Performance test shows hyperthreading effect

There have been a number of discussions here about hyperthreading, and
if it really provides a performance boost. An independent test of the
performance of virtual machines did reference testing on base
(non-virtual) hardware as a reference point, and just out of curiousity
ran the test with and without HT enabled. The performance gain was
significant, and is described in

http://www.webperformanceinc.com/lib...nce/index.html

My personal testing, on the real time to compile a Linux kernel, show a
consistent 25-30% decrease in clock time, using the same course and
hardware.

Other discussions, here and elsewhere, indicate that CPUs tend to run
cooler with HT enabled. Maybe a hardware type can explain that one.
--
Bill Davidsen
He was a full-time professional cat, not some moonlighting
ferret or weasel. He knew about these things.

On Apr 3, 9:24 am, Bill Davidsen <david...@tmr.com> wrote:
> There have been a number of discussions here about hyperthreading, and
> if it really provides a performance boost. An independent test of the
> performance of virtual machines did reference testing on base
> (non-virtual) hardware as a reference point, and just out of curiousity
> ran the test with and without HT enabled. The performance gain was
> significant, and is described in
>
> http://www.webperformanceinc.com/lib...stingVirtualiz...
>
> My personal testing, on the real time to compile a Linux kernel, show a
> consistent 25-30% decrease in clock time, using the same course and
> hardware.
>
Is this a surprise? I thought it was well understood that
hyperthreading is helpful in (and always was aimed at) server
applications, because server pipelines stall a high percentage of the
time. Most of the nay-saying and pooh-poohing of hyperthreading in
these groups was from people who wanted to do other things. Intel, of
course, never said hyperthreading will be much more useful in some
applications than in others because, regardless of its usefulness to
an arbitrary user, it was good marketing copy.

Robert.

Robert Myers wrote:
> On Apr 3, 9:24 am, Bill Davidsen <david...@tmr.com> wrote:
>> There have been a number of discussions here about hyperthreading, and
>> if it really provides a performance boost. An independent test of the
>> performance of virtual machines did reference testing on base
>> (non-virtual) hardware as a reference point, and just out of curiousity
>> ran the test with and without HT enabled. The performance gain was
>> significant, and is described in
>>
>> http://www.webperformanceinc.com/lib...stingVirtualiz...
>>
>> My personal testing, on the real time to compile a Linux kernel, show a
>> consistent 25-30% decrease in clock time, using the same course and
>> hardware.
>>
> Is this a surprise? I thought it was well understood that
> hyperthreading is helpful in (and always was aimed at) server
> applications, because server pipelines stall a high percentage of the
> time. Most of the nay-saying and pooh-poohing of hyperthreading in
> these groups was from people who wanted to do other things. Intel, of
> course, never said hyperthreading will be much more useful in some
> applications than in others because, regardless of its usefulness to
> an arbitrary user, it was good marketing copy.
>
It's a definitive end to the whining about HT being a marketing trick. ;-)

The interesting thing is that many people report their CPUs run slightly
cooler with HT enabled, so I'm puzzled by Intel not enabling it in the
single core mobile models, for both performance and power benefits. I'm
reasonably sure there was some wording in the original core2 information
which said something like "HT is not supported in these parts" or similar.

Question: does anyone on the list have any experience with the 955
PentiumD? If I remember the part number correctly, it's the only dual
core HT enabled CPU actually released.

--
Bill Davidsen
He was a full-time professional cat, not some moonlighting
ferret or weasel. He knew about these things.