Duel Intel Xeon Gold 6152 Lack Of Performance

[MBell9]

Purchased a new DELL precision 7920 work station, Windows 10 Pro, 64bit, 1803, For Workstations. (power setting is "Ultimate")

Duel Intel Xeon Gold 6152, 2.1GHz, 3.7GHz Turbo 22C (88 Hyper threaded logical processors)

Benched marked this against our existing DELL precision 7910 work station, Windows 7 Professional, 64bit. (3 years old computer)

Dual Intel Xeon Processor E5-2697 v3, 2.6GHz, 3.6GHz Turbo 14C (56 Hyper threaded logical processors)

Software used is PTC Creo (4) CAD/CAM.

This uses a single thread for calculating tool cutter paths, and multi threads (all logical processors) when gouge checking.

Xeon Gold single thread performance, did not see this speed go above 2.8GHz (not 3.7GHz as expected), ran 67% slower than E5-2697.

Multi threading, Windows 10 "Affinity" made two processor groups "group 0" & "group 1", with 44 logical processors per group.

Can only activate one group at a time and can only gouge check with 44 logical processors (not the total 88), this ran 357% slower than E5-2697.

Once again did not see Xeon Gold speed go above 2.8GHz

E5-2697 (Win 7) has one group for processor affinity with all 56 logical processors ticked.

I tried changing the environment variable "number of processors" to 88, upon computer restart, this automatically changes back to 44.

Can anyone help?

Has anyone experienced this reduced performance issue?

Is there a windows 10 setting that I have missed that I can set affinity to a single group or increase processor speed?

Is there a hardware setting or bios setting to improve processor speed affinity group?

See attached workstation details

[MBell9]

Interesting, just done a "Intel® Processor Diagnostic", it reports only processor at 2.79Ghz, where is the 3.7GHz? (see attached word doc)

Task manager shows two.

[idata]

Hi mathew.bell,

The Intel® Xeon Gold 6152 belongs to our latest processor family Scalable and some operating systems are not capable to recognize properly all the new features and processor architecture. My best recommendation is to let the operating system chooses automatically the number of Cores to handle your application.In regard to the fastest processor speed will be able to reach; depends on the processor voltage and temperature allow it. Additionally, it will speed up if the operating system requested it, only.My suggestion is the following: Confirm with Dell support if the Intel® Turbo Boost Technology is active in the BIOS https://www.dell.com/support/home/us/en/04/?lwp=rtand run our processor stress test to confirm if the processor is working properly.Intel® Processor Diagnostic Tool https://downloadcenter.intel.com/download/19792/Intel-Processor-Diagnostic-ToolRegards, Miguel C.Intel Customer SupportTechnician Under Contract to Intel Corporation

[idata]

Hi Mathew.bell,

I am following up on your case. Have you been able to run our Intel® Processor Diagnostic Tool? We are really interested in the processor performance of your new processor.Regards,Miguel C.Intel Customer Support TechnicianUnder Contract to Intel Corporation

[MBell9]

Hi Miguel,

Thanks for following up.

See attached full test (TXT) of the CPU's, only registering 2.8GHz. Hopefully you can tell me how to get it to 3.7GHz.

Screen shot of task manager.

Are you also able to answer why Gold 6152 is slower than E5-2697 v3?

Also speed step and turbo boost is active within the bios.

[idata]

Hi mathew.bell,

I reviewed the results of Intel® Processor Diagnostic Tool and the processor is working as it should. It passed the stress test. I am sending you the Intel® Xeon® Processor Scalable Family NDA Specification Update. In figure 4 you will see the expected maximum speed when all the Cores are working. In this case, the maximum speed is 2.8MHz. https://www.intel.com/content/www/us/en/processors/xeon/scalable/xeon-scalable-spec-update.htmlI reviewed the datasheet and specification update and the Intel® Xeon® Gold 6152 Processor has 22 Cores and 44 Threads.Review at Dell support website if you are already using the latest bios of your system. https://www.dell.com/support/home/us/en/04/product-support/product/precision-7920-workstation/driversRegards,Miguel C.Intel Customer Support TechnicianUnder Contract to Intel Corporation

[idata]

Hi mathew.bell,

I am following up on your last post. Did you have any luck with a newer BIOS version?Regards,Miguel C.Intel Customer Support TechnicianUnder Contract to Intel Corporation

[MBell9]

Hello Miguel,

Yes, BIOS is fully up to date. I have also sent this email to the below people.

mailto:Barry.Chew@dell.com Barry.Chew@dell.com, mailto:Brianna.Obrien@dell.com Brianna.Obrien@dell.com, mailto:Carmen.Sea@dell.com Carmen.Sea@dell.com, mailto:Ee.Noch.Koo@dell.com Ee.Noch.Koo@dell.com, mailto:Guo.Xin.Ye@dell.com Guo.Xin.Ye@dell.com, mailto:Jet.Maisog@dell.com Jet.Maisog@dell.com, mailto:Kelly.Phillips@gallagher.com Kelly.Phillips@gallagher.com, mailto:Kevin.R.Davies@dell.com Kevin.R.Davies@dell.com, mailto:Marcus.Wixted@dell.com Marcus.Wixted@dell.com, mailto:Todd.Mills@Dell.com Todd.Mills@Dell.com, intel_corp <</span>mailto:webadmin@intel.com webadmin@intel.com>, Jojo Adolfo/GGL@GGL, "Tang, Xiao Yun Ariel" mailto:ng@ptc.com ng@ptc.com>, LEAP PTC Support <</span>mailto:ptcsupport@leapaust.com.au ptcsupport@leapaust.com.au>, Paul Reid/GGL@GGL,

This is a blanket email that covers all interested parties, DELL, INTEL, PTC CREO (4), & GGL IS Department.

DELL Precession 7920, now installed with Win7 Prof. (was Win10 for workstations) has moved forwards positively in speed, but not to expected levels.

Best BIOS setting to balance thread performance and multi threading is having 56 Logical cores. (same as our old computer DELL Precision 7910)

Single thread is now comparable to our old computer about 2.40min (was about 3.10min - 4min approx depending on No. threads), on paper it should faster.

Multi threading bench marks have improved from 6.37min down to 2.37min. (still disappointed here, old computer does it in 1.56min.)

Changed BIOS setting to 64 logical processors, the multi threading time is comparable to 56 logical processors. (maybe better by a few seconds, not chasing this...)

64 logical processors gives one processor group. Affinity, all processors can be ticked.

88 logical processors gives two logical groups of 44. Affinity, only one group can be ticked.

With all 88 logical processors switched on, with affinity only 44 used, multi threading slower again, 3.04mins, however still faster than Win10.

We did a final bench mark test on another older computer with details:

Win 7 Pro. Xeon E5-1620 V3, 3.5GHz, 4 cores, 8 logical processors.

Single thread = 2.48min

Multi thread = 3.23mins

In conclusion, using PTC Creo (4) M060 on a DELL Precision 7920:

Overall the Xeon Gold 6152 is slower than Xeon E5-2697 v3.

Compromised BIOS setting of 56 logical processors switched on, out of 88.

NZ$16K on new workstation that has not given any performance increase, only less than what it should be.

Gallagher Group IS Depart and JoJo Adolfo have spent considerable time and effort to convert DELL Precession 7920 (Win10) to Win7, and also carries a cost of 1.5 - 2 weeks labour, thankyou IS.

It may not be a processor problem, what now can be done? INTEL, DELL & PTC, Please, all three talk to each other to resolve and let me know options.

This computer runs better on Win 7, and will remain.

Click on the below links to download recordings of the multi threading testing.

Precision 7910_test2 (3 year old computer) Download size 1.2GB

Precision 7920_test2 (new computer) Download size 400MB

[idata]

Hi mathew.bell,

I have received an update from my upper level. It seems the processor might be running at lower frequencies than expected due to one of the following reasons: 1. Depends upon the mode on which it is being operated on AVX-512 or AVX which manages a higher workload. The CPU may not ramp up core frequencies when running some Intel AVX or AVX-512 workloads depending upon the number of active cores.2. Workload balancing mechanism in Xeon Scalable Processors may have the potential to incorrectly apply a larger than expected weighting value to some Intel AVX and AVX-512 workload. Highly parallel/vectorized applications generate the highest load on the processor cores which requires more power and thus generates more heat.3. Since its a third party system and if they have further concerns, we are recommending you, going back to your system manufacturer and request a workaround to the issue.Regards,Miguel C.Intel Customer Support TechnicianUnder Contract to Intel Corporation