[x265] Recent Frames/second benchmarks per platform and per clip?

[Tom Vaughan]

Mario,
You might want to try running encodes in the cloud.  For example, you can
rent a dual Xeon E5-2680 v2 server from Amazon EC2, running Linux, (a
c3.8xlarge instance) for spot pricing of about US $0.26/hour (spot pricing
varies, of course).  A single socket version is half the price, and there is
a range of smaller, cheaper instance types.  Linux instances are 1/6th the
price of Windows instances.  These are fast, powerful machines with lots of
memory and SSD storage, and they won't tie up your personal PC.

See http://aws.amazon.com/ec2/pricing/

Tom

-----Original Message-----
From: x265-devel [mailto:x265-devel-bounces at videolan.org] On Behalf Of Mario
Rohkrämer
Sent: Friday, September 26, 2014 1:11 PM
To: Development for x265
Subject: Re: [x265] Recent Frames/second benchmarks per platform and per
clip?

Thanks, Tom.

Yes, I already did batch testing, even just on a small video (QCIF foreman);
one with larger dimensions would be more accurate, but running it 3 times
each (to avoid mistakes by random slowdown effects due to surprising
activities, you know, it is Windows...) in the slowest presets will take
several hours per 10 seconds, so my hardware is not suitable anymore.
Nevertheless, I will stay curious, and I trust your efforts.

I already published a PDF with a few graphs here in this mailing list, I may
do that again soonish with additional results of more recent builds.


Am 26.09.2014, 21:59 Uhr, schrieb Tom Vaughan
<tom.vaughan at multicorewareinc.com>:

> Hi Mario,
> It's easy for anyone to create a script to run a batch of performance
> tests.
> With x265's csv log file feature, it's easy to review and chart your
> results.  We can share example scripts if that helps anyone.
>
> In addition to hand-tuned assembly code optimization for specific
> processor instruction sets (SSE, AVX, AVX2), the team is refactoring
> code to improve data movement and multithreading.  In genera the work
> we are doing to improve performance will be very positive for people
> running the latest hardware architectures, but it should not be
> negative for people running on older platforms.
>
> Tom
>
> -----Original Message-----
> From: x265-devel [mailto:x265-devel-bounces at videolan.org] On Behalf Of
> Mario Rohkrämer
> Sent: Friday, September 26, 2014 12:44 PM
> To: Development for x265
> Subject: Re: [x265] Recent Frames/second benchmarks per platform and
> per clip?
>
> Hi Raul.
>
> I am curious about the development of speed and efficiency too. But
> their use for me is rather marginal, without any AVX capable CPU.
>
> Especially recently, there was a lot of speed optimization mainly for
> AVX2.
> Way beyond AMD Phenom-II. I can only hope that it doesn't get slower
> for me...
>
>
> Am 26.09.2014, 20:59 Uhr, schrieb Tom Vaughan
> <tom.vaughan at multicorewareinc.com>:
>
>> Hi Raul,
>>
>> Yes, you can ask those kinds of questions on this list.
>>
>>
>>
>> We run a wide range of regression tests and performance tests weekly,
>> but for a number of reasons we don’t publish this test data publicly.
>> If there is a specific project that I can help you with, please
>> contact me directly.
>>
>>
>>
>> Thanks,
>>
>> Tom
>>
>>
>>
>>
>>
>> *From:* x265-devel [mailto:x265-devel-bounces at videolan.org] *On
>> Behalf Of *Raul Lopez
>> *Sent:* Thursday, September 25, 2014 4:37 PM
>> *To:* Development for x265
>> *Subject:* [x265] Recent Frames/second benchmarks per platform and
>> per clip?
>>
>>
>>
>> Hi All,
>>
>>
>>
>> Are there any recent reports of x265 encoding runs of known clips
>> that show the Frames/second and Y PSNR achieved in a certain platform
>> for a range of bitrates with certain encoding parameters?
>>
>>
>>
>> I am looking for something like:
>>
>>
>>
>> BQTerrace 1080p60, IDR refresh of 120, Delay of 1 second, preset
>> 'faster'
>> 15.3 Frames/second Y PSNR 33.23 dB for 3Mbps on Ivy Bridge i7 4C/8T
>> 2.3 GHz (I just generated that today using x265 code from Sep 23,
>> 2014).
>>
>>
>>
>> Thanks for any leads.  BTW, is this the correct x265 list to use for
>> these questions?
>>
>>
>>
>> RAUL
>
>


-- 

Fun and success!
Mario *LigH* Rohkrämer
mailto:contact at ligh.de
_______________________________________________
x265-devel mailing list
x265-devel at videolan.org
https://mailman.videolan.org/listinfo/x265-devel