[mythtv-users] [mythtv-commits] Ticket #11539: New capture card Hauppauge WinTV-HVR-950Q FE_GET_INFO No Such Device

[Gary Buhrmaster]

On Tue, May 14, 2013 at 1:46 PM, Raymond Wagner <raymond at wagnerrp.com> wrote:
> From: Jean-Yves Avenard
> Sent: Tuesday, May 14, 2013 9:19 AM
>> The fact is. Not all processors are equal with 64 bits mode.
>>
>> A core 2 duo, running a 64 bits kernel *is* slower than a 32 bits one.
>> Not so long ago, got a few reports that a 0.25 version of myth wasn't
>> playing a file well, when 0.24 was just fine. This was on a mac.
>> Turned out, by default I had made the app use 64 bits; because hey, 64
>> is at least twice better than 32 :)
>
>> Turned out, that's true on i3,i5 and i7, but not so on core 2 duo.
>
> There are some performance deficits when dealing with 64-bit applications,
> as a direct result of them being 64-bit. Many things are now immediately
> twice the size, which means for the same operation, you have twice the
> memory IO, and twice the cache usage, resulting in potential issues with
> cache coherency. Still, the advantages typically outweigh those deficits,
> with the improvement from an expanded instruction set and register space
> accounting for the losses in memory and cache.

Full disclosure, I run my Core 2 system in 64-bit mode.

Agree that on average, 64 bits will be better.  But the Core 2 Duo arch
had a couple of "gotchas" in 64 bit mode.  The first was the inability
to fuse instructions into a single micro-op.  The other was that while
the processor could decode up to 4 instructions per cycle, they had
to come from a 16 byte line, and instructions with the 64 bit prefixes
averaged more than that, resulting in occasional starving of the pipeline.
For most codes, the advantage of the additional registers, and, when
possible, taking advantage of new instructions, outweighed the small
penalties that the Core 2 Duo arch imposed, but there are always edge
cases (and your compiler optimizer could make a huge difference too).

Gary