MP All Models Mac Pro CPU Compatibility List

[fhturner]

Yep

LOL...makin' me work here! Okay, so the socket # trails behind the Xeon E5 family # by 1? Like so...?

E5 v2 = FCLGA2011("-1")
E5 v3 = FCLGA2011-2
E5 v4 = FCLGA2011-3

So each family revision gets a new socket...bummer. Certainly limits going forward on machines based on these (like Mac Pro, but also PC builds, I guess). We must've been spoiled being able to put Westmere chips in Nehalem-based boards/sockets...

 

[ActionableMango]

Okay, so the socket # trails behind the Xeon E5 family # by 1? Like so...?

It is inconsistent.

In any case, the socket incompatibility is only part of the story. Haswell uses a completely different chipset than Ivy Bridge, so the nMP doesn't have the right chipset anyway.

 

[Theophany]

Upgraded the X5660 in my 4,1>5,1 last night to a W3690 and took the 16GB of RAM up to 32GB. It was fairly straightforward, but for anybody who runs into the issue of really goddamn tight hex screws I found using a torx driver instead of a hex driver worked really well. I was worried about stripping the heads as the hex driver kept slipping but had no such issue with a torx driver.

Also, when tightening the nuts back on, I found tighten until you can feel the bolts start becoming difficult to turn and then unscrew about a quarter turn. When you look through the heatsink all the bolt heads should be level.

Temps were ~83'c under Prime95 last night and after running the FPU torture test overnight, this morning they have stabilised at ~76'c, so I think the Arctic Silver 5 has burned in properly now. Fans still whisper quiet too. GeekBench 4 scores have increased 14% in single core and 13% in multicore.

 

[Neodym]

Complemented the MP 1,1/2,1 section with stepping / Watt information and corrected a few values there.

If anyone else wants to edit the table: I found it useful to copy the part I wanted to work on to Excel, do my changes there (highlighting added or changed cells) and only then actually enter the changes into the BBCode of ActionableMango's post #1.

 

[joebclash]

Upgraded the X5660 in my 4,1>5,1 last night to a W3690 and took the 16GB of RAM up to 32GB. It was fairly straightforward, but for anybody who runs into the issue of really goddamn tight hex screws I found using a torx driver instead of a hex driver worked really well. I was worried about stripping the heads as the hex driver kept slipping but had no such issue with a torx driver.

Also, when tightening the nuts back on, I found tighten until you can feel the bolts start becoming difficult to turn and then unscrew about a quarter turn. When you look through the heatsink all the bolt heads should be level.

Temps were ~83'c under Prime95 last night and after running the FPU torture test overnight, this morning they have stabilised at ~76'c, so I think the Arctic Silver 5 has burned in properly now. Fans still whisper quiet too. GeekBench 4 scores have increased 14% in single core and 13% in multicore.

What torx size did you use?

 

[Theophany]

What torx size did you use?

I'll have to check my set to be absolutely sure, but it was roughly the same size head as a 3mm hex.

ETA: Checked last night - it was a T20.

 

[ActionableMango]

Complemented the MP 1,1/2,1 section with stepping / Watt information and corrected a few values there.

If anyone else wants to edit the table: I found it useful to copy the part I wanted to work on to Excel, do my changes there (highlighting added or changed cells) and only then actually enter the changes into the BBCode of ActionableMango's post #1.

Wow, there are a lot more entries now. Thanks for contributing!

 

[yukdave]

The X5690 supports 288GB of RAM (with 4 slots of 16GB you get 64GB)

The W3690 support 24GB of RAM

Assuming a 2009/2010/2012 Single Proc Mac Pro


https://ark.intel.com/products/5257...r-X5690-12M-Cache-3_46-GHz-6_40-GTs-Intel-QPI

https://ark.intel.com/products/5258...r-W3690-12M-Cache-3_46-GHz-6_40-GTs-Intel-QPI

 

[h9826790]

The X5690 supports 288GB of RAM (with 4 slots of 16GB you get 64GB)

The W3690 support 24GB of RAM

Assuming a 2009/2010/2012 Single Proc Mac Pro


https://ark.intel.com/products/5257...r-X5690-12M-Cache-3_46-GHz-6_40-GTs-Intel-QPI

https://ark.intel.com/products/5258...r-W3690-12M-Cache-3_46-GHz-6_40-GTs-Intel-QPI

Those info is not updated. E.g. When W3690 released, there was no 32GB stick. So, this option is not covered.

The W3690 is already demonstrated able to support at least 3x16 + 8 =56GB RAM.

X5690 able to support some 32GB stick, but no one figure out the rule yet. So far only know can support up to 5x32GB sticks to get 160GB RAM in total (for dual CPU option). No one ever report any test about 32GB stick on the single CPU model.

So, X5690 may able to support more than 64GB on a single processor cMP. And W3690 also can at least support 56GB RAM.

 

[Theophany]

The X5690 supports 288GB of RAM (with 4 slots of 16GB you get 64GB)

The W3690 support 24GB of RAM

Assuming a 2009/2010/2012 Single Proc Mac Pro


https://ark.intel.com/products/5257...r-X5690-12M-Cache-3_46-GHz-6_40-GTs-Intel-QPI

https://ark.intel.com/products/5258...r-W3690-12M-Cache-3_46-GHz-6_40-GTs-Intel-QPI

I've got 32GB in my W3690.

 

[Synchro3]

The X5690 supports 288GB of RAM (with 4 slots of 16GB you get 64GB)

The W3690 support 24GB of RAM

Assuming a 2009/2010/2012 Single Proc Mac Pro


https://ark.intel.com/products/5257...r-X5690-12M-Cache-3_46-GHz-6_40-GTs-Intel-QPI

https://ark.intel.com/products/5258...r-W3690-12M-Cache-3_46-GHz-6_40-GTs-Intel-QPI

These Intel product pages are outdated. I have 48 GB RAM in my W3690 Mac Pro. Max RAM for W3690 is 56 GByte (3 x 16GB + 8GB).

 

[ActionableMango]

The X5690 supports 288GB of RAM (with 4 slots of 16GB you get 64GB)

The W3690 support 24GB of RAM

Assuming a 2009/2010/2012 Single Proc Mac Pro


https://ark.intel.com/products/5257...r-X5690-12M-Cache-3_46-GHz-6_40-GTs-Intel-QPI

https://ark.intel.com/products/5258...r-W3690-12M-Cache-3_46-GHz-6_40-GTs-Intel-QPI

I appreciate this, but what Intel lists as a specification is often different than what we can do in our Mac Pros. In this case I am confident that what Intel states is not correct for our Mac Pros.

 

[rjtiedeman]

I appreciate this, but what Intel lists as a specification is often different than what we can do in our Mac Pros. In this case I am confident that what Intel states is not correct for our Mac Pros.

I have 6 core cMP 5.1 with X5690 - 3.46GHz with 32GB. Would upgrading the ram to 48GB or 64GB show any performance improvement? I use it for CAD renderings that rely on the processor.

 

[kschendel]

I have 6 core cMP 5.1 with X5690 - 3.46GHz with 32GB. Would upgrading the ram to 48GB or 64GB show any performance improvement? I use it for CAD renderings that rely on the processor.

Probably belongs in a separate thread ...

The rather unhelpful answer is "It depends on whether you're short of memory." If you are, then yes. Very occasionally, adding memory when there are no overt signs of memory pressure can speed things up, but only if the application can detect or be configured to use the extra memory to reduce the amount of work it needs to do, in some way (eg partial results caching). I've no idea whether that would apply to CAD rendering in general or whatever your rendering app is in particular.

If I had to guess in the blind, I'd probably say No.

 

[Revoxy]

wow looks like even intels curant gen CPU's still drop 20c when delidded and have the stock thermal gunk relapsed
http://www.eteknix.com/intel-core-i7-7700k-kaby-lake-delidding-analysis-make-difference/4/

The New CPU's are very different from these old high end Xeons. Those Xeons always used indium as a termal interface material not thermal glue.

 

[h9826790]

The New CPU's are very different from these old high end Xeons. Those Xeons always used indium as a termal interface material not thermal glue.

That's my understanding as well, our Xeon's IHS is soldered to the CPU, de-lid it won't make much difference on thermal conductivity.

 

[jclmavg]

I can get a Xeon E5-2673 v2, 8 Cores, 3.3 GHz, TDP 110W, for an okay price. Would this one work in the nMP? It's the same as the E5-2667 V2, but with a lower TDP.

 

[Zanr Zij]

I can get a Xeon E5-2673 v2, 8 Cores, 3.3 GHz, TDP 110W, for an okay price. Would this one work in the nMP? It's the same as the E5-2667 V2, but with a lower TDP.

2600v3 will not work on any cMP. It is 2011-3 socket ( R3 ) not compatible with 2011 socket

 

[jclmavg]

2600v3 will not work on any cMP. It is 2011-3 socket ( R3 ) not compatible with 2011 socket

It's not a v3, it's a v2.

 

[Zanr Zij]

It's not a v3, it's a v2.

Ah my misread. It should work but I'm not sure 100% Apple add support that model

 

[ActionableMango]

I can get a Xeon E5-2673 v2, 8 Cores, 3.3 GHz, TDP 110W, for an okay price. Would this one work in the nMP? It's the same as the E5-2667 V2, but with a lower TDP.

It's not been verified by anyone else that I am aware of or it would be on the list. I did a fresh look and could not find any examples in this forum, OWC doesn't offer it in their CPU upgrade program, and no Geekbench listings for that CPU in a Mac.

I would not try it unless the place selling it has a good return policy, in case it is incompatible. If they have a good return policy it may be worth a try. You would be the first as far as I can find.

 

[jclmavg]

It's not been verified by anyone else that I am aware of or it would be on the list. I did a fresh look and could not find any examples in this forum, OWC doesn't offer it in their CPU upgrade program, and no Geekbench listings for that CPU in a Mac.

I would not try it unless the place selling it has a good return policy, in case it is incompatible. If they have a good return policy it may be worth a try. You would be the first as far as I can find.

OWC never updated their 2014 list with compatible Xeons, which is a shame. Could not find any other examples either, probably no one has even tried. Based on specs it should simply work, other than the lower tdp there's no difference with the 2667.

What precisely could prevent it from working?

 

[ActionableMango]

What precisely could prevent it from working?

I couldn't say, but I can provide a historical example. The X5482 works in the 3,1. But only SLANZ. The SLBBG does not. They are the same CPU model and as far as I can tell, all of the specs are the same between the two steppings except power consumption.

There are also historical examples where lower power consumption CPUs work just fine.

This is why I think it is worth a try, but only if you can return it (or don't mind reselling it) if the CPU turns out to be incompatible.

 

[h9826790]

OWC never updated their 2014 list with compatible Xeons, which is a shame. Could not find any other examples either, probably no one has even tried. Based on specs it should simply work, other than the lower tdp there's no difference with the 2667.

What precisely could prevent it from working?

It seems the Mac Pro's firmware only allow certain CPU to boot. Even a CPU for the same socket, from the same family, but if not "listed" in the firmware, it won't work at all. I doubt if anyone can tell the exact reason apart from Apple. However, this is one of the theory that match with the result.

e.g. via 5,1 firmware hack, the 4,1 can accept more CPU.

Since the only thing do changed during the process is just the firmware. So, I will say it's a prove that the Mac Pro's firmware decide which CPU can work. In other words, it's also the reason that preventing some CPU to work.

 

[flowrider]

^^^^Case in point the X5687.

Lou