Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

Snow Tiger

macrumors 6502a
Original poster
Dec 18, 2019
854
634
Just as a formality , I am in the process of independently proving the discount 28 Core W-3275M Xeon I purchased used from China is a bonafide production processor . This chip was used to upgrade my factory 8 core Mac Pro 7,1 .

The processor is high performance and is completely stable in my MP7,1 . I have not discovered any missing features .

The IHS markings indicate that the chip is a genuine production version , but what would happen if a previous owner placed a production IHS on a ES / QS chip ?

It is not impossible to delid a CPU , after all , and there is a financial motivation by a seller engaging theoretically in fraudulent behavior .

Buyers of high end boxed processors directly from Amazon ( as a seller and not just on its marketplace ) itself , a legitimate seller , have discovered their purchases as counterfeit . Amazon unintentionally bought fake silicon from one of their vendors .

So , best to independently check and verify .

Here's the actual picture of my W-3275M :

Front :
P1235904.JPG


Reverse :
P1235912.JPG


One way is to determine if a chip is really a production version is to generate its full serial number and contact Intel , either directly or through its warranty coverage webpage . Here's how to determine the full serial number of a modern Intel processor . Don't worry , its not an eye readable number :rolleyes: . You didn't think Intel would make it that easy , eh ?

Here's how to decipher the information on the LGA3647 processor's IHS and substrate :

lga3647.png

Screen Shot 2020-01-29 at 12.40.10 PM.png


The serial number of a processor is typically a combination of the finished process order (FPO) number and the assembly test process order (ATPO) number.

The FPO is easy to determine . It's printed on the IHS in eye readable format , on the front side of the chip . The FPO is also printed in eye readable format in gold on the reverse side of the chip . Make certain both these matching numbers are present on the chip . If they don't match , then the IHS likely has been swapped . The ATPO is not eye readable , but is in a 2D matrix ( AKA Data Matrix ) format that requires a discrete physical scanner or an app on a smart phone to read . I use the Barcode Scanner SDK app from Cognex on my iPhone to read this matrix .

If the IHS doesn't match the processor die , then I would think the full serial number would fail to properly generate and Intel would flag it as so .

Another , more thorough , method to determine if a chip is really a production version is to determine the chip's CPUID , S-spec and stepping information . This can be done through software utilities , usually in Windows . The S-spec number is printed on the front of the chip , on the IHS .

I have used a total of three utilities in order to attempt to obtain this info .

1 ) The Intel Processor Identification Utility ( which is not optimized for this processor , so some readings are not provided ) .
2 ) CPU-Z v. 1.91.1 - the latest version .
3 ) AIDA64 Extreme v. 6.20.5300 - the latest version .

CPUID : We know from LightBulbFun using an program called MacEFI that the MP7,1 has microcode support for only three CPUID processor numbers ( 50655 , 50656 and 50657 ) . 50655 and 50656 are ES steppings . And 50657 being the production and QS stepping .

S-spec : S-specs are only printed on the IHS . This is the most accurate verification that a chip is a production version . There appears to be no electronic method of obtaining this info . So , if someone replaced the IHS from the rest of the package , the S-spec number is suspect .

Stepping : Steppings are a verification that a chip is compatible with your System . The same stepping can be used for both an ES / QS version and a production version of a chip , so use this number with care . There are public data bases of steppings from sources like CPU World and wikichip . For instance , wikichip says for my W-3275M that its production S-Spec is SRFFK and its Core Stepping is B1 . Unfortunately , the CPU utilities have to be updated with the latest info in order to read the stepping info of actual silicon correctly . This takes time and the Cascade Lake Xeons are state of the art silicon .

Anyways , here are the screen grabs from the three utilities I ran in Windows 10 Pro Workstation with my W-3275M upgraded MP7,1 .

Capture.PNG


Well , Intel says the chip has a CPUID of 50657 , which is a known production and QS CPUID . It is not an ES version chip .

-----------------------------------------------------------------------------------------------------------------------------------------

Capture2.PNG


CPU-Z says the chip has either a stepping of B1 or L1 . B1 is a W-3275M stepping . But what is L1 ?

---------------------------------------------------------------------------------------------------------------------------------

Capture3.PNG


AIDA64 performed the worst of all - it reported the chip as having either B1 , L1 or R1 stepping . We don't know what L1 or R1 are .

When I ran this chip's serial number ( FPO and partial ATPO ) on Intel's online warranty coverage webpage , I get a notification it is a Tray processor with warranty supplied by the seller :

Screen Shot 2020-01-29 at 1.43.39 PM.png


As a result of all my investigations , I have determined that the chip is definitely not an ES version . It is either a production or QS version chip . It is almost certainly a production chip since I am able to obtain a warranty statement from Intel . Since ES and QS chips cannot be legally sold they would not qualify for any kind of warranty .

But I am annoyed I could not precisely determine the stepping info of my particular processor .
 
Last edited:
Snow Tiger,
Have you tried MacCPUID?
I am not sure this will run on Catalina (32 vs 64 bit app?).
As an example, my CPU info is below on my iMac 5K 27-inch 2017 desktop computer.
I will test this on my Mac Pro 2019 8 core system soon.
Yes I know that the included example is not a Mac Pro. I just included it for demonstration purposes.
Hopefully someone else with a known retail W-3275m processor can provide cpu data for comparison purposes.
Is it possible that your CPU is a QS chip with "better" performance given your GB benchmarks beat the published retail W-3275m? Silicon lottery?
Thanks,
rxs0
 

Attachments

  • Screen Shot 2020-01-29 at 7.17.35 PM.png
    Screen Shot 2020-01-29 at 7.17.35 PM.png
    86.2 KB · Views: 362
Last edited:
Snow Tiger,
Have you tried MacCPUID?
I am not sure this will run on Catalina (32 vs 64 bit app?).
As an example, my CPU info is below on my iMac 5K 27-inch 2017 desktop computer.
I will test this on my Mac Pro 2019 8 core system soon.
Yes I know that the included example is not a Mac Pro. I just included it for demonstration purposes.
Hopefully someone else with a known retail W-3275m processor can provide cpu data for comparison purposes.
Is it possible that your CPU is a QS chip with "better" performance given your GB benchmarks beat the published retail W-3275m? Silicon lottery?
Thanks,
rxs0

The stepping information is in the wrong format , with the MacCPUID utility . So , we won't learn anything meaningful . What we need is called the manufacturing design stepping number , ( e.g. B1 ) .

From Intel's support document on their utility :


Screen Shot 2020-01-29 at 6.48.54 PM.png
 
B1 and L1 are the steppings, and CPU-Z is giving the proper information. You have 28 cores, so it has to be an XCC, which makes it a B-1 stepping.

See page 10: https://www.intel.com/content/dam/w...updates/2nd-gen-xeon-scalable-spec-update.pdf

Note the stepping ID, the least significant hex digit in the CPUID is always 7, so that's why the apps can't tell the difference. I don't know if AIDA will tell you the chop, but HWiNFO will. But as I said, because you have more than 18 cores, you're guaranteed to be XCC.
 
Last edited:
Tray processors are processors sold at volume to an OEM or System Integrator, without any box, fan, manual etc.

It doesn't necessarily mean they are of the lower grade batch?

On the flip-side, retail boxed versions are of higher grade? (unless they are of lower stepping)
 
It doesn't necessarily mean they are of the lower grade batch?

On the flip-side, retail boxed versions are of higher grade? (unless they are of lower stepping)
Nope, just processors sold at volume. There aren't differences between tray or boxed processors of the same sSpec besides box, manual, heatsink, marketing materials.

OEMs and System Integrators don't buy boxed processors, just tray ones. Tray and boxed are exactly the same silicon.
 
  • Like
Reactions: kittiyut
I believe it was $4,500 per another thread: here post #25
Got it! I'm thinking upgrade mine from 8-Core to 28-Core as well, and didn't want to pay $8K for the upgrade from Apple.

By the way, I found some Chinese websites that explain about the Intel processor:

 
  • Haha
Reactions: Scansterz
OK , between konqerror's Intel Specification Update article

and another article on Cascade Lake ( CL ) Xeons at Anandtech ,


I have deduced what the B1 , L1 and R1 stepping versions refer to . CL Xeons have a specific stepping version depending on how large a die Intel used to make them .

It works like this :

An LCC die is used to make CL Xeons up to 10 cores .

An HCC die is used to make CL Xeons up to 18 cores .

An XCC die is used to make CL Xeons up to 28 cores .

LCC die Xeons have a stepping of R-1 .

HCC die Xeons have a stepping of L-1 .

XCC die Xeons have a stepping of B-1 .

Screen Shot 2020-01-30 at 12.55.44 AM.png


A CL Xeon with more than 18 cores can only have a B-1 stepping since it must be manufactured with a XCC die .
 
Last edited:
It works like this :

An LCC die is used to make CL Xeons up to 10 cores .
An HCC die is used to make CL Xeons up to 18 cores .
An XCC die is used to make CL Xeons up to 28 cores .

Right. Obviously any CPU with greater than 18 cores must be XCC, but also note that Intel does use larger chops on smaller configurations, probably due to supply availability or binning. For example, I have some 12-cores that have the cache of 18 cores, but identify themselves as XCC. This lets Intel pick 12 cores that meet the spec out of 28, instead of picking 12 from 18.

Wikichip's got diagrams too: https://en.wikichip.org/wiki/intel/microarchitectures/cascade_lake#Entire_SoC_Overview
 
I installed HWiNFO64 in Windows 10 on my W-3275M upgraded MP7,1 and got some interesting information .

Here are two screen grabs . Be prepared to be puzzled .

one report.PNG


two report.PNG


It reports as a Platinum 8279W with a S-spec of SRGWA , but with a brand name of W-3275M . It also insists the chip is a production unit .

The big issue is , no such processor was ever made by Intel .

Platinum 8279W is not listed anywhere as a retail released processor from Intel . So , maybe it was an internal experimental chip , an OEM chip not publicly available or there is an error in the CPU utility .

OEM chips are silicon manufacturers specially make only for a single customer .

Thoughts , anyone ?
 
Last edited:
I installed HWiNFO64 in Windows 10 on my W-3275M upgraded MP7,1 and got some interesting information .

Here are two screen grabs . Be prepared to be puzzled .

View attachment 891375

View attachment 891376

It reports as a Platinum 8279W with a S-spec of SRGWA , but with a brand name of W-3275M . It also insists the chip is a production unit .

The big issue is , no such processor was ever made by Intel .

Platinum 8279W is not listed anywhere as a retail released processor from Intel . So , maybe it was an internal experimental chip , an OEM chip not publicly available or there is an error in the CPU utility .

OEM chips are silicon manufacturers specially make only for a single customer .

Thoughts , anyone ?
This report seems bogus, you are using a RX 480? Polaris 10 is RX 480…
 
This report seems bogus, you are using a RX 480? Polaris 10 is RX 480…

factory stock 580X in a MP7,1 running in Windows 10 Pro Workstation . I'm not interested in the GPU report , though . Just the CPU .
 
If the GPU is misrepresented, how you can take anything else as correct? ;)

I fell down the rabbit hole a long time ago with the MP7,1 , Alex . Help me claw out . 🐇
[automerge]1580406751[/automerge]
I also can't believe the product "Platinum 8279W" is a randomly generated identity .

There has to be a darn good reason why a program is talking about it .
 
I fell down the rabbit hole a long time ago with the MP7,1 , Alex . Help me claw out . 🐇
[automerge]1580406751[/automerge]
I also can't believe the product "Platinum 8279W" is a randomly generated identity .

There has to be a darn good reason why a program is talking about it .
Don't take any CPU information software too seriously, the only thing really reliable for uncommon CPUs is using sysctl machdep.cpu, using -x when applicable, and searching the info yourself with Intel documentation.
 
An LCC die is used to make CL Xeons up to 10 cores .

An HCC die is used to make CL Xeons up to 18 cores .

An XCC die is used to make CL Xeons up to 28 cores .

I'd like some confirmation other than wikichip (or similar) that the HCC die has 18 cores. Skylake used the 10/18/28 configurations - and I think there was some Intel confirmation of this. (Maybe die photos?)

Cascade Lake SKUs offer a different set of core counts. In particular, there is no 14 core SKU! Where Skylake offered 14 and 18 core SKUs, Cascade Lake instead offers 12 and 16 cores.

My theory is that the HCC die now has only 16 cores - in a 6 x 3 arrangement with two blocks replaced with memory controllers. (Skylake was 4 x 5 - 2.) Going 6 blocks wide would allow the same "north cap" as the XCC die, i.e. the same PCIe and multi-socket interconnect counts.

And I have to wonder about the LCC die. Could it be 6 x 2 - 2? (Skylake was 4 x 3 - 2.) Could it be a harvested HCC die? Note that there is no 10 core W series Cascade Lake. (There are "metal" series 10 cores.)
 
My theory is that the HCC die now has only 16 cores - in a 6 x 3 arrangement with two blocks replaced with memory controllers. (Skylake was 4 x 5 - 2.) Going 6 blocks wide would allow the same "north cap" as the XCC die, i.e. the same PCIe and multi-socket interconnect counts.

My hypothesis is all Gold 6xxx and above, and W series are XCC. Gold 6+ because the XCC is the only one with the 3rd UPI link, and W because only the Gold 6+ have the second AVX-512 MAC. Otherwise it would mean fabricating and disabling the second AVX-512 unit on the two lower chops, for the benefit of a relatively low-volume product.

Does somebody have a W series that isn't a XCC?
 
Thx, very useful. I just got one from China, same FPO as yours. I also checked intel warranty same result.
 
Thx, very useful. I just got one from China, same FPO as yours. I also checked intel warranty same result.

There are some easy tests consumers can run to verify the processor is basically good :

Your Mac Pro has a self diagnostic routine at start up called Apple Diagnostics .

Just press the D key as your Mac starts up .


Here is the list of results codes :


I'll also copy and paste them here as a PDF file in case the link from Apple goes dark .

You can also run a CPU benchmark like Geekbench . There are a lot of integer and floating point type operations in this program , which have been used in the computer industry for decades to benchmark and certify processors as healthy . These would be the old Dhrystone and Whetstone tests from the 1970s and 1980s .

So , if your processor successfully completes a GB 5 test with the appropriate score result , the silicon should be good .


You can also load a copy of Windows 10 Pro Workstation on your Mac Pro 7,1 via bootcamp . It is very easy . Once this is done , you can run GIMPS ( a prime number discovery program which is sure to give your processor a real workout ) .


Their forums are not very friendly to benchmark enthusiasts , though , so unless you are a mathematician ...

You might also install a windows utility to verify processor thread and core activity during this test , like Aida64 .
 

Attachments

  • Apple Diagnostics: Reference codes - Apple Support.pdf
    111 KB · Views: 1,247
There are some easy tests consumers can run to verify the processor is basically good :

Your Mac Pro has a self diagnostic routine at start up called Apple Diagnostics .

Just press the D key as your Mac starts up .


Here is the list of results codes :


I'll also copy and paste them here as a PDF file in case the link from Apple goes dark .

You can also run a CPU benchmark like Geekbench . There are a lot of integer and floating point type operations in this program , which have been used in the computer industry for decades to benchmark and certify processors as healthy . These would be the old Dhrystone and Whetstone tests from the 1970s and 1980s .

So , if your processor successfully completes a GB 5 test with the appropriate score result , the silicon should be good .


You can also load a copy of Windows 10 Pro Workstation on your Mac Pro 7,1 via bootcamp . It is very easy . Once this is done , you can run GIMPS ( a prime number discovery program which is sure to give your processor a real workout ) .


Their forums are not very friendly to benchmark enthusiasts , though , so unless you are a mathematician ...

You might also install a windows utility to verify processor thread and core activity during this test , like Aida64 .
Thx, I saw GIMPS they provide both Windows and Mac version, did you run both or just windows testing?
 
Thx, I saw GIMPS they provide both Windows and Mac version, did you run both or just windows testing?

GIMPS in macOS does not stress the main system memory very much , so I'd hesitate to use it .

GIMPS in Windows 10 via bootcamp in a MP7,1 does fully stress the memory .

There is a close relationship between the processor and main system memory , so its best to stress both concurrently to verify high performance , stability .

Install matching memory modules in your MP7,1 in groups of 6 or 12 for high performance , or your benchmarks will look funny .
 
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.