[prev in list] [next in list] [prev in thread] [next in thread] 

List:       opensolaris-discuss
Subject:    Re: [osol-discuss] Another [OT] Hardware Post
From:       Erik Trimble <erik.trimble () oracle ! com>
Date:       2011-02-26 4:31:17
Message-ID: 1298694677.23338.61.camel () ghostbox
[Download RAW message or body]

On Fri, 2011-02-25 at 08:56 -0800, Erik Trimble wrote:
> On 2/25/2011 1:23 AM, Orvar Korvar wrote:
> > Regarding i7 uses too much power, and i5 is better - i5 doesnt suit my needs. I \
> > want plenty of CPU power when I need to and it does not matter if the cpu sucks \
> > much power when I am doing heavy calculations. The rest of the time, when I dont \
> > do heavy calculations, I want it to idle at very low wattage. I basically want a \
> > cpu with low wattage in idle and when I surf the web. When I need to, I want to \
> > have plenty of power and in that case it it allowed to suck much wattage. 
> > I would not like to have a weak cpu, and when I need to do some heavy work, I \
> > need to switch to another PC with a beefy CPU. Then I need two PC, one for web \
> > surf etc and one for heavy work. I want one PC, that has low power wattage in \
> > idle and surf the web, and plenty of muscles when the need rises. I suspect \
> > Sandybridge will idle at 50W. Or maybe I wait for Ivy bridge...
> 
> Like I said before, modern CPUs idle down quite far when under low load. 
> The Clarksdale i5, all SandyBridges (i5 & i7), Athlon II X4, and even 
> Phenom II X6, will shut down cores that aren't being used. So, for web 
> surfing, you're likely idling half of the CPU, a 65W TDP CPU should run 
> at 30W or less.  That's ALL the current-gen CPUs.
> 
> i7 single-threaded performance isn't really any better than i5, even on 
> SandyBridge-cores. SandyBridge single-thread is pretty much the same as 
> Clarksdale single-thread, and both are 10-15% faster than Athlon/Phenom 
> single-thread.   Exact performance depends on your application, as both 
> the i7 and Phenoms have significant L3 cache which can really help 
> certain apps (and be totally worthless for others).
> 
> Realistically, an i7 will be faster than an i5 on heavy memory-access 
> applications (they have a faster/wider memory bus). On low-memory access 
> stuff, an i7 won't be noticeably faster than an i5.  Also, an i7's 
> HyperThreading can be both a bane and a boon. For certain apps, having 
> more threads available is great; however, you can easily get into cache 
> thrashing with HyperThreading, so it's really best used when you have a 
> single app that wants 80% of the threads, not 10 apps all with one 
> thread each.  In the later case, you'd be better off with a Phenom II X6 
> (or, shutting off Hyperthreading, but then again, why not get an i5 in 
> that case?).   (I work in the Java VM group, and see lots of issues 
> around getting the best performance with HyperThreading on - how to 
> maximize your performance without stepping in the "death zones" where it 
> tanks.)
> 
> 

I'm going to followup on my post above, with a bit more information, and
a admission of incorrectness.

My experience with the SandyBridge architecture has been in the Xeon
dual- and quad-socket systems we use here, and test against similarly
configured Nehalem/Westmere Xeons and AMD Shanghai and MangyCours CPUs.

What I'm looking at now is several places which have benchmarked and
tested the i5-based SandyBrigde CPUs against i5/i7 Nehalem/Westmere
desktop chips, and the Athlon II/Phenom II chips.

The results differ from what my (server-oriented) experience has been:
to whit, that the SandyBridge architecture provides significant
performance boost over the Nehalem microarchitecture on most usages, and
is very noticably faster than the Athlon II/Phenom II configuration.

So, I take it back:  if you do indeed want maximum performance right
now, a i5-2000/3000-series SandyBridge chip is the way to go. And they
do pull significantly lower power than the Nehalem-based chips (though
not less than the Energy Efficient Athlon/Phenoms).


The issues around the graphics driver still are salient, however.  As,
of course, it the question of How Much Compute Power Does One Really
Need.


-- 
Erik Trimble
Java System Support
Mailstop:  usca22-317
Phone:  x67195
Santa Clara, CA
Timezone: US/Pacific (GMT-0800)

_______________________________________________
opensolaris-discuss mailing list
opensolaris-discuss@opensolaris.org


[prev in list] [next in list] [prev in thread] [next in thread]