[WBEL-users] White Box wins election on Super Tuesday
Bill Davidsen
davidsen@tmr.com
Fri, 09 Apr 2004 09:01:12 -0400
William Warren wrote:
>
>
> Bill Davidsen wrote:
>> Note that in most applications this is a non-issue, but one 120w HT
>> CPU has less heat than two 84w CPUs if you want to check real SMP
>> operation. I'll build the dual Opteron sometime in the next year, but
>> I haven't even ordered parts because I'm waiting for the O/S to be
>> stable.
>
> HT is not true SMP jsut some more registers that allow the cpu to fake
> it..this is not true smp..you are not getting full smp benefits from HT
> jsut a faked dual-cpu...
> http://www20.tomshardware.com/cpu/20021202/hyperthreading-02.html
> I would take the benefits of two 89w cpu's over one 120w faked dual cpu
> any day..:)
I have about 30 Xeon systems with HT, and not only does the application
run 20-30% faster (tps) with HT on, the CPUs run 6-8C cooler. This has
been discussed on lkml and comp.sys.intel if you want to follow the
comments of folks who have seen similar results.
All running Linux as servers for an integer heavy application, bib
memory, fast disk, etc, etc.
I can't say that HT makes the CPU use less power, I can say that many
people have noted it generates less heat (so I infer the power). I would
say that 30% more power and 30% more performance are about a wash in
tps/w for my applications, clearly that won't be true on the desktop.
The other advantage of HT is is cost and real estate on the system
board. With Intel SMP means Xeon, although the price has come down and
the larger cache gives a slight performance advantage. With Opteron you
need to go to the SMP capable CPU, so cost is non-linear there as well.
The real issue is that HT is free in current Intel CPUs, and it does
provide the gain without the cost and complexity of multiple CPUs. No
reason not to use it.
--
-bill davidsen (davidsen@tmr.com)
"The secret to procrastination is to put things off until the
last possible moment - but no longer" -me