Lost cores???

[Mojoxer]

I have a very strange situation. I just upgraded my cMP 1,1 from dual Xeon 5150s to dual Xeon 5355s. The upgrade went smoothly enough, but somehow, I have lost 2 CPU cores out of the possible 8 I should now have.

I updated the SMC and firmware to 2,1 before the upgrade. And now, when I check System Profile, I get the correct listing for the CPU (Quad Core Intel Xeon), but the Total Number of Cores is listed as 6 and the L2 Cache (per Processor) is listed as 6MB, which is 3/4 of the true amount.

Has anyone ever seen this before? I didn't seen any reference to this in the forums and can't find anything about it with my poor Google-fu.

Screencap from System Profile and listing from MachDep.cpu below. I appreciate any help or ideas anybody can toss my way. Surely these aren't CPUs with disabled cores???

sysctl -a |grep machdep.cpu
machdep.cpu.max_basic: 10
machdep.cpu.max_ext: 2147483656
machdep.cpu.vendor: GenuineIntel
machdep.cpu.brand_string: Intel(R) Xeon(R) CPU X5355 @ 2.66GHz
machdep.cpu.family: 6
machdep.cpu.model: 15
machdep.cpu.extmodel: 0
machdep.cpu.extfamily: 0
machdep.cpu.stepping: 7
machdep.cpu.feature_bits: 3219913727 320445
machdep.cpu.extfeature_bits: 537921536 1
machdep.cpu.signature: 1783
machdep.cpu.brand: 0
machdep.cpu.features: FPU VME DE PSE TSC MSR PAE MCE CX8 APIC SEP MTRR PGE MCA CMOV PAT PSE36 CLFSH DS ACPI MMX FXSR SSE SSE2 SS HTT TM PBE SSE3 DTES64 MON DSCPL VMX EST TM2 SSSE3 CX16 TPR PDCM
machdep.cpu.extfeatures: SYSCALL XD EM64T LAHF
machdep.cpu.logical_per_package: 4
machdep.cpu.cores_per_package: 4
machdep.cpu.microcode_version: 102
machdep.cpu.processor_flag: 2
machdep.cpu.mwait.linesize_min: 64
machdep.cpu.mwait.linesize_max: 64
machdep.cpu.mwait.extensions: 3
machdep.cpu.mwait.sub_Cstates: 32
machdep.cpu.thermal.sensor: 1
machdep.cpu.thermal.dynamic_acceleration: 0
machdep.cpu.thermal.invariant_APIC_timer: 0
machdep.cpu.thermal.thresholds: 2
machdep.cpu.thermal.ACNT_MCNT: 1
machdep.cpu.thermal.core_power_limits: 0
machdep.cpu.thermal.fine_grain_clock_mod: 0
machdep.cpu.thermal.package_thermal_intr: 0
machdep.cpu.thermal.hardware_feedback: 0
machdep.cpu.thermal.energy_policy: 0
machdep.cpu.arch_perf.version: 2
machdep.cpu.arch_perf.number: 2
machdep.cpu.arch_perf.width: 40
machdep.cpu.arch_perf.events_number: 7
machdep.cpu.arch_perf.events: 0
machdep.cpu.arch_perf.fixed_number: 0
machdep.cpu.arch_perf.fixed_width: 0
machdep.cpu.cache.linesize: 64
machdep.cpu.cache.L2_associativity: 16
machdep.cpu.cache.size: 4096
machdep.cpu.tlb.inst.small: 128
machdep.cpu.tlb.inst.large: 8
machdep.cpu.tlb.data.small: 16
machdep.cpu.tlb.data.small_level1: 256
machdep.cpu.tlb.data.large: 16
machdep.cpu.tlb.data.large_level1: 32
machdep.cpu.address_bits.physical: 36
machdep.cpu.address_bits.virtual: 48
machdep.cpu.core_count: 4
machdep.cpu.thread_count: 4

 

[Amethyst1]

• The sysctl output looks ok AFAICS.
• Does Activity Monitor’s CPU usage graph show six or eight cores?
• It’s possible to limit Mac OS X to a certain number of cores with the cpus=n boot argument. Check if this has been set (sudo nvram boot-args).

 

[Mojoxer]

Thanks for the reply. Activity Monitor shows 6 cores, both in the floating CPU window and in the CPU window of the main program. the nvram boot-args is clear, and I also tried to force the issue the other way with cpus=8, with no luck. Currently installing Bootcamp and Win7 to see if it's a hardware thing or an OSX thing.

 

[Amethyst1]

Currently installing Bootcamp and Win7 to see if it's a hardware thing or an OSX thing.

That would have been my next suggestion. Keep us posted.

 

[Mojoxer]

Just FYI, cMP 1,1 needs 32-bit Windows 7 for ease of installation. 32-bit EFI don't ya know! if I had remembered this, it would have saved me a bit of frustration.

 

[Mojoxer]

That would have been my next suggestion. Keep us posted.

OK. BootCamp is less than useful. Win7 32-bit apparently will only see one socket in the MacPro? It correctly IDs the Clovertown X5355 with 4 Cores, and Win TaskManager sees all four cores, as does HWiNFO32.

Does anyone know if there is a live-CD/USB version of Linux available for the MacPro 1,1 I could use for testing? This is really weird.

 

[Mojoxer]

I've found a spot for 32-bit EFI/64-bit distros. Slow downloads, but I'm working on it. Fingers crossed.

 

[Mojoxer]

Finally, after getting a Ubuntu 16.04 Live Disk up and running, it seems I have some rare 3-core Xeon X5355 CPUs. I have no other explanation for the results. Running cat /proc/cpuinfo gives me info on 6 cpu cores (0-5) reported as GenuineIntel Xeon X5355 family 6 model 15 stepping 7 processors. Running System Monitor shows me 6 cpus running as well. Since OSX and Ubuntu are both showing the same number of cores, I'm going to say it must be a hardware issue, or something buried down deep in opencore(?) that I know nothing about.

With that. apparently, I'm now the owner of the only (?) 6-core MacPro 1,1 in existence! I don't know if should feel honored or not. The pair of CPUs only cost me $20 on eBay, so for that plus a little time and thermal paste, I have 50% more CPU cores with the resultant 50% uplift in multiprocessor performance. Cool beans, and thanks for the help Amethyst1!

 

[Amethyst1]

Win7 32-bit […] correctly IDs the Clovertown X5355 with 4 Cores, and Win TaskManager sees all four cores, as does HWiNFO32.

That’s interesting. Maybe one of your CPUs is a quad-core and the other is a dual-core…? The 6 MB L2 per CPU wouldn’t quite make sense in that case though…

Finally, after getting a Ubuntu 16.04 Live Disk up and running, it seems I have some rare 3-core Xeon X5355 CPUs. I have no other explanation for the results. Running cat /proc/cpuinfo gives me info on 6 cpu cores (0-5) reported as GenuineIntel Xeon X5355 family 6 model 15 stepping 7 processors. Running System Monitor shows me 6 cpus running as well.

Is there a way to check which cores are on which CPU/socket?

 

[Mojoxer]

That’s interesting. Maybe one of your CPUs is a quad-core and the other is a dual-core…? The 6 MB L2 per CPU wouldn’t quite make sense in that case though…


Is there a way to check which cores are on which CPU/socket?

Maybe so. Will have to work up to taking the whole thing apart again to look at the processors themselves again. I thought I checked their markings before installing them, but can't find where I wrote down what was marked on them, and I can't remember what they were marked. (Getting oid suxxors!)

Not sure if the MP 1,1s will run with only one CPU installed. I think I've read that folks have problems with that, but I think it should work. (I know the 3,1 runs with only 1 CPU installed, I upgraded my 2.8 with a 2nd CPU and heatsink.) I will try with CPU A only and RAM in slots 1&2 only on tray A to see if that gives me any new info. If that works, I can swap out the CPUs in the same socket and see if they both report the same, or something different.

 

[Mojoxer]

I don't think I've ever said, but just in case it makes any difference, I'm running OSX 10.7.5 on this machine.

 

[EugW]

That’s interesting. Maybe one of your CPUs is a quad-core and the other is a dual-core…? The 6 MB L2 per CPU wouldn’t quite make sense in that case though…


Is there a way to check which cores are on which CPU/socket?

Hmm... X5355 is quad-core 2.66. 5150 is dual-core 2.66.

What are the leftover CPUs after the install?

 

[Amethyst1]

Hmm... X5355 is quad-core 2.66. 5150 is dual-core 2.66.

I’d think if an X5355 had been paired with a 5150 cat /proc/cpuinfo on Linux would have shown different brand strings for the two CPUs though. 🤷‍♂️

 

[Icelus]

Linux's $ dmidecode might give some additional info.

 

[Nguyen Duc Hieu]

Just FYI, cMP 1,1 needs 32-bit Windows 7 for ease of installation. 32-bit EFI don't ya know! if I had remembered this, it would have saved me a bit of frustration.

I got Windows 10-64 bit on my cMP1,1.
It's a little bit complicated to install, but it's doable.
If you install Windows 32 bit, only 3GB of RAM is usable.

 

[Amethyst1]

I got Windows 10-64 bit on my cMP1,1.
It's a little bit complicated to install, but it's doable.

Don’t click me

 

[TheShortTimer]

Don’t click me

I clicked.

 

[Mojoxer]

Latest update. I got the beast apart and the CPUs are both XEON x5355s. Unfortunately they are not the "best" stepping. They are the SLAC4 stepping, instead of the SLAEG stepping. I don't know if this makes a difference to the MP, and I can't find any sort of errata about there being any 3-core or disabled core variants on the google.

 

[Mojoxer]

Welp, I was wrong. Couldn't get the MP to boot with 1 CPU. Guess I'll just live with a 6-core system until I can spring for another pair of CPUs. Although with a system this old, I'm not sure it really matters. As I said, I'm only out $20.

 

[Amethyst1]

If they're both three-core CPUs (which I thought only AMD had done a while back ) why does 32bit Windows "see" four cores despite being restricted to one socket...? Anyway, that's a cool find and a cool Mac Pro you have there.

 

[GMShadow]

It's possible these were failed chips that someone re-lidded, I suppose. Something's off about this for sure.

 

[Mojoxer]

Just an update on this situation.

I just purchased a new pair of X5355 CPUs from a new vendor. These chips are S-SPEC SLAEG chips. After installation into the cMP, there is no change to the thread count on the machine. It's still listing 2 X5355 CPUs with a total of 6 Threads. This shows in System Information, Activity Monitor, and even is reported as such in Geekbench 2.2.7.

I'm now pretty certain this must be some sort of open firmware setting or OSX setting. I could accept that I got one "bad" set of CPUs, but I'm not sure of the coincidental probability of two sets coming to the same person out of all the people who have upgraded a cMP 1,1 to a 2,1 and then installed X5355s.

Does anyone out there have any new ideas?

 

[m1maverick]

Did you reset the open firmware defaults. From the open firmware prompt:

set-defaults​

What does the following open firmware output report:

dev /cpus .properties​

 

[Mojoxer]

Linux's $ dmidecode might give some additional info.

I finally got back to Linux to check this information out an was surprised at what I found:

CPU A - Processors 0-3 available. CPU B - Processors 0 & 1 only available.

Surprisingly (to me!) I'm getting all of one CPU and only half of the second CPU working. Fortunately (???) I have one more X5355 CPU (SLAEG) that I can swap out with CPU B to see if that fixes the situation.

Also, I forgot, Intel Macs don't do Open Firmware. They are EFI. Sorry for my confusion.

 

[Amethyst1]

Surprisingly (to me!) I'm getting all of one CPU and only half of the second CPU working.

At least this explains why 32bit Windows sees four cores.