This discussion is now locked
Posted by Mr. Paul Clarke on 8 May 2014 at 12:09
Section 2.2.3, "Memory Bandwidth". The L3 cache bandwidth description says "at half-clock speed", yet that doesn't seem to be accounted for in the subsequent formula. Which is correct?
Also, there are nice desciptions and formulas for all of the channels (L1, L2, L3, SMP interconnect, PCIe interconnect), but nothing about the "Total Memory" itself. How is the "192 GBps" or "384 GBps" calculated?
Posted by Mr. Scott Vetter on 9 May 2014 at 8:18
I will ask the team! More later.
Posted by Mr. Scott Vetter on 12 May 2014 at 8:50
Adding this text:
Total Memory Bandwidth: Each POWER8 processor has 8 memory channels running at 8 GBps capable of writing 2 bytes and reading 1 byte at a time. The bandwidth formulas is calculates as follows:
8 channels * 8 GBps * 3 Bytes = 192 GBps per processor module
Posted by Gary Rockley on 29 June 2014 at 14:22
Memory has issues.
1. (All memory must be ordered in pairs
For S812L, If Min is 16GB, and must be 'ordered' in pairs then min is 32GB not 16GB. (There is no 8GB option and the FC is single DIMM). Per Salesmanual min is 1x #EL3P.
2. Plugging order for S822L is questioned.
If 1st Pair is plugged to C16 and C18, then the 2nd Proc is unpolulated. Is this correct? If so, Memory Load Balancing not acheived.
Posted by Mr. Scott Vetter on 9 October 2014 at 13:52
An updated version of this publication is available today with corrections based on your comments, and additional major features if required.
Posted by Herman Dierks on 20 July 2015 at 8:52
Nice job on the Redbook. I like the fact that you show how to compute the various bus bandwidth's (however those are all "peak bandwidths" and may not be "sustained bandwidths".
It looks like in Fig 2-12 that it shows a slot "C4" which really should be "C5". The Fig 2-13 has it correct so I think 2-12 was a typo and should be fixed. Thanks
Posted by Mr. Scott Vetter on 21 July 2015 at 15:36
Thanks. We used to have a caveat in the start of chapter 2. I need to double check to ensure that it didn't drop out somewhere. Yes.. all the speeds are in theory peaks, and real workload testing should be made for proper evaluation. I'll get that slot number updated and a new version online.
Posted by Deepak Shetty on 28 April 2017 at 0:34
Table 2-26 needs to be updated for the KVM (EC70) column. Currently its blank. Also the USB adapters tables/matrix @ Table 2-17 just say Linux under "OS Support". which version of Linux under which virtualization path (PowerVM, KVM on Power) is not mentioned and thus is not useful at all!
Posted by Mr. Scott Vetter on 1 May 2017 at 13:02
Hello Deepak! I see you discovered the online comments. Always best to directly email me as you have in the past. If you send me the information to include for the versions of Linux, I will be glad to update however Linux support always changes and new releases come out. It is better to check the distro directly to ensure the latest information.