Processor upgrade 1.1 macpro. is it worth it?

@brettallica... I'm not sure why you would want to replace what you have now (4 cores) with 4 cores. Why not just get one new quad now, stick it in a drawer and then do the upgrade when you get the second?

The technical reason I'd do it that way is (yes, I'm a practicing electrical engineer) that the lack of a processor has real potential to cause problems that will be very difficult to unravel. Things may work out just fine, but here's the deal. There are 177 connections to each processor, and each connection has been finely balanced to electrical conditions where BOTH processors are installed. The proper term is impedance matching. What happens when one of the processors is missing is that the 177 pins are terminated not by a hunk of silicon, but by air--and the system was designed to have a hunk of silicon there, not air. The unexpected impedance presented by air will result in degradation of the signals in the whole processing system and since the degraded signals are literally deformed, there is an increased chance that the remaining processor will receive unrecognizable and thus corrupted data. Which of course leads to crashes and kernel panics and all sorts of headaches that you will be hard pressed to track down.

This is not to guarantee that it won't work with 1 processor, but you will certainly be operating with less design margin than was intended. Since you won't get any performance boost until you replace both processors, it seems like an unnecessary tearing apart of your computer!

The 5355s use up to 2x more electrical power than the 5150s -- there are twice as many processors cranking away! Since the cooling system hasn't changed, they will run hotter for the same fan speed.

The 50,000 dollar question is will the Processor preferences pane - the one that's in the Developer's Tools - disable the 4-core Xeon in the same fashion as it does the 2-core now? I currently run the system on one processor when I'm doing an overnight torrent seed on files and it really is a Godsend for keeping the power consumption down.

Has anyone tried this yet?

Hey Kim,

Thanks for that detailed answer! This is the exact type of response I sought when I posed my question originally. It sounds like you know a thing or two about this stuff. I definitely wanted to know if it was possible, but I also wanted to know why it was or was not. I am gathering from your post that "it is possibly doable, but not necessarily recommended."

To answer your question you asked of me about replacing four cores with four cores, the answer is purely economics: two of these processors will cost a person $600 (ish), while one would cost $300 (ish). That's quite a difference in price; the extra money I can allocate to other expenses for the time being while I get by with one quad processor (or so I had planned anyway). I have read on various forums what people think about two processors with two cores vs. one processor with four cores, and there is certainly no definitive or overly persuasive answer that I have found that says one would perform better than the other. I haven't read a response like yours in all of that research, though. From your response it sounds like there are other, deeper, more technical reasons why the system as a whole may not perform better with an individual quad processor, leaving an empty processor socket.

Kim Hansen1 wrote:

@brettallica... I'm not sure why you would want to replace what you have now (4 cores) with 4 cores. Why not just get one new quad now, stick it in a drawer and then do the upgrade when you get the second?

The technical reason I'd do it that way is (yes, I'm a practicing electrical engineer) that the lack of a processor has real potential to cause problems that will be very difficult to unravel. Things may work out just fine, but here's the deal. There are 177 connections to each processor, and each connection has been finely balanced to electrical conditions where BOTH processors are installed. The proper term is impedance matching. What happens when one of the processors is missing is that the 177 pins are terminated not by a hunk of silicon, but by air--and the system was designed to have a hunk of silicon there, not air. The unexpected impedance presented by air will result in degradation of the signals in the whole processing system and since the degraded signals are literally deformed, there is an increased chance that the remaining processor will receive unrecognizable and thus corrupted data. Which of course leads to crashes and kernel panics and all sorts of headaches that you will be hard pressed to track down.

This is not to guarantee that it won't work with 1 processor, but you will certainly be operating with less design margin than was intended. Since you won't get any performance boost until you replace both processors, it seems like an unnecessary tearing apart of your computer!

Thanks, harryb! You've been quite helpful and insightful, too. I've begun to now seriously contemplate the value of going with two x5365 chips vs. two x5355. Part of me just wants to bite the bullet and go for it with the x5365, but dang they are so much more expensive! Then again, there is significantly more processor bandwidth in the x5365.

harryb wrote:

Okay Brett misunderstood sorry mate.

Running one CPU is not something I would do personally, but having said that Google shows many reports of folks doing just that, particularly this link and the thread by Purple Patch:-

http://www.insanelymac.com/forum/lofiversion/index.php/t123604.html

Dell,HP and IBM of course offer single CPU X5355/5365 so there would be not great danger in it that I could see. Only thing to be doubly sure of when it comes time for the additional CPU< get the same step model.

Be interesting to see performance differences between the 2.66GHz Dual Core and the X5365 3GHx Quad with only one processor.

Steve pleased you got the Kext Utility up and goind. Does it shows 'Processors 2 x 2.66GHz Quad-Core Intel Xeon'?

Are you going to overclock your 2.66 GHz machine? If you aren't then don't spend the extra money for the 5365. Remember that the 3.0 GHz clock speed quoted is the highest speed at which the processor will run reliably, but if the clock in your computer remains set to 2.66 GHz, the 5365 will run at 2.66 GHz and will perform EXACTLY the same as a 5355 at 2.66 GHz. It's like a Ferrari in a school zone... sure, the 5365 could go faster, but the speed control (clock speed) is set at 2.66 GHz so that's how fast it actually will go.

FWIW, I paid $200 total for my matched pair of 5355s. Insane deal yes, but if you are patient you can find a deal.

I'd try it for you, but it's no longer in /Developer/Extras after installing Xcode 4.2 so I can't!

Wow, that is insane! I had no idea that one couldn't just pop in the 5365 and get the full speed out of them without overclocking the system first. I had just assumed that since so many people were putting those 5365 chips in these computers that it was just a simple install and nothing else was needed in order to realize that full clock speed. Consider that avenue almost officially closed, as I won't be doing any "serious" modifying to my machine that way, unless it is insanely easy. I had a friend of mine overclock my 1 x 1.25 GHz G4 MDD to 1.5 GHz a while back, and it involved messing with the traces and/or resistors on the logic board. If anything like that is needed to overclock the Mac Pro, I would definitely not want to go down that road.

The hunt for X5355 now commences in earnest!

Kim Hansen1 wrote:

Are you going to overclock your 2.66 GHz machine? If you aren't then don't spend the extra money for the 5365. Remember that the 3.0 GHz clock speed quoted is the highest speed at which the processor will run reliably, but if the clock in your computer remains set to 2.66 GHz, the 5365 will run at 2.66 GHz and will perform EXACTLY the same as a 5355 at 2.66 GHz. It's like a Ferrari in a school zone... sure, the 5365 could go faster, but the speed control (clock speed) is set at 2.66 GHz so that's how fast it actually will go.

FWIW, I paid $200 total for my matched pair of 5355s. Insane deal yes, but if you are patient you can find a deal.

I'm running 5365 and they're running 3Ghz (at least showing it in about this mac) without any overclocking, simply 'dropped in'.

Regardless, if doing it again I'd be buying 5355. It's simply not worth the $ for the minimal performance difference. In my experience 5365 also generate more heat and therefore noise, which again for me isn't worth it for the minimal performance difference.

Interesting. I have no idea what about-your-mac uses to determine what to display though. What do your benchmarks look like? I could be very wrong here--I understand the motherboard has a programmable clock chip (which means theoretically it can be overclocked with software, there are plusses and minuses with that) so it's possible. The clock speed delta from 2.66 to 3.0 GHz is about 13%, which isn't anything near the bump you get from doubling the number of processors. An 8 core 3 GHz machine would perform like a 9 core 2.66 GHz machine. Unless the memory bus is also running faster in which case I would expect to see an improvement that might be noteworthy.

If your 64 bit geekbench is near 12,000 I'd bet you are at 3 GHz and I'm all wet! It seems many/most people with 2.66 GHz machines are seeing 10,400-10,500 ish numbers with no other applications running except geekbench.

But more importantly, did your KPs go away after redoing the thermal compound?

My geekbench is 11484 with an app or two running, so I presume a little wetness is likely? I hadn't heard of what you mention regarding the need to overclock or motherboard needing to be altered via software.

Still having the occasional KP after renewing compound. Waiting for one more, then I'll be uninstalling Little Snitch and seeing how things go (I'm not convinced they've fixed their KP issues regardless of the last update). After that if I'm still getting KPs I'll be removing the CPUs again to see what sort of contact I've had this time between the CPUs and heatsink (by looking at the pattern left on the thermal compound, can't think of any other way to do it). After reapplying for what will be a third time, I'll try rolling back OSX to 10.6.7 since others with the same sort of issue have had no problem running that revision of OSX, with problems starting as of 10.6.8 and Lion.

Personally I think some sort of subtle change in OSX combined with my hardware is most likely the cause of my KP issues at present. It seems very unlikely that it is temperature related to me. My machine certainly runs hotter than stock, but I've had KPs overnight with my machine at idle running 1800RPM on all fans (via SMC fan control) providing temps lower than I've had them with original processors (monitoring via hardware monitor). At present I'm experiencing a KP every 3 days or so, so less frequently than I originally had.

So assuming I get to the point of applying thermal compound for a third time, how much Arctic Silver is safe to use? I obviously don't want it oozing out anywhere if it warms up and melts out, so how thick can/should it be applied? My memory of the original thermal pad (by which I mean the stock 'screened' version from Apple) was about 2mm or 3mm thick. Arctic Silver seems a little too gooey and runny to hold in place at that sort of thickness once up to operating temperature.

Yep. I appear to be all wet! It is surprising in some ways, not so in others. Perhaps I'll do some research to see what intel did with the clovettowns.

As to the amount of thermal compound, you want to use the minimum possible... But you need to use "enough". The thing is you are trying to move heat from the surface of the processor to the heat sinks and the challenge is to eliminate the air between the two because the two metal surfaces will never be in perfect contact with each other. The thermal compound is intended to fill the tiny gaps that exist because the two surfaces are not perfectly flat. And while the compound conducts heat much much better than air, it isn't perfect. So the thicker the layer of compound, the further the heat has to travel before it gets to the heat sink. It's only measured in thousands of an inch, but the resistance to the flow of heat means the processor can't get rid of it's heat as quickly.

Think of it as a traffic jam. You have a truck load of heat at one end trying to get through traffic to the heat sink. But one lane of the freeway is closed which restricts the flow of traffic... As the length of the lane closure gets longer, it rakes longer for any single truckload of heat to get through because fewer trucks can pass through the area with the lane closure in any given period of time. Meantime, more truckloads of heat keep arriving at the front and they begin to stack up. Having no thermal compound (air) is much like closing the road all together. Ideally your compound completely covers the faying surfaces (parts in contact) completely with no air bubbles, no foreign matter and 100% of the top if the processor covered. That makes the road as wide as possible.

After that, you want it as thin as possible to minimize the distance the heat has to travel through the compound.

Long answer eh? If you redo the job again, look carefully when you take off the heat sinks. If the entire top surface of the processor was covered, you have enough compound. If there are clean spots, use a tiny bit more. If it was oozing over the side, use less. I redid my job a couple of times, changing compounds to see what happened. It looked to me, both visually and from the temperatures I observed that with AS5 using just a bit more than they show on their website and applying it in a horizontal line resulted in complete coverage, minimal oozing, and reasonable temps. At 2.66 GHz though! Your truckloads of heat are arriving faster than mine so they're stacking up faster and deeper.

Well Kim Intel disagree with that twice as much power stating both the 5100 and 5300 use very little power, and of course the number of processors has not changed. They both use two processors.

http://www.multicoreinfo.com/research/papers/whitepapers/multicore_virtualizatio n.pdf

And the overheating thing is an individual issue depending a lot on location, venue temperature, cleanliness and dust, quantity of Arctic Silver used or lack of the compund. Have upgraded 2GHz machines to 5100 2.66Ghz, and to 2.,66GHz 5355 and have yet to see any change in temperatures.

Whenever have to take fan units out, always remeove the little centre patch and lubricate the fans with a drop or two of WD40.

Cutrently my Mac Pro ius running with fans set at 499rpm through smcFanControl, with ambient temperature currently at 17ºC, other readings are CPU A 29º, CPU B 31º, Expansion slot 17º, HDD Bay 1 20º and HDD Bay 2 22º,

Neat paper Harryb -- virtualization and the performance of the hypersupervisor along with its impact on operating cost are a big concern in data centers for sure.

I was refering to Intel's spec sheets, which indicates the max total dissipated power of the 5150 is 65W http://ark.intel.com/products/27218 and the TDP of the 5355 is 120W http://ark.intel.com/products/28035

Not quite 2x, but when operating at full load the spec sheet says the 5355 will produce more heat than the 5150. For a given cooling system that is unchanged (lots of factors!), when you put more heat into the system, everything will be warmer.

I positioning myself to purchase a pair of NEW X5355 for $250. My source actually has four of these available and would sell them all to me for $500. I'm thinking about buying all four and distributing the other pair here for those who're looking to make this mod on their rigs.

Kim Hansen1 wrote:

<...snip...>

FWIW, I paid $200 total for my matched pair of 5355s. Insane deal yes, but if you are patient you can find a deal.