6511P/Granite Rapids in general OCability

Hi,
Recently specced out a server using a Xeon 6507P only to learn it doesn’t have enough PCIe lanes for the board and application I want to use it with. The next viable CPU is the 6511P, with double the cores, and all the PCIe lanes I need - but it is 200MHz slower boosting than the 6507P (4.1GHz vs the 4.3GHz of the 6507P). My workload is single-threaded, which puts me in an annoying dilemma.

a) Am I able to OC the part effectively to either recuperate that 200MHz or surpass it? More do be better…

b) Is the 200MHz even worth caring about?

c) Are there any other options that are high-clocking, not necessarily high in core count (again, not one of the requirements) but also with ~128 PCIe 5.0 lanes, that aren’t earth-shatteringly expensive?

$546 for 8 cores 4.1 ghz low power usage 128 pcie gen5 lanes 12 channels ddr5 ecc rdimm… PROVANTAGE: AMD 100-000001553 AMD Epyc 9015 8 Core 155 TDP
And the zen5 cores out perform my genoa chips despite the 4.1ghz clock speed. 1 worker was able to FIO ~8gigabytes/sec to my NFS storage target. Dont know what your workload is but this thing is impressive for the low price.

1 Like

Will look into
(though would still like other questions answered)

I haven’t used Granite Rapids, but with Skylake/Cascade Lake era chips and Rome/Milan, 200MHz does not make a difference.

In my experience moving from 3.7GHz to 4.0 GHz SKUs, for example, properly setting up and configuring Linux tunables like thread pinning, numad/numactl, hugepages, adjusting mitigations/KASLR/microcode and so on will have a far larger impact for most workloads.

The fact of the matter is you’ve got the same number of registers and same cache hierarchy to work with, so more clockspeed may not provide a linear increase in performance.

In terms of SKUs to upgrade to, STH has a good breakdown.

2 Likes

This topic was automatically closed 273 days after the last reply. New replies are no longer allowed.