[FFmpeg-user] Help with excessive CPU usage

[Reindl Harald]

Am 19.02.2014 20:00, schrieb Loadlinx:
> On Wednesday, February 19, 2014, Reindl Harald <h.reindl at thelounge.net>
> wrote:
> 
>>
>>
>> Am 19.02.2014 17:01, schrieb Tom Evans:
>>> On Wed, Feb 19, 2014 at 3:34 PM, Reindl Harald <h.reindl at thelounge.net<javascript:;>>
>> wrote:
>>>>
>>>> Am 19.02.2014 16:25, schrieb Tom Evans:
>>>>> On Wed, Feb 19, 2014 at 2:34 PM, Loadlinx <limiteddi at gmail.com<javascript:;>>
>> wrote:
>>>>>> Not identical, but 16 cores.
>>>>>> server1: E7330
>>>>>> server2: E7340 (slightly larger L2)
>>>>>>
>>>>>> Everything else but CPU is 1:1
>>>>>
>>>>> So why doesn't your E7340 support SSE4.1?
>>>>
>>>> why should it?
>>>
>>> One server runs fast enough for him; on that server ffmpeg reports it
>>> is using CPU features SSE 4.1
>>>
>>> One server does not run fast enough for him; on that server ffmpeg
>>> does not use SSE 4.1
>>>
>>> Don't be so pedantic
>>
>> thank you for strip the technical informations i posted
>> and instead stating that i am pedantic
>>
>> i used google to search the ark database for you and provided the
>> information that this CPU simply has no SSE4.1 capability, frankly
>> the E7330 as built by Intel also has *no* SSE4.1 support
>>
>> https://www.google.at/search?q=ark+englisch+E7330
>> Status: End of Life
>> Launch Date: Q3'07
>> Expected Discontinuance: Q1'2010
>> Processor Number: E7330
>> Lithography: 65 nm
>>
>> again http://en.wikipedia.org/wiki/SSE4#SSE4.1
>> These instructions were introduced with Penryn microarchitecture,
>> the 45 nm shrink of Intel's Core microarchitecture
>>
>> and no you do not just throw away a CPU like a Xeon-E7 which
>> costs around 2000 EURO
>>
>>
> Hard to believe that the process is the issue here. It it simply because of
> SSE4.1 support?

no, it improves performance but for sure not that massive
in general i responded to "So why doesn't your E7340 support SSE4.1" with facts

what makes me *really* wonder is that ffmpeg pretends that E7330 has SSE4.1
because as said above this CPU was built before SSE4.1 was introduced at all

so that is interesting in case where takes ffmpeg that information, or is
it *really* that CPU and what happens if ffmpeg believes for whatever
reason a CPU supports SSE4.1 or whatever SIMD instrcutions while it
actually does not - normally i would expect a segfault with a error
message like "unkown instruction"

at least such errors are what i have seen in "-mtune native" builds for
several packages trying to run them on a older VNMware host :-)

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 246 bytes
Desc: OpenPGP digital signature
URL: <http://ffmpeg.org/pipermail/ffmpeg-user/attachments/20140219/a253ad84/attachment.asc>