MP 7,1 CPU upgrade in 2019 Mac Pro

Snow Tiger

macrumors 6502a
Dec 18, 2019
579
362
Well , to those still following my progress I regret to inform you that for the second time my Gold 6212U upgraded MP7,1 has failed to survive a NVRAM reset . It is now offline again . Just a black screen , no power to the mouse / keyboard and consumes around 100 W at the wall continuously .

Looks like for now the only processor upgrades that work with the MP7,1 are the W-series Xeons .
 

Snow Tiger

macrumors 6502a
Dec 18, 2019
579
362
A pity, but thank you so much for testing AND sharing your results!
I struggled with this for the last two weeks and I'm quite certain Apple security is behind the upgrade block .

What layer of security , I'm not certain . Maybe that darn encrypted SSD . For what its worth , I attempted three times today to revive the firmware . No go .

Tsialex will figure it out . He has complete dumps of the System . Whether he can undo the block remains to be seen .

The 6212U has support in the firmware of the MP7,1 . At least we learned that much .

But at the moment , the best we can do is install another W series Xeon . When they get cheap enough , it'll be a better deal than Apple for a self install .
 

erroneous

macrumors member
Jul 25, 2004
46
15
I struggled with this for the last two weeks and I'm quite certain Apple security is behind the upgrade block .
Thanks for trying some crazy mad science!👨‍💻

Did the pci lane difference on the gold Xeon (48 vs the W range’s 64) show up at all in the pci slot/pool balance uni?
 

Snow Tiger

macrumors 6502a
Dec 18, 2019
579
362
Thanks for trying some crazy mad science!👨‍💻

Did the pci lane difference on the gold Xeon (48 vs the W range’s 64) show up at all in the pci slot/pool balance uni?
Never got to the point I was concerned about it ! I was more concerned with the processor upgrade surviving more mundane operations . Funny thing about this upgrade was that it was stable and fairly high performance for as long as I wanted , until the NVRAM reset thing . But I will say something about the Gold 6212U - she's a really nice piece of silicon in an UP Cascade Lake Quad 300 W GPU rendering array PC !
 

danwells

macrumors 6502a
Apr 4, 2015
590
443
Does anyone know if a W-3275 (non M) works? That's the upgrade that saves money right now. Apple offering only the large memory versions of the 24 and 28 core chips adds a $3000+ premium to those configurations that doesn't need to be there.
 

Snow Tiger

macrumors 6502a
Dec 18, 2019
579
362
Does anyone know if a W-3275 (non M) works? That's the upgrade that saves money right now. Apple offering only the large memory versions of the 24 and 28 core chips adds a $3000+ premium to those configurations that doesn't need to be there.
likely works , but not verified .
 
  • Like
Reactions: erroneous

Rehtori

macrumors newbie
Jan 9, 2020
12
9
Well well. As it happened I just ordered Xeon 3265 24 core (non M) model to replace 3245 16 core cpu in my Mac Pro 7,1 but obviously after reading comments about the possibility of damaging the motherboard etc. ..tbh I'm not sure If I should continue with the installation or wait until some proper instruction guides pop up...
 
  • Like
Reactions: Graham King

Snow Tiger

macrumors 6502a
Dec 18, 2019
579
362
Well well. As it happened I just ordered Xeon 3265 24 core (non M) model to replace 3245 16 core cpu in my Mac Pro 7,1 but obviously after reading comments about the possibility of damaging the motherboard etc. ..tbh I'm not sure If I should continue with the installation or wait until some proper instruction guides pop up...
The safest processor upgrade method for laypeople is to do a complete system tear down via iFixit instructional , unless you can find a method to only remove both the heatsink standoff plates . I tried on my MP7,1 and there was too much Loctite on the fastener threads . All I got was this scary squeaking noise like I was stripping something , so I aborted the process . I had no desire to be the first person in history to strip a torx fastener .

I hope you have better success . Use a T handle T8 driver for the four HS standoff plate fasteners and a T handle T15 driver for the two CPU heatsink fasteners . Important : the heavy and large heatsink fasteners are not retained and will fall out of the heatsink if you do not keep the heatsink level . Do not attempt to use straight handle drivers for either fastener type , as you need the sufficient torque a T handle will provide .

In order to deal with that issue , I developed a processor removal and installation method suitable for technicians . There is still a high level of risk of damaging the socket and I'm still debating whether to make the method public . But its safer than finger installation .

Since I personally believe in the free flow of information , I told two high profile individuals in the Mac community the method ( with a pictorial based instructional ) and embargoed public release of the method pending developments . I'm not real eager to encourage someone to fry an eight thousand dollar processor and a thousand dollar logic board .

Three of the pins on the LGA3647 socket of my own MP7,1 were slightly bent from all the attempts I have made so far . The bent pins have not done any harm as far as I can tell , but I am more than annoyed since I always examine processor socket pin arrays with a flashlight beamed from several directions before proceeding . The bent pins are not serious enough to straighten . I have 40 years of experience with straightening interface pin arrays all the way from the simple 40 pin 6502 chip ( 1980s ) to the current super complex 3,647 pin LGA3647 chip . But I'm ashamed that there was any damage at all , folks , as I have installed over a thousand processors in my career . I have some pretty strict shop protocols . I really wish Apple had used a PHM in the MP7,1 , then we wouldn't be having issues with finger installs .

I have performed a total of eight successful removals and installations with this individual socket and the manufacturer , TE Connectivity , says the socket is good for 30 cycles ( installations ) before it wears out . So , I can't continue to test forever with this particular Mac . For the time being , I'm done experimenting . The factory 8 Core chip is back in the MP7,1 and my Gold 6212U is back in Blackbird - my high performance UP Cascade Lake four 300W compute GPU rendering PC .

At any rate , I hope someone ( Alex , are you listening ? ) can figure out how to get Skylake and all the other Cascade Lake processors compatible with the MP7,1 . Right now , we're restricted to the W Xeons .
 

Rehtori

macrumors newbie
Jan 9, 2020
12
9
The safest processor upgrade method for laypeople is to do a complete system tear down via iFixit instructional , unless you can find a method to only remove both the heatsink standoff plates . I tried on my MP7,1 and there was too much Loctite on the fastener threads . All I got was this scary squeaking noise like I was stripping something , so I aborted the process . I had no desire to be the first person in history to strip a torx fastener .

I hope you have better success . Use a T handle T8 driver for the four HS standoff plate fasteners and a T handle T15 driver for the two CPU heatsink fasteners . Important : the heavy and large heatsink fasteners are not retained and will fall out of the heatsink if you do not keep the heatsink level . Do not attempt to use straight handle drivers for either fastener type , as you need the sufficient torque a T handle will provide .

In order to deal with that issue , I developed a processor removal and installation method suitable for technicians . There is still a high level of risk of damaging the socket and I'm still debating whether to make the method public . But its safer than finger installation .

Since I personally believe in the free flow of information , I told two high profile individuals in the Mac community the method ( with a pictorial based instructional ) and embargoed public release of the method pending developments . I'm not real eager to encourage someone to fry an eight thousand dollar processor and a thousand dollar logic board .

Three of the pins on the LGA3647 socket of my own MP7,1 were slightly bent from all the attempts I have made so far . The bent pins have not done any harm as far as I can tell , but I am more than annoyed since I always examine processor socket pin arrays with a flashlight beamed from several directions before proceeding . The bent pins are not serious enough to straighten . I have 40 years of experience with straightening interface pin arrays all the way from the simple 40 pin 6502 chip ( 1980s ) to the current super complex 3,647 pin LGA3647 chip . But I'm ashamed that there was any damage at all , folks , as I have installed over a thousand processors in my career . I have some pretty strict shop protocols . I really wish Apple had used a PHM in the MP7,1 , then we wouldn't be having issues with finger installs .

I have performed a total of eight successful removals and installations with this individual socket and the manufacturer , TE Connectivity , says the socket is good for 30 cycles ( installations ) before it wears out . So , I can't continue to test forever with this particular Mac . For the time being , I'm done experimenting . The factory 8 Core chip is back in the MP7,1 and my Gold 6212U is back in Blackbird - my high performance UP Cascade Lake four 300W compute GPU rendering PC .

At any rate , I hope someone ( Alex , are you listening ? ) can figure out how to get Skylake and all the other Cascade Lake processors compatible with the MP7,1 . Right now , we're restricted to the W Xeons .
3265/3275 Works perfectly fine.
Thx for you detailed setup experience! would you care to share the instructions via email ?
3265/3275 Works perfectly fine.
Nice! How did your cpu / heatsink removing and re-installation go ? was there any bent pins ? Do you have any thoughts regarding this procedure to share with us ?
- - Post merged: - -

For sure some kind of informative instructions would be very helpful in the tcoming days...
 
Last edited:

devnull16

macrumors newbie
Nov 19, 2018
9
4
@Snow Tiger I really appreciate your efforts. I’d probably get the Base Mac Pro and eventually upgrade the CPU, mind to share your guide via email? Thanks a lot
 

Softco

macrumors newbie
Jan 12, 2020
17
8
I hope it isn’t seen as intruding that I registered on MacRumors just to comment on this thread, but I would like to represent a group that I think is actively reading like I have.

I really respect the expertise that has been contributed (even by PassingBy). I also respect that many of those who are commenting are professional system creators, etc. Many years ago I was also in that business and still do my own modding and upgrading. I have also created and used Hackintoshes.

Since then I have transitioned into my own successful video content creation business. That business was built on the backs of upgraded MP 5,1s that have served well, but are effectively topped out at High Sierra making the MP 7,1 a welcome sight.

So here’s my question. My workflow has come to anticipate a certain number of BSoDs or kernel panics, or miscellaneous ghosts in the machine. In the ~10 years I’ve lived on Macs I’ve never done a NVRAM reset. Given all that, why wouldn’t I just do the 6212U uprgrade that Snow Tiger has tested? Searching Google I found the 6212U being sold for under $2000 and its benchmarks are incredible.

I realize those of you in the business have reputations to protect and don’t want to create machines that are less than 100% stable. But for those of us that are competent modders, and are comfortable with recovering from a backup once or twice a year, why wouldn’t we take advantage of the huge cost/benefit advantage of a CPU like the 6212U?

Many thanks for your consideration!
 
  • Like
Reactions: John_B Beta

Kris Kelvin

macrumors regular
Dec 28, 2005
220
131
why wouldn’t we take advantage of the huge cost/benefit advantage of a CPU like the 6212U?
Because we don't know exactly why the NVRAM reset broke the Mac Pro with the 6212U and there's no guarantee this won't happen in production. And IF it happens, you have to...
  1. Recognize the problem (which is next to impossible if you're not the one reading this thread)
  2. Swap the CPU for a compatible model, which hopefully is available right away
Even in the best case, that's considerable downtime and uncertainty.

Of course, if you don't mind that, nobody's preventing you from sticking a 6212U into your Mac Pro. :) But in that scenario, a Hackintosh might be an even more attractive way to get this level (or even higher) of performance.

Also: Welcome!
 

Softco

macrumors newbie
Jan 12, 2020
17
8
Because we don't know exactly why the NVRAM reset broke the Mac Pro with the 6212U and there's no guarantee this won't happen in production. And IF it happens, you have to...
  1. Recognize the problem (which is next to impossible if you're not the one reading this thread)
  2. Swap the CPU for a compatible model, which hopefully is available right away
Even in the best case, that's considerable downtime and uncertainty.

Of course, if you don't mind that, nobody's preventing you from sticking a 6212U into your Mac Pro. :) But in that scenario, a Hackintosh might be an even more attractive way to get this level (or even higher) of performance.

Also: Welcome!
Thank you!

I also see both of the points you raised as critical.

I assume 1 is taken care of because I’m really only expecting people who have read this thread to know about the pros and cons of upgrading with this CPU.
2 is really important too, but unless you’re planning to sell the stock CPU, it should always be available as an emergency backup.

Of course, the biggest factor of all is stability. How often will 1 or 2 even be an issue. That was the reason I ultimately swore off hackintoshes. Such a huge headache keeping them stable, and anytime your machine did something strange you didn’t even know where to begin the search.

Gratefully, my business is busy enough that I can afford to buy a better stock CPU from Apple, and so busy that I really can’t afford to have my machine be hackintosh flakey. But I really really hate paying Apple double for half the performance just because Apple wants me to.

I suppose the only way to know the answer is for someone to do the upgrade and report. Still trying to decide if that someone will be me.
 

Snow Tiger

macrumors 6502a
Dec 18, 2019
579
362
There has been a minor but noteworthy discovery in the firmware of the MP7,1 by the talented and mild mannered members here at MR .

There is another CPU check , independent of the CPUID check .

The System tests for a "CPU Type Mismatch" event , while the Mac is powered on .

We think it has something to do with the number of PCIe lanes being reported , how they are organized , or both .

What this boils down to is this - as the W series Xeons are the only subfamily of Cascade Lake Xeons that have 64 PCIe lanes ( while the rest of these Xeons have only 48 lanes ) , we suspect that somehow the firmware only gives the okey dokey to the W Xeons when it performs this check because they report the necessary lanes as available ( and maybe organized in a certain way ) . And also , it must be somehow related to NVRAM resets , of course .

We'll report as further discoveries are made .

But personally , I'm in the process of upgrading my MP7,1 with a W-3275M .

It's the best piece of silicon that I can prove , at this time , works as intended in my Mac .

From another member here :

Screen Shot 2020-01-15 at 12.58.37 PM.png
 
Last edited:

LightBulbFun

macrumors 68020
Nov 17, 2013
2,174
1,993
London UK
Called it! :)

I have a feeling the whole Xeon SP CPU issue

is because when you do a PRAM reset, you reset everything including low level memory training and other such things

that when you do a CPU swap are still in place from when the old CPU was installed

but when you reset the NVRAM it clears all that and perhaps one of those low level bring ups is not compatible with the Xeon SP CPU

(it may even be some sort of platform compatibility check that gets ignored until you do a PRAM reset at which point the system goes Oi this CPU is not for this platform!)

what diagnostic LEDs light up on the logic board when its NOT posting?
but they are different CPU still for technically different platforms

a good example of this is the LGA1151 Xeon E CPUs dont (normally) work in Consumer Core ix Motherboards despite being the same socket and same CPUID etc

(the only thing that blocks them from working is the intel Management engine and if you can get round that then you can boot them on consumer mobos)

also from what i have been told, the MacPro7,1 does have 4 diagnostic LEDs,

1 is standby power, first power regulator, and T2 is powered

2 is all power regulators active, and CPU is powered to startup

3 is memory setup complete

4 is EFI completed startup and passed startup to the operating system wherever it is

it is also worth keeping in mind the Xeon SP CPUs only have 48 lanes where as the Xeon W CPUs have 64 lanes

it could be something as simple as the previous config from the Xeon W is forcing all 64 lanes on the Xeon SP CPU then you PRAM reset it gets set back to 48 lanes but 48 lanes may mean something important is no longer connected so the machine no longer POSTS

and it could even be something to do with your specific Model of CPU (I know its a gold CPU but specifically a single CPU only one, it would be interesting to try a non U CPU)
 

daveedjackson

macrumors regular
Aug 6, 2009
169
143
London
Sounds to me a lot like Apple are doing everything in their power to stop the upgrade of the CPU. have they spent so much time on making sure it’s expandable under their own parameters?
 

fuchsdh

macrumors 68000
Jun 19, 2014
1,525
864
Sounds to me a lot like Apple are doing everything in their power to stop the upgrade of the CPU. have they spent so much time on making sure it’s expandable under their own parameters?
How is not supporting CPUs that don’t work for the system “doing everything in their power”? There’s no evidence they’re blocking -W CPUs they don’t sell.
 
  • Like
Reactions: Macintosh IIcx

tsialex

macrumors 603
Jun 13, 2016
6,462
7,098
Brazil
Sounds to me a lot like Apple are doing everything in their power to stop the upgrade of the CPU. have they spent so much time on making sure it’s expandable under their own parameters?
This has nothing to do with Apple blocking anything since 2019 Mac Pro needs a Xeon processor with 64 lanes.

Apple has a validation test in place for the type of the CPU (probably checking if it has 48 or 64 lanes available) and I found that it returns "Configuration validation failed. CPU type mismatch." when it fails. This is probably why @Snow Tiger Cascade Lake 6212U don't survive a NVRAM reset, since allocation of the PCIe lanes to the PEX8796 switch and the Pool A/B automatic configuration needs to be redone at POST time.

There are further validation tests, we can see several of the strings returned here:
Screen Shot 2020-01-15 at 18.17.54.png
 

beaker7

Cancelled
Mar 16, 2009
920
5,011
Sounds to me a lot like Apple are doing everything in their power to stop the upgrade of the CPU. have they spent so much time on making sure it’s expandable under their own parameters?
Yeah thats a bit unfair. I am stunned that a non-W CPU even booted.
 

Snow Tiger

macrumors 6502a
Dec 18, 2019
579
362
Yeah thats a bit unfair. I am stunned that a non-W CPU even booted.
Not really . The Gold 6212U I tested had a compatible CPUID ( as we found out later , but I deduced this at first ) and Stepping version ( which I knew from day one , which is critically important ) in the firmware ; two things very much in its favor .

Where the non-authorized processor upgrade failed was precisely in the total number of PCIe lanes provided to the System and maybe how they are organized ( since not all PCIe lanes are generated by the CPU ) .

The big question now is , can and how this limitation be patched away ? And what would be the limitations , if possible ? The unauthorized processor might survive all the tests , but we could wind up with a situation where not all the lanes of a PCIe slot are available , maybe we'll lose a slot even , or gosh forbid if the missing lanes are tied into the factory NAND drive ? ( another failed boot situation , since that's where the key is that starts up the System , metaphorically speaking ) .
- - Post merged: - -

Why are you buying the M?? the non-M works just as well. are you planning to add more than 768GB of RAM? :O
Why climb only half way up Mt. Everest ? 🤔 Someday , I'm gonna test 2TB in one of these rigs ...
 
Last edited:
  • Like
Reactions: erroneous