Thread: Performances with new Intel Core* processors

Performances with new Intel Core* processors

From
Jonathan Ballet
Date:
Hello,

I've read a lot of mails here saying how good is the Opteron with PostgreSQL,
and a lot of people seems to recommend it (instead of Xeon).

However, it seems that new Intel processors, Core Duo and Core 2 Duo, performs
very well, in desktop environment at least.


I wonder what can we expect with them, do anybody have done any experiments with
those processors ?


Regards,
    Jonathan Ballet

Re: Performances with new Intel Core* processors

From
"Merlin Moncure"
Date:
On 7/31/06, Jonathan Ballet <jon@multani.info> wrote:
> Hello,
>
> I've read a lot of mails here saying how good is the Opteron with PostgreSQL,
> and a lot of people seems to recommend it (instead of Xeon).
>

I am a huge fan of the opteron but intel certainly seems to have a
winner for workstations. from my research on a per core basis the c2d
is a stronger chip with the 4mb cache version but it is unclear which
is a better choice for pg on 4 and 8 core platforms.  I have direct
personal experience with pg on dual (4 core) and quad (8 core) opteron
and the performance is fantastic, especially on 64 bit o/s with > 2gb
memory (vs 32 bit xeon).

also opteron is 64 bit and mature so i think is a better choice for
server platform at the moment, especially for databases.  my mind
could be changed but it is too soon right now.  consider how long it
took for the opteron to prove itself in the server world.

merlin

Re: Performances with new Intel Core* processors

From
Arjen van der Meijden
Date:
On 31-7-2006 17:52, Merlin Moncure wrote:
> On 7/31/06, Jonathan Ballet <jon@multani.info> wrote:
>> Hello,
>>
>> I've read a lot of mails here saying how good is the Opteron with
>> PostgreSQL,
>> and a lot of people seems to recommend it (instead of Xeon).
>
> I am a huge fan of the opteron but intel certainly seems to have a
> winner for workstations. from my research on a per core basis the c2d
> is a stronger chip with the 4mb cache version but it is unclear which
> is a better choice for pg on 4 and 8 core platforms.  I have direct
> personal experience with pg on dual (4 core) and quad (8 core) opteron
> and the performance is fantastic, especially on 64 bit o/s with > 2gb
> memory (vs 32 bit xeon).

As far as I know there is no support for more than two Woodcrest
processors (Core 2 version of the Xeon) in a system. So when using a
scalable application (like postgresql) and you need more than four
cores, Opteron is still the only option in the x86 world.

The Woodcrest however is faster than a comparably priced Opteron using
Postgresql. In a benchmark we did (and have yet to publish) a Woodcrest
system outperforms a comparable Sun Fire x4200. And even if you'd adjust
it to a clock-by-clock comparison, Woodcrest would still beat the
Opteron. If you'd adjust it to a price/performance comparison (I
configured a HP DL 380G5-system which is similar to what we tested on
their website), the x4200 would loose as well. Mind you a Opteron 280
2.4Ghz or 285 2.6Ghz costs more than a Woodcrest 5150 2.66Ghz or 5160
3Ghz (resp.), but the FB-Dimm memory for the Xeons is more expensive
than the DDR or DDR2 ECC REG memory you need in a Opteron.

> also opteron is 64 bit and mature so i think is a better choice for
> server platform at the moment, especially for databases.  my mind
> could be changed but it is too soon right now.  consider how long it
> took for the opteron to prove itself in the server world.

Intel Woodcrest can do 64-bit as well. As can all recent Xeons. Whether
Opteron does a better job at 64-bit than a Xeon, I don't know (our test
was in 64-bit though). I have not seen our Xeon 64-bits production
servers be any less stable than our Opteron 64-bit servers.
For a database system, however, processors hardly ever are the main
bottleneck, are they? So you should probably go for a set of "fast
processors" from your favorite supplier and focus mainly on lots of
memory and fast disks. Whether that employs Opterons or Xeon Woodcrest
(no other Xeons are up to that competition, imho) doesn't really matter.

We'll be publishing the article in the near future, and I'll give a
pointer to it (even though it will be in Dutch, you can still read the
graphs).

Best regards,

Arjen van der Meijden
Tweakers.net

Re: Performances with new Intel Core* processors

From
Bruce Momjian
Date:
Good to know.  We have been waiting for performance comparisons on
the new Intel CPUs.

---------------------------------------------------------------------------

Arjen van der Meijden wrote:
> On 31-7-2006 17:52, Merlin Moncure wrote:
> > On 7/31/06, Jonathan Ballet <jon@multani.info> wrote:
> >> Hello,
> >>
> >> I've read a lot of mails here saying how good is the Opteron with
> >> PostgreSQL,
> >> and a lot of people seems to recommend it (instead of Xeon).
> >
> > I am a huge fan of the opteron but intel certainly seems to have a
> > winner for workstations. from my research on a per core basis the c2d
> > is a stronger chip with the 4mb cache version but it is unclear which
> > is a better choice for pg on 4 and 8 core platforms.  I have direct
> > personal experience with pg on dual (4 core) and quad (8 core) opteron
> > and the performance is fantastic, especially on 64 bit o/s with > 2gb
> > memory (vs 32 bit xeon).
>
> As far as I know there is no support for more than two Woodcrest
> processors (Core 2 version of the Xeon) in a system. So when using a
> scalable application (like postgresql) and you need more than four
> cores, Opteron is still the only option in the x86 world.
>
> The Woodcrest however is faster than a comparably priced Opteron using
> Postgresql. In a benchmark we did (and have yet to publish) a Woodcrest
> system outperforms a comparable Sun Fire x4200. And even if you'd adjust
> it to a clock-by-clock comparison, Woodcrest would still beat the
> Opteron. If you'd adjust it to a price/performance comparison (I
> configured a HP DL 380G5-system which is similar to what we tested on
> their website), the x4200 would loose as well. Mind you a Opteron 280
> 2.4Ghz or 285 2.6Ghz costs more than a Woodcrest 5150 2.66Ghz or 5160
> 3Ghz (resp.), but the FB-Dimm memory for the Xeons is more expensive
> than the DDR or DDR2 ECC REG memory you need in a Opteron.
>
> > also opteron is 64 bit and mature so i think is a better choice for
> > server platform at the moment, especially for databases.  my mind
> > could be changed but it is too soon right now.  consider how long it
> > took for the opteron to prove itself in the server world.
>
> Intel Woodcrest can do 64-bit as well. As can all recent Xeons. Whether
> Opteron does a better job at 64-bit than a Xeon, I don't know (our test
> was in 64-bit though). I have not seen our Xeon 64-bits production
> servers be any less stable than our Opteron 64-bit servers.
> For a database system, however, processors hardly ever are the main
> bottleneck, are they? So you should probably go for a set of "fast
> processors" from your favorite supplier and focus mainly on lots of
> memory and fast disks. Whether that employs Opterons or Xeon Woodcrest
> (no other Xeons are up to that competition, imho) doesn't really matter.
>
> We'll be publishing the article in the near future, and I'll give a
> pointer to it (even though it will be in Dutch, you can still read the
> graphs).
>
> Best regards,
>
> Arjen van der Meijden
> Tweakers.net
>
> ---------------------------(end of broadcast)---------------------------
> TIP 6: explain analyze is your friend

--
  Bruce Momjian   bruce@momjian.us
  EnterpriseDB    http://www.enterprisedb.com

  + If your life is a hard drive, Christ can be your backup. +

Re: Performances with new Intel Core* processors

From
Scott Marlowe
Date:
On Mon, 2006-07-31 at 11:30, Arjen van der Meijden wrote:
> On 31-7-2006 17:52, Merlin Moncure wrote:

> For a database system, however, processors hardly ever are the main
> bottleneck, are they? So you should probably go for a set of "fast
> processors" from your favorite supplier and focus mainly on lots of
> memory and fast disks. Whether that employs Opterons or Xeon Woodcrest
> (no other Xeons are up to that competition, imho) doesn't really matter.

Just making a quick comment here.  While the CPU core itself nowadays
certainly is not the most common bottleneck for a fast db server, the
ability of the CPU/Memory combo to act as a datapump IS often a limit.

In that case, you want to go with whichever setup gives you the fastest
access to memory.

Re: Performances with new Intel Core* processors

From
Vivek Khera
Date:
On Jul 31, 2006, at 12:30 PM, Arjen van der Meijden wrote:

> For a database system, however, processors hardly ever are the main
> bottleneck, are they? So you should probably go for a set of "fast
> processors" from your favorite supplier and focus mainly on lots of
> memory and fast disks. Whether that employs Opterons or Xeon
> Woodcrest (no other Xeons are up to that

No, but it *does* matter how fast said processor can sling the memory
around, and in my experience, the opterons have been much better at
that due to the efficiency of the memory transport layer.

Attachment

Re: Performances with new Intel Core* processors

From
"Luke Lonergan"
Date:
Vivek,

On 7/31/06 2:04 PM, "Vivek Khera" <vivek@khera.org> wrote:

> No, but it *does* matter how fast said processor can sling the memory
> around, and in my experience, the opterons have been much better at
> that due to the efficiency of the memory transport layer.

My Mac laptop with a Core 1 and DDR2 RAM does 2700 MB/s memory bandwidth.
The Core 2 also has lower memory latency than the Opteron.

That said - Intel still hasn't figured out how to do cache-coherent SMP
scaling yet - the Opteron has the outstanding EV6/HTX bus and the cc-numa
cache coherency logic working today.

- Luke



Re: Performances with new Intel Core* processors

From
Florian Weimer
Date:
* Arjen van der Meijden:

> For a database system, however, processors hardly ever are the main
> bottleneck, are they?

Not directly, but the choice of processor influences which
chipsets/mainboards are available, which in turn has some impact on
the number of RAM slots.  (According to our hardware supplier, beyound
8 GB, the price per GB goes up sharply.)  Unfortunately, it seems that
the Core 2 Duo mainboards do not change that much in this area.

--
Florian Weimer                <fweimer@bfk.de>
BFK edv-consulting GmbH       http://www.bfk.de/
Durlacher Allee 47            tel: +49-721-96201-1
D-76131 Karlsruhe             fax: +49-721-96201-99

Re: Performances with new Intel Core* processors

From
"Denis Lussier"
Date:
My theory, based entirely on what I have read in this thread, is that a low end server (really a small workstation) with an Intel Dual Core CPU is likely an excellent PG choice for the lowest end.

I'll try to snag an Intel Dual Core workstation in the near future and report back DBT2 scores comparing it to a similarly equiped 1 socket AMD dual core workstation.   I'll keep the data size small to fit entirely in RAM so the DBT2 isn't it's usual disk bound dog when you run it the "right" way (according to tpc-c guidelines).

--Denis
   Dweeb from EnterpriseDB

On 8/1/06, Florian Weimer <fweimer@bfk.de> wrote:
* Arjen van der Meijden:

> For a database system, however, processors hardly ever are the main
> bottleneck, are they?

Not directly, but the choice of processor influences which
chipsets/mainboards are available, which in turn has some impact on
the number of RAM slots.  (According to our hardware supplier, beyound
8 GB, the price per GB goes up sharply.)  Unfortunately, it seems that
the Core 2 Duo mainboards do not change that much in this area.

--
Florian Weimer                <fweimer@bfk.de>
BFK edv-consulting GmbH       http://www.bfk.de/
Durlacher Allee 47            tel: +49-721-96201-1
D-76131 Karlsruhe             fax: +49-721-96201-99

---------------------------(end of broadcast)---------------------------
TIP 9: In versions below 8.0, the planner will ignore your desire to
       choose an index scan if your joining column's datatypes do not
       match