Re: Server performance & enhancement

2017-02-16 Thread Josh Paetzel


On Thu, Feb 16, 2017, at 05:20 AM, Karl Denninger wrote:
> I'm wondering what people think of a forward path for the following:
> 
> 
> CPU: Intel(R) Xeon(R) CPU   E5620  @ 2.40GHz (2400.14-MHz
> K8-class CPU)
>   Origin="GenuineIntel"  Id=0x206c2  Family=0x6  Model=0x2c  Stepping=2
>  
> Features=0xbfebfbff
>  
> Features2=0x29ee3ff
>   AMD Features=0x2c100800
>   AMD Features2=0x1
>   VT-x: PAT,HLT,MTF,PAUSE,EPT,UG,VPID
>   TSC: P-state invariant, performance statistics
> real memory  = 25773998080 (24580 MB)
> avail memory = 24915828736 (23761 MB)
> Event timer "LAPIC" quality 600
> ACPI APIC Table: <111612 APIC1749>
> FreeBSD/SMP: Multiprocessor System Detected: 16 CPUs
> FreeBSD/SMP: 2 package(s) x 4 core(s) x 2 hardware threads
> 
> The system board is a Supermicro and has embedded IPMI (which I require
> for remote admin purposes) on a separate interface.  It's plenty fast
> but very power-hungry.
> 
> I'm interested in other's experience with various system board options,
> and processors, that will return /better /performance and which have a
> good history running FreeBSD /but with lower power consumption.
> 
> /Any suggestions welcome.
> 
> -- 
> Karl Denninger
> k...@denninger.net 
> /The Market Ticker/
> /[S/MIME encrypted email preferred]/
> Email had 1 attachment:
> + smime.p7s
>   4k (application/pkcs7-signature)


You could run a single E5-2640v4 CPU which is 10 cores at 2.4Ghz and
90watts.

Dual 5620's are 8 cores at 2.4Ghz and 170watts.

** E5 v4 is faster clock for clock than Westmere, so the 2.4Ghz isn't
equivalent between the two CPUs.

It's somewhere between 2x and 3x more computing power than dual Westmere
5620's, and uses nearly half the power at full load.  In practice it
will use far less than half the power as E5 is much better at powering
down idle cores and so forth than Westmere was.  Not to mention single
socket, which will reduce your platform cost and power draw there. 
There's an additional win, in that FreeBSD doesn't scale well with
multi-package.

As far as motherboards there are a lot of choices,
https://www.supermicro.com/products/nfo/Xeon_X10_E5.cfm?pg=MB=SELECT=UP#List_MBD
shows a bunch of different PCI and SAS/SATA port and memory
configurations. Something in the X10SR family should work well for you.

That's the "less than half the power draw" option.  You can go with
something that has 1/4 the power draw of your current setup, but it will
likely be an incremental performance improvement (~25%)

-- 

Thanks,

Josh Paetzel
___
freebsd-hardware@freebsd.org mailing list
https://lists.freebsd.org/mailman/listinfo/freebsd-hardware
To unsubscribe, send any mail to "freebsd-hardware-unsubscr...@freebsd.org"


Re: Server performance & enhancement

2017-02-16 Thread Bob Bishop
Hi,

> On 16 Feb 2017, at 11:20, Karl Denninger  wrote:
> 
> I'm wondering what people think of a forward path for the following:
> 
> 
> CPU: Intel(R) Xeon(R) CPU   E5620  @ 2.40GHz (2400.14-MHz
> K8-class CPU)
>  Origin="GenuineIntel"  Id=0x206c2  Family=0x6  Model=0x2c  Stepping=2
> 
> Features=0xbfebfbff
> 
> Features2=0x29ee3ff
>  AMD Features=0x2c100800
>  AMD Features2=0x1
>  VT-x: PAT,HLT,MTF,PAUSE,EPT,UG,VPID
>  TSC: P-state invariant, performance statistics
> real memory  = 25773998080 (24580 MB)
> avail memory = 24915828736 (23761 MB)
> Event timer "LAPIC" quality 600
> ACPI APIC Table: <111612 APIC1749>
> FreeBSD/SMP: Multiprocessor System Detected: 16 CPUs
> FreeBSD/SMP: 2 package(s) x 4 core(s) x 2 hardware threads
> 
> The system board is a Supermicro and has embedded IPMI (which I require
> for remote admin purposes) on a separate interface.  It's plenty fast
> but very power-hungry.
> 
> I'm interested in other's experience with various system board options,
> and processors, that will return /better /performance and which have a
> good history running FreeBSD /but with lower power consumption.
> 
> /Any suggestions welcome.

We’ve had good experiences with Intel’s own servers based on the S2600Gx 
m’boards but might be overkill for you.

> -- 
> Karl Denninger
> k...@denninger.net 
> /The Market Ticker/
> /[S/MIME encrypted email preferred]/

--
Bob Bishop
r...@gid.co.uk




___
freebsd-hardware@freebsd.org mailing list
https://lists.freebsd.org/mailman/listinfo/freebsd-hardware
To unsubscribe, send any mail to "freebsd-hardware-unsubscr...@freebsd.org"