Barcelona / Phenom and x264

[plonk420]

now, you all must forgive me for not understanding what's involved in h.264/x264 encoding. it looks like AMD's answer to C2D accels at floating point calculation and "SSE128"... will this benefit x264 encoding?


http://anandtech.com/cpuchipsets/showdoc.aspx?i=3092
http://anandtech.com/IT/showdoc.aspx?i=3091

[Sharktooth]

as it does with C2D except for the FP that is not used almost at all in x264.

[Sagittaire]

hmmm ... for the same clock core seem better. And core is certainely able to have higher clock than phenon. This CPU is not a revolution ...

[Sharktooth]

due to the massive amount of cache coz DDR-2 667 was used for testing and that was a clear advantage for core.
however we should wait the phenom to draw some serious conclusions.
however as i always said native quads scale better.

[Sagittaire]

however as i always said native quads scale better.

Certainely if you compare 1x4 core and 4x1 core but I have serious doubt for 1x4 vs 2x2. Core 2 is certainely really more efficient than phenon at the same clock and core 2 is certainely able to have higher clock and by far. IMO c2q 6600 at 2.4 Ghz will produce better overall result than phenon at 2.5 Ghz.

[Sharktooth]

im not sure. the barcelona L3 cache speed is related to the core speed.
however here they have a 2.5Ghz sample to test.
http://techreport.com/articles.x/13176/1

[simonhowson]

Hopefully we will see some doom9 x264 (GG you say its the Liberals fault but i dont see the Constitution saying the Commonwealth is the only authority on housing issues….) benchmarks on these Barcelona Opterons. I'm anxious to see how much the improvement is. My guess is 10% clock for clock over K8.

[Sharktooth]

10% is just BS. look at the WME comparison: http://techreport.com/articles.x/13176/9
the test is limited to 4 cores but with more threads barcelona scales better.
looking at the whole benchmarks results, the new opteron has an average gain of 80% clock for clock over the K8.

[simonhowson]

10% is just BS. look at the WME comparison: http://techreport.com/articles.x/13176/9
the test is limited to 4 cores but with more threads barcelona scales better.
looking at the whole benchmarks results, the new opteron has an average gain of 80% clock for clock over the K8.

I hadn't seen that benchmark, but I'm not sure how well WMV encoding compares to x264. The comparson I'd like to see would be 2 X Dual Core opterons compared to 1 X Quad Core Barcelona at the same clock speed.

There are no benchmarks with only a single quad core opteron, so it is hard to make a direct comparison, but for this benchmark, my calculation is the Opteron is 32% faster at the same clock speed? I could be wrong though.

[Sharktooth]

uhm, no... in another benchmark they used 2 dual core opterons comparing them to the new barcelona, and the result is greatly in favour of the new barcelona quadcore.
anandtech states the difference is about 15% but there are other tests he didnt perform showing even a 100% increase...

some other benchs can be found here: http://www.techwarelabs.com/reviews/processors/barcelona/index_3.shtml

[simonhowson]

Does anyone know the difference between SSE4.1 and SSE4A?

Is SSE4A simply AMDs version of SSE4.1?

As I undersatnd it Peryn CPUs show SSE4.1, because they only include a subset of the full SSE4 features. The CPU-ID screen shots for the Barcelona CPUs show SSE4A, which I guess is the same thing, but don't know for sure.

Hopefully SSE4A includes the most important SSE4 features.

[akupenguin]

SSE4A contains only a few instructions, none of which are useful to x264.

[simonhowson]

SSE4A contains only a few instructions, none of which are useful to x264.

Do you know if SSE4.1 has anything useful to x264, or is it too early to say?

[Sharktooth]

another report: http://www.techarp.com/showarticle.aspx?artno=443&pgno=0

[JnZ]

I'd like to see would be 2 X Dual Core opterons compared to 1 X Quad Core Barcelona at the same clock speed.....

I have 2 x dual core Opteron 285@ 2.8GHz. So if someone tests Barcelona and give me this sample file, I can make such comparsion.

I must say, that this my quad is very good for x264 encoding. I have almost 10fps with HDDVD encoding with almost maximum settings.

Boys, do you have someone Intels quad? It will be interesting to make comparsion Intel Qxxxx vs 2xOpteron 285 on x264 encoding speed. At least it will be very interest for me.

[Manao]

JnZ : go check http://forum.doom9.org/showthread.php?t=122318. It contains a benchmark based on x264, and a CPU database containing already a lots of different CPUs, either Intel's or Amd's. There are some Quads in there.

[akupenguin]

SSE4.1 contains some useful instructions. However, I'm not sure the instruction touted as being designed for video compression (mpsadbw) is among the useful ones.

[Mutant_Fruit]

10% is just BS. look at the WME comparison: http://techreport.com/articles.x/13176/9
the test is limited to 4 cores but with more threads barcelona scales better.
looking at the whole benchmarks results, the new opteron has an average gain of 80% clock for clock over the K8.
Every benchmark i've read has put the new chips as 'slower than the core architecture' which is all that really is important at this stage. K8 is old and slow. Thats not what AMD are competing with.

If you look at the conclusion in that report you'll see they say:

or one, it is not going to capture the overall performance lead from Intel soon, not even in "Q4," which is when higher-clocked parts like the Opteron 2360 SE are expected to arrive.


In other words, unless AMD start pricing competitively (which will unfortunately make them hemorrhage money unless they can reduce the costs of fabrication beyond what intel can reduce their costs to), the core2 is still a better option.

[plonk420]

i'd say it's at LEAST a .5 iteration above the original K8 platform... i guess we just need to see actual benches with x264 :P ... the WME encoder benches are at least not a BAD sign...

[simonhowson]

I have 2 x dual core Opteron 285@ 2.8GHz. So if someone tests Barcelona and give me this sample file, I can make such comparsion.


Thanks for making this offer. I am wondering if you can run Graysky's new x264 benchmark? You can read about it here (http://www.techarp.com/showarticle.aspx?artno=442&pgno=0) and download it here (
http://www.techarp.com/x264_Benchmark/x264-Benchmark.rar). Use 7-Zip (http://www.7-zip.org/download.html) to decompress RAR files. You must extract it to ' c:\work2 ' else it won't work.

If you have time, can you also run this older benchmark (http://forum.doom9.org/showthread.php?t=122318). It's still very useful, because there is so much data created using it.

I've emailed some people with Barcelona based servers asking them to run the same benchmarks. Hopefully I will have those results to compare against yours in the next couple of days.