.

In Mac OS X the system call "clock_get_uptime" is often used as a timing source, which is dependent from the bus clock and returns wrong information. E.g. see here for problems.

These problems occur on MacBooks. They are quite obvious, since Notebook computers face heat problems, when running under full load, e.g. video encoding. The processor then throttles by leaving out bus cycles and/or the thermal management reduces the bus clock. Both lead to deviation of real time and bus time and thus USB/Firewire devices get confused.

See also this thread on the apple support forum and this one too.

Apple really caused themselves a lot of unneccessary problems. These do normally not occur on Desktop computers (iMac and MacPro), because neither underclocking nor throtteling is needed under normal circumstances. However, if you change your bus clock, exactly the same problems occur as on Macbooks.

-Christoph

 

[christoh]

I am kind of surprised this program doesn't allow you to lock your other bus speeds down and only change the FSB (or the multiplier). I would be really wary of overclocking the bus with out that mechanism as your data could eventually get corrupted.

It is cool that there is a means to OC Intel Macs now.

All Mac Pros use ICS932S4xx clock chips. These chips generate 3 independent frequencies.

• 48 MHz for USB. It cannot be changed.
• 100 MHz for PCIe, also generates 33 MHz for PCI by dividing the 100 MHz by 3. This can be changed programmatically
• Bus clock of 333.33 MHz or 400 MHz. Can be changed programmatically.

The bus clock is the speed the front side bus runs with. Since modern FSBs have the quadruple width of the original Intel FSB. Intel says a 400 MHz FSB runs "effectively" at 1600 MHz. I call it "marketing bus clock".

The CPU's speed is determined by the multiplier. A 2.8 GHz CPU has a multiplier of 7. So, if the bus speed is 400 MHz, the CPU runs at 7 * 400 MHz = 2800 MHz.

Memory is also running at bus clock speed. Since DDR2-RAM answers twice per clock cycle, the effective speed is always 2 * bus clock. In stock speed this is 800 MHz.

As you may have noticed, most users here cannot overclock beyond a certain speed, because of limitations in their FB-DIMMs. The CPU would be able to allow even more.

The only way to speed up the CPU while leaving RAM unchanged is to change the CPU multiplier. Most Intel CPUs have this multiplier locked. It cannot be changed. An exception are extreme editions. These have an X in their model number.

The only way to speed up machines with the Intel L5xxx and E5xxx processors is to increase the bus clock. So, that's, what ZDNet Clock does. If you set the bus clock to 450 MHz from 400 MHz you get the following:

FSB marketing speed: 450 MHz * 4 = 1800 MHz (stock speed 1600 MHz)
Memory speed: 450 MHz * 2 = 900 MHz (stock speed 800 MHz)
CPU speed: 450 MHz * 7 = 3150 MHz (stock speed 2800 MHz)

These three speeds are coupled and cannot be changed idependently, except on those machine with an X5xxx processor, where you can increase the CPU multiplier.

We also experimented with the PCIe clock. But switching from 100 to 103 MHz resulted in a coruppted display immediately. This is obviously a syncing problem between the PCIe bus and the PCIe northbridge.

-Christoph

 

[DaBrain]

How exactly does this relate to the 3G iPhone? I was under the impression that all news had to relate, even tangentially so, to the 3G iPhone.

Your not the only one! The obsession here over the iPhones is crazy I tell ya!--)))

 

[rxse7en]

Wonder if this is worth it for me. I've got a 2.66GHz Mac Pro, and I'm fairly happy with the speed but I've been wondering about upgrading to a 3.0GHz Mac Pro..

My 2.66 Mac Pro has been running at 3.1ghz since yesterday. Aside from the clock and restart issues this is a real winner. I'm gonna run some encoding tests in Visual Hub today and see what we get.

 

[IlluminatedSage]

Now the Mac Pro can claim some true cred as a gaming machine.

Apple will only gain real cred when it releases a mid range tower. these towers are just too high priced.

apple is missing a huge market demographic and with their sales increasing the "mythical midrange tower" continues to a big mistake for apple to continue avoid making.

midrange macs sold well in the past, to continue to drive people to imac's which use a LAPTOP cpu, is ridiculously lame.

 

[diamond.g]

All Mac Pros use ICS932S4xx clock chips. These chips generate 3 independent frequencies.

• 48 MHz for USB. It cannot be changed.
• 100 MHz for PCIe, also generates 33 MHz for PCI by dividing the 100 MHz by 3. This can be changed programmatically
• Bus clock of 333.33 MHz or 400 MHz. Can be changed programmatically.

The bus clock is the speed the front side bus runs with. Since modern FSBs have the quadruple width of the original Intel FSB. Intel says a 400 MHz FSB runs "effectively" at 1600 MHz. I call it "marketing bus clock".

The CPU's speed is determined by the multiplier. A 2.8 GHz CPU has a multiplier of 7. So, if the bus speed is 400 MHz, the CPU runs at 7 * 400 MHz = 2800 MHz.

Memory is also running at bus clock speed. Since DDR2-RAM answers twice per clock cycle, the effective speed is always 2 * bus clock. In stock speed this is 800 MHz.

As you may have noticed, most users here cannot overclock beyond a certain speed, because of limitations in their FB-DIMMs. The CPU would be able to allow even more.

The only way to speed up the CPU while leaving RAM unchanged is to change the CPU multiplier. Most Intel CPUs have this multiplier locked. It cannot be changed. An exception are extreme editions. These have an X in their model number.

The only way to speed up machines with the Intel L5xxx and E5xxx processors is to increase the bus clock. So, that's, what ZDNet Clock does. If you set the bus clock to 450 MHz from 400 MHz you get the following:

FSB marketing speed: 450 MHz * 4 = 1800 MHz (stock speed 1600 MHz)
Memory speed: 450 MHz * 2 = 900 MHz (stock speed 800 MHz)
CPU speed: 450 MHz * 7 = 3150 MHz (stock speed 2800 MHz)

These three speeds are coupled and cannot be changed idependently, except on those machine with an X5xxx processor, where you can increase the CPU multiplier.

We also experimented with the PCIe clock. But switching from 100 to 103 MHz resulted in a coruppted display immediately. This is obviously a syncing problem between the PCIe bus and the PCIe northbridge.

-Christoph

Is there no way to lock the memory bus speed down? I know on some motherboards you can choose to run the memory at 1:1, 5:4 or some other variation. That would minimize the impact of upping the bus speed on the RAM. It also happened to be what I was thinking about locking down, but for some reason forgot to mention it. Besides I am not sure how else people would be hitting higher clock speeds as memory stability tends to hold everyone back.

 

[sbarton]

I don't recommend this for your average Mac user, but its' a fun tool for those that want to experiment.

 

[sbarton]

Apple will only gain real cred when it releases a mid range tower. these towers are just too high priced.

apple is missing a huge market demographic and with their sales increasing the "mythical midrange tower" continues to a big mistake for apple to continue avoid making.

midrange macs sold well in the past, to continue to drive people to imac's which use a LAPTOP cpu, is ridiculously lame.

Amen brother. Besides, the Mac Pro would make a horrible gaming machine. I could put something together that would destroy it (for gaming) for less than $1200. The Mac Pro is designed for professionals, but seriously, I don't think Apple is selling a shed-load of these machines either. Does anyone have actual numbers for Mac Pro's sold last year or does Apple even release those figures? On a side note, as one poster mentioned earlier, *if* you are looking for an 8 way media monster then you would be a fool not to look at the MacPro. You can't touch it price-wise through Dell or anyone else unless you start buying your own components.

I think one of the main reasons we haven't seen a mid-range mac is due to the slow adoption of EFI by the PC industry. It is primarily because of this, we still have Nvidia and ATI having to make custom versions of their video cards for Macs vs Mac's being able to use the same video cards as PC's with just an OS X driver. By far the biggest advantage of a Mid-range Mac would be to have the ability to upgrade the graphics subsystem. If you can't do that anyway, then perhaps the rational at Apple is to wait until it is feasible for end-users before releasing one.

 

[brent1a]

Elaborate?

I don't recommend this for your average Mac user, but its' a fun tool for those that want to experiment.

You're not the onlyh one to be cautious. Would you care to elaborate? I've OC'd pc's before and only had to worry about heat issues. If the pc crashed all I had to do was restart and knock off the OC or juct knock it down.

Is there aything else to be worried about on my Mac Pro than overheating and the fast clock?

 

[brent1a]

What about aftermarket clock software?

I have a sort-of-not-really solution for the fast clock issue:
Download and install MenuCalendarClock for iCal then disable the system clock display and enable just the Day/Date/Year display on MenuCalendarClock. Therefore you do't have to see that your time is 3 hours fast or whatever........but then this only works until your clock tells you that it's 2 days ahead.
Crap....this really doesn't have a fix except restarting your whole system.

 

[sbarton]

You're not the onlyh one to be cautious. Would you care to elaborate? I've OC'd pc's before and only had to worry about heat issues. If the pc crashed all I had to do was restart and knock off the OC or juct knock it down.

Is there aything else to be worried about on my Mac Pro than overheating and the fast clock?

Probably not. Just acknowledgment of the fact that crossing over into the world of over-clocking may pose issues beyond the expected ~out of the box and everything works~ Mac experience. Your typical Mac user may not know how to adequately test a particular over-clock for long-term stability. The buffered ram in the Mac Pro is going to be the main limiter in over-clocking those systems. The cooling as others have pointed out, is already pretty robust.

 

[Schizoid]

2.8 to 3.0

Anyone actually got any gain from this overclocker app?

Been running my 2,8 at 3.0Ghz for a while now with no problems, but did a VisualHub compression on a DVD-size movie and here's the results!
3Ghz: 40m:29s
2.8Ghz: 40m:25s

So we'll call that negligible!
Begs the question - are we actually overclocking our Mac Pros with this application?
Do Apple's own top-speed processor Macs yield a similar result?

 

[Photek]

taking my 8core 2.8ghz to a 3.0ghz and then giving it a very heavy C4D scene to render resulted in black restart screen...

might stay away from this voodoo

 

[p33]

Nice to know it's there, but no way I am going to try this on my 2.8

A marginal speed upgrade with big risks. It might be stable for some time, but I'm sure it will shorten the life of the CPU considerably.

Some years back I OC'ed a brand new Celeron 300 to 450 MHz (you can figure out when it was roughly). That box was working 24x7 several years in a row for me and it's still working to the best of my knowledge. So, I don't really know what your confidence is basing on...

I agree though that the speedup in your case would be marginal and it's simply not worth it.

 

[Toe]

Been running my 2,8 at 3.0Ghz for a while now with no problems, but did a VisualHub compression on a DVD-size movie and here's the results!
3Ghz: 40m:29s
2.8Ghz: 40m:25s

So we'll call that negligible!
Begs the question - are we actually overclocking our Mac Pros with this application?

That certainly is curious. Have you looked at your logs from that time period to see if there were any random errors? I am wondering if the over-clocked CPU had some faults, had to compensate for them, and lost time as a result...?

A faster runner who stumbles more often is not necessarily the best bet.

 

[Schizoid]

That certainly is curious. Have you looked at your logs from that time period to see if there were any random errors? I am wondering if the over-clocked CPU had some faults, had to compensate for them, and lost time as a result..

Nope no errors - just no difference in speed at all
Been watching the system.log - nothing out of the ordinary
My question really was what's the actual difference between a 2.8Ghz and a "real" 3Ghz?

 

[brent1a]

taking my 8core 2.8ghz to a 3.0ghz and then giving it a very heavy C4D scene to render resulted in black restart screen...

might stay away from this voodoo

I took my 2.8Ghz Mac Pro to 3080Mhz (3.08Ghz?) and ran Cinebench R10 twice and the result was I got about 3% increase in performance....haven't seen any black restart screen just yet.

 

[gnasher729]

Bus clock timing is more precise in the nanosecond resolution. This is needed, when profiling and tuning up your code. However, every game developer knows, that you shouldn't use it as a real time clock source.

Many computers, e.g. most notebooks, underclock themselves when idle to save power. AMD multi-core-processors can slow down each core in a single package seperately. This results in a different "real time" on each core.

So hands off timing sources, that derive from the bus clock as local APIC interrupts or the RDTSC instruction.

I'd say you are blaming Apple for something that is not their fault at all.

On AMD processors, the rdtsc instruction has real problems because the timers for multiple CPUs drift apart, making it very hard indeed to use rdtsc for any clock purposes (some linux guys probably know all the details). Intel CPUs don't have that problem. And since Apple doesn't use AMD CPUs, that AMD problem is irrelevant. If an AMD Hackintosh has problems because of that, tough. Not Apple's fault.

The rdtsc assembler instruction itself has the problem that it counts motherboard clocks (multiplied by the maximum multiplier of the processor) and doesn't directly give a time, so people usually use mach_absolute_time. Apart from that, rdtsc is _not_ influenced by what you call "underclocking" in a notebook, which is really reducing the clock multiplier. In every second, rdtsc is increased by a number equal to the maximum clock speed of the processor, no matter what the actual clock speed is.

mach_absolute_time, on the other hand, returns time in nanoseconds, and it does that correctly on multi-CPU systems: First, it uses rdtsc to read the current clock. Then it subtracts a correction factor that can be different per CPU. Then it multiplies by a factor to change clocks to nanoseconds. It then adds another correction factor, and finally checks whether it is still running on the same CPU it was running on when the call started - if not, then it does the whole calculation again.

Here is the point where this overclocking tool is missing something: As it changes the motherboard clock speed and therefore the rate at which rdtsc is counting, it should change the multiplier used by mach_absolute_time and change the correction factors that are subtracted and added so that mach_absolute_time can continue to give the correct results. It doesn't do that. Therefore mach_absolute_time will give incorrect results. There is no way you can blame Apple for that - I am sure that if Apple provided an overclocking tool then they would make this work, and the stupid rebooting would probably not be needed.

 

[VanArnam]

Mac Pro 2.1: 8 core 3.0GHz 1333--> 3.3 GHz/1466

Stable for an hour now, running 3x SMP Folding@Home. Fans were at 2000rpm already, temps went up by 2 degrees Celsius. neat!

 

[Digitalclips]

I took my 2.8Ghz Mac Pro to 3080Mhz (3.08Ghz?) and ran Cinebench R10 twice and the result was I got about 3% increase in performance....haven't seen any black restart screen just yet.

I cannot get past a 6% increase on a new 2.8 8 Core Mac Pro without a KP. I have an extra 4 GIGs of RAM from a supposedly good manufacturer and also both the standard ATI graphics card plus a new 8800 both running 30" ACDs. I wonder which is the problem, the RAM or the cards?

 

[brent1a]

I cannot get past a 6% increase on a new 2.8 8 Core Mac Pro without a KP. I have an extra 4 GIGs of RAM from a supposedly good manufacturer and also both the standard ATI graphics card plus a new 8800 both running 30" ACDs. I wonder which is the problem, the RAM or the cards?

From reading the other posts I'd say the ram is the problem. I have an extra 2gb's of iRam from Newegg but I haven't had any problems so far.

8800 gfx card, huh? I'm jonesing for the ATI 3870 real bad, myself.

 

[kabunaru]

Apple really caused themselves a lot of unneccessary problems. These do normally not occur on Desktop computers (iMac and MacPro), because neither underclocking nor throtteling is needed under normal circumstances. However, if you change your bus clock, exactly the same problems occur as on Macbooks.

-Christoph

So are you saying that PowerPC Mac laptops ran cooler and had better heat management?

 

[idyll]

My 2.66 Mac Pro has been running at 3.1ghz since yesterday. Aside from the clock and restart issues this is a real winner. I'm gonna run some encoding tests in Visual Hub today and see what we get.

What are the clock and restart issues? Sorry I didn't read all of the thread

 

[brent1a]

What are the clock and restart issues? Sorry I didn't read all of the thread

System time (clock) runs fast, & if you want to disable the overclocking issue you have to restart your mac.