Re: Advice on system purchase
On 11/1/2012 11:42 AM, Martin Steigerwald wrote:
> Am Montag, 29. Oktober 2012 schrieb Stan Hoeppner:
>>> For powerful laptops and power saving desktops I think Intel
>>> Sandybridge/Ivybridge is best bet currently - except for the
>>> political dimension.
>> Sure, but 90% of users don't need "powerful". All the cores sit idle
>> most of the time, and a faster CPU doesn't make Thunderbird or Firefox,
>> IE or Outlook express, go any faster. Nor any of the standard desktop
>> apps. 90% of users would benefit more from a low wattage dual or even
>> single core CPU, with an SSD instead of a rust drive.
>> But it's hard to sell people on the truth after you've been lying to
>> them about the benefits of 4-8 core CPUs for many years...
> Yeah, as you pointed out its about peak performance.
No, it has nothing to do with peak performance. What I said was that,
in a nut shell, a 2.5-3GHz dual core CPU from AMD or Intel is more
powerful than what 99.99% of users need. Yet AMD/Intel are only selling
4/6/8 core desktop CPUs today. It's a waste of cores and a waste of money.
> But if you look at my powertop snapshots you will see that the Linux
> Kernel mostly switched between 800 MHz and turbo mode, i.e. give me all
> MHz you can squeeze out of the CPU given the current load on other cores
> and GPU. Especially if regularily it does it mostly on one core. If that
> core can go faster it may have a noticable effect.
> So a higher peak performance may have a beneficial effect on latency in
> short high load situations like starting an application or other high
This has nothing to do with the case I've made. You're talking about
battery saving CPU frequency scaling on a laptop. This has nothing to
do with the case I've made.
> Whether thats noticable? Well, one would have to test it.
> I think to see any difference during application load times tough you need
> to have a good SSD alongside.
CPU performance has little to do with app load times, with most
productivity apps anyway. Load time is dictated by disk latency. If
the application's binary and libraries have been cached then CPU makes
more difference, but it's a small difference.
> But I do believe that the kernel pings
> between 800 MHz and turbo mode not for nothing.
I have no idea what point you're making here.
> That said: By all means, I bet there are numerous AMD CPUs out there that
> are just fast enough already. So I think we basically agree with
> difference in detail scope.
The top performing CPUs from 3-4 generations ago (4-5 years), both AMD
and Intel, were already fast enough for the average user. Simply
switching to an SSD gives these users far more additional performance
than a faster many-core chip.