Do AMD cpus at least give a smoother desktop experience w/more cores?

[TheELF]

And still nobody uses x264.exe to convert video,it's a commandline tool and it's not very good.
The only reason benchmarking sites use it is because it has a ready made benchmark...

Generic software out there uses ffmpeg or mencoder and everything else uses it's own implementation.

 

[The Stilt]

And still nobody uses x264.exe to convert video,it's a commandline tool and it's not very good.
The only reason benchmarking sites use it is because it has a ready made benchmark...

Generic software out there uses ffmpeg or mencoder and everything else uses it's own implementation.

And which x264 encoder do you think the "alternative solutions" use? ffmpeg uses libx264 and so does hybrid and mencoder. Or you recon the CLI of x264 makes the encoding slower?

Exactly the same thing, only the GUI differs.

Since nobody uses the GUI of x264 or x265 to encoder their videos, I guess I'm nobody then. I use them because I don't need any fancy GUIs to do what I want to do, and there is no third party *uckery potentially involved.

 

[AtenRa]

You said it, not fair. That's because the Intel SMT "core" is < 1/10 of the relative size of AMD CMT core :sneaky: On Excavator the hardware for the non-BSC core require around < 56% of the area of a BSC core. On Broadwell and Skylake the hardware required for SMT takes < 5% of the area of a native core.

EDIT: Broadwell & Skylake > 5% to < 5%.

Well, in order to have SMT you need a wider core thus bigger (bigger front-end, bigger execution = more ports etc).
The CMT needs two smaller cores, so comparing only one CMT core to one SMT core is not "fair". You need to compare the entire CMT module to the SMT core (+HT) in order to get the maximum throughput potential from both of them.

That is also why comparing a dual Module four threads BD vs Quad Core Phenom II is not "fair" as well. It is like comparing a Core i3 (4x Threads) vs Core i5 (4x threads).

 

[TheELF]

And which x264 encoder do you think the "alternative solutions" use? ffmpeg uses libx264 and so does hybrid and mencoder. Or you recon the CLI of x264 makes the encoding slower?

Exactly the same thing, only the GUI differs.

Since nobody uses the GUI of x264 or x265 to encoder their videos, I guess I'm nobody then. I use them because I don't need any fancy GUIs to do what I want to do, and there is no third party *uckery potentially involved.

They sure don't use the the same version the benchmark uses which is from 2012...

 

[The Stilt]

They sure don't use the the same version the benchmark uses which is from 2012...

That's a valid argument only if the different CPUs are tested with different x264 versions. Otherwise it would be the same for all, obviously.

Anyway I made a quick test between the newest x264 build from Git and the one supplied in TechARP x264 5.01 benchmark package. The newest build, compiled with much newer GCC (5.3 x86-64) and YASM (1.30) was ~8% faster than the old one on Piledriver. I used raw 1080P YUV input with "slow" preset "crf 17.0" rate control.

 

[coercitiv]

A quick and dirty test on 6600k @ 4Ghz. The archived folder was the 3dMark setup folder (3GB, 7 files). All 7Zip settings were default. CB was run with 4 threads, 7z with 2-8t.

CB MT - 668cb
7z 4t - 3m:48s
7z 8t - 3m:25s
WinRar - 1m:42s (MT setting on, assume 4t)

CB + WinRar - 194 cb + 1m:48s

CB + 7z low priority - 666 cb + 2 hours estimate
CB + 7z 2t - 340-360cb + 7 minutes
CB + 7z 4t - 106 cb + 3m:56s
CB + 7z 8t - 33 cb + 3m:27s

In order to finish one CB run with 7z 4t running, I had to archive the folder aprox 1.5x times. When doing CB + 7z 8t that number increased to 4+ times. There was no variation in 7z times (±1s), but there was variation in CB scores while running CB multiple times during one archival run (during the CB + 7z 2t test)

As usual, increasing number of 7z threads squeezes a bit more performance out of the system, but does so at the cost of system responsiveness and any other thread that needs computation.

 

[The Stilt]

Now repeat the test with increased NB/L3 (uncore frequency) and MEMCLK :sneaky:

 

[coercitiv]

Now repeat the test with increased NB/L3 (uncore frequency) and MEMCLK :sneaky:

I would rather repeat the test with CB + WinRar for starters, see how that influences results. [added, see above]

If we were to drop operating frequency to 3Ghz to make it easier to compare with other CPUs, what uncore / memclk setting variation would you suggest as a 2nd data set?