I have a x86 router with 4eth and 4BGP peers Full view uplinks,
now it work on vmware esxi and have uptime > 60days without problem.
On last week i instal clean system 6.32.2 on x86 (supermicro X9SCM-F, cel-G540, 2Gb ECC, sata DOM 128gb, 4Eth intel(2onboard+2pci-e)) and export config through export compact and corrected many variable.
On new system Mikrotik work not stable, when router get all BGPpeers routes, router have free 1Gb of ram (2gb total), and work normal, on random momet free memory begin to runs out and all CPU cores have load 100% (/tools profile - managment and routing have heavy load)
I downgrade system to 6.30.4 but, it don't have result. memory leak... uptime 3-7h - get EOF...

observations:
Then router geting routes from BGP-peers - cpu load 100% and winbox not connected.
frequently one cpu core have 100% load (on x86 and CCR)
any other router on x86 and routerboard without BGP work fine on v6.30.4 and 6.32.2.

6.32.3 same....
uptime 10h
memory <100mb (in normal state router have 1G free ram)
cpu load all cores - 100%

Isn't it Eth drivers problem ? Are the previous Eths in the same configuration ? Onboard cards could saturate processor.

all eth cards is intel server nic,
i don't think what problem in NIC...
becouse in leak moment routing+management load cpu 100%

weird. no XEN, VMWare/EMC stuff installed on system ? or anything elese, interfering with netflow ?

so far its more usual for obsolete(1-2yrs ago)and fixed Atheros drivers for x86 arch(both UX and Win affected in past days), than intel interfaces.
do you tried to play with NIC "advanced" settings for awhile ?

Do you use SNMP and/or The Dude?

I've seen on occasions that The Dude will cause 100% cpu usage (and profile shows 'management' at 100%).

It may be a long shot but you could try disabling SNMP to see if that changes anything.

Also do you get a full internet routing table from your BGP peers?
If the issue is actuallt with The Dude/SNMP, it may be polling for the whole routing table of the router causing also 'routing' to take 100% of the CPU in profile.

For me the only fix I found for this was to completely disable any polling (snmp or routeros) on the Dude for the Mikrotiks affected.