Advanced Benchmark v.2 for H264, H265, AV1 and H266

[Sagittaire]

Originally Posted by rco133

|-----------------------------------------------------------------|--------------------------------------------------
| CPU                                                             | AMD Ryzen 9 7950X 16-Core Processor
|-----------------------------------------------------------------|--------------------------------------------------
| H264 - x264 - 1080p - HDR - Multipart encoding - 1611 frames    |  36.296 s
|-----------------------------------------------------------------|--------------------------------------------------
| H265 - x265 - 1080p - HDR - Multipart encoding - 1611 frames    |  41.920 s
|-----------------------------------------------------------------|--------------------------------------------------
| AV1 - SVT-AV1 - 1080p - HDR  - Multipart encoding - 1611 frames |  54.638 s
|-----------------------------------------------------------------|--------------------------------------------------
| H266 - VVEnc - 1080p - HDR  - Multipart encoding - 1611 frames  |  83.910 s
|-----------------------------------------------------------------|--------------------------------------------------
| Overall - 1080p - HDR  - ABR Ladder encoding - 1611 frames      |  251.548 s
|-----------------------------------------------------------------|--------------------------------------------------

You are at default BIOS setting for 7950X or you use overclocking and/or PBO?

[rco133]

Originally Posted by Sagittaire

You are at default BIOS setting for 7950X or you use overclocking and/or PBO?

I don't use any overclocking, and not quite sure what PBO is. Don't think I have anything special done in BIOS.

[GeoffreyA]

|-----------------------------------------------------------------|--------------------------------------------------
| CPU                                                             | AMD Ryzen 3 2200G with Radeon Vega Graphics
|-----------------------------------------------------------------|--------------------------------------------------
| H264 - x264 - 1080p - HDR - Multipart encoding - 1611 frames    |  342.661 s
|-----------------------------------------------------------------|--------------------------------------------------
| H265 - x265 - 1080p - HDR - Multipart encoding - 1611 frames    |  456.775 s
|-----------------------------------------------------------------|--------------------------------------------------
| AV1 - SVT-AV1 - 1080p - HDR  - Multipart encoding - 1611 frames |  621.730 s
|-----------------------------------------------------------------|--------------------------------------------------
| H266 - VVEnc - 1080p - HDR  - Multipart encoding - 1611 frames  |  828.223 s
|-----------------------------------------------------------------|--------------------------------------------------
| Overall - 1080p - HDR  - ABR Ladder encoding - 1611 frames      |  2176.045 s
|-----------------------------------------------------------------|--------------------------------------------------

[Selur]

You are at default BIOS setting for 7950X or you use overclocking and/or PBO?

The difference is probably due to memory clocking, iirc. I clocked my RAM down for it to be more stable with 64GB RAM in the system.
My RAM was even slower than I thought, accidentally it was @3600 ,

@4800:

|-----------------------------------------------------------------|--------------------------------------------------
| CPU                                                             |  Ryzen 9 7950X 16-Core Processor
|-----------------------------------------------------------------|--------------------------------------------------
| H264 - x264 - 1080p - HDR - Multipart encoding - 1611 frames    |  37.496 s
|-----------------------------------------------------------------|--------------------------------------------------
| H265 - x265 - 1080p - HDR - Multipart encoding - 1611 frames    |  44.487 s
|-----------------------------------------------------------------|--------------------------------------------------
| AV1 - SVT-AV1 - 1080p - HDR  - Multipart encoding - 1611 frames |  56.433 s
|-----------------------------------------------------------------|--------------------------------------------------
| H266 - VVEnc - 1080p - HDR  - Multipart encoding - 1611 frames  |  85.840 s
|-----------------------------------------------------------------|--------------------------------------------------
| Overall - 1080p - HDR  - ABR Ladder encoding - 1611 frames      |  248.640 s
|-----------------------------------------------------------------|--------------------------------------------------

5400:

|-----------------------------------------------------------------|--------------------------------------------------
| CPU                                                             |  Ryzen 9 7950X 16-Core Processor
|-----------------------------------------------------------------|--------------------------------------------------
| H264 - x264 - 1080p - HDR - Multipart encoding - 1611 frames    |  36.231 s
|-----------------------------------------------------------------|--------------------------------------------------
| H265 - x265 - 1080p - HDR - Multipart encoding - 1611 frames    |  44.252 s
|-----------------------------------------------------------------|--------------------------------------------------
| AV1 - SVT-AV1 - 1080p - HDR  - Multipart encoding - 1611 frames |  55.768 s
|-----------------------------------------------------------------|--------------------------------------------------
| H266 - VVEnc - 1080p - HDR  - Multipart encoding - 1611 frames  |  84.954 s
|-----------------------------------------------------------------|--------------------------------------------------
| Overall - 1080p - HDR  - ABR Ladder encoding - 1611 frames      |  247.151 s
|-----------------------------------------------------------------|--------------------------------------------------

Running @6400 isn't stable with 64GB on my system.

[Sagittaire]

Originally Posted by Selur

The difference is probably due to memory clocking, iirc. I clocked my RAM down for it to be more stable with 64GB RAM in the system.
My RAM was even slower than I thought, accidentally it was @3600 ,

Running @6400 isn't stable with 64GB on my system.

Really interessing. For me codec compression was always independant to the RAM speed ... but in fact no.

Anyway run RAM at 6400 Mhz is useless. The sweet spot for Zen 4 is 6000 Mhz and could be really stable with same synchronisation for RAM and CPU. Have higher speed for RAM is useless and even negative for performances.

If 6000 Mbps isn't stable at default EXPO, Higher CAS Latency will be better for performances than RAM-CPU desynchronisation.

In any case, this bench will have been very useful for you Selur: your CPU was not at good peak performance.

[Z2697]

Originally Posted by Selur

The difference is probably due to memory clocking, iirc. I clocked my RAM down for it to be more stable with 64GB RAM in the system.
My RAM was even slower than I thought, accidentally it was @3600 ,

DDR5-3600 is very unusual
According to AMD, 6000 is the "sweet spot", I'm currently running 32x2 at 6000 MT/s with no problem.

[Z2697]

My 7950X is lame
Or is it the result of having too many background programs running

|-----------------------------------------------------------------|-------------------------------------------------- 
| CPU                                                             |  Ryzen 9 7950X 16-Core Processor                              
|-----------------------------------------------------------------|-------------------------------------------------- 
| H264 - x264 - 1080p - HDR - Multipart encoding - 1611 frames    |  36.081 s                
|-----------------------------------------------------------------|-------------------------------------------------- 
| H265 - x265 - 1080p - HDR - Multipart encoding - 1611 frames    |  45.244 s                
|-----------------------------------------------------------------|-------------------------------------------------- 
| AV1 - SVT-AV1 - 1080p - HDR  - Multipart encoding - 1611 frames |  56.145 s                
|-----------------------------------------------------------------|-------------------------------------------------- 
| H266 - VVEnc - 1080p - HDR  - Multipart encoding - 1611 frames  |  91.032 s                
|-----------------------------------------------------------------|-------------------------------------------------- 
| Overall - 1080p - HDR  - ABR Ladder encoding - 1611 frames      |  264.344 s                
|-----------------------------------------------------------------|--------------------------------------------------

[Z2697]

Crank up the fan and leave the computer undisturbed (background programs are limited not completely closed)
I'm planning on 9950X upgrade, gotta do a better comparison when that happens, and perhaps compare that "awful" inter CCD latency with the corresponding AGESA fix.

|-----------------------------------------------------------------|-------------------------------------------------- 
| CPU                                                             |  Ryzen 9 7950X 16-Core Processor                              
|-----------------------------------------------------------------|-------------------------------------------------- 
| H264 - x264 - 1080p - HDR - Multipart encoding - 1611 frames    |  35.127 s                
|-----------------------------------------------------------------|-------------------------------------------------- 
| H265 - x265 - 1080p - HDR - Multipart encoding - 1611 frames    |  43.569 s                
|-----------------------------------------------------------------|-------------------------------------------------- 
| AV1 - SVT-AV1 - 1080p - HDR  - Multipart encoding - 1611 frames |  53.986 s                
|-----------------------------------------------------------------|-------------------------------------------------- 
| H266 - VVEnc - 1080p - HDR  - Multipart encoding - 1611 frames  |  87.040 s                
|-----------------------------------------------------------------|-------------------------------------------------- 
| Overall - 1080p - HDR  - ABR Ladder encoding - 1611 frames      |  250.485 s                
|-----------------------------------------------------------------|--------------------------------------------------

[Sagittaire]

Originally Posted by Z2697

Crank up the fan and leave the computer undisturbed (background programs are limited not completely closed)
I'm planning on 9950X upgrade, gotta do a better comparison when that happens, and perhaps compare that "awful" inter CCD latency with the corresponding AGESA fix.

Well I think that you must wait 9950X3D because I will be certainely the best CPU in area to make encoding ... and by far.

[Z2697]

Originally Posted by Sagittaire

Well I think that you must wait 9950X3D because I will be certainely the best CPU in area to make encoding ... and by far.

I heard that 3D V-Cache does not help encoding much (if any?), the lowered clock of the X3D CCD will be more impactful, IF the design of 9950X3D this time is similar to 7950X3D.

[Sagittaire]

Originally Posted by Z2697

I heard that 3D V-Cache does not help encoding much (if any?), the lowered clock of the X3D CCD will be more impactful, IF the design of 9950X3D this time is similar to 7950X3D.

1) My bench for H264, H265, H266 or AV1 say something else ...

2) 9800X3D is by far better than 9700x (105W bios) in SVT-AV1 encoding test:


3) 9950X3D will be not like 7950X3D. Frenquency will be the same than 9950X with optimized 3D V-Cache installation in CPU package.

[Z2697]

9800X3D looks interesting, but 9950X3D is only rumor now, of course I fully expect it to be as interesting as 9800X3D, but it still has several months to wait. I might just get a 9950X.

[birdie]

Originally Posted by Sagittaire

1) My bench for H264, H265, H266 or AV1 say something else ...

2) 9800X3D is by far better than 9700x (105W bios) in SVT-AV1 encoding test:

3) 9950X3D will be not like 7950X3D. Frenquency will be the same than 9950X with optimized 3D V-Cache installation in CPU package.

It's not far better, it just has a much higher TDP.

The 9700X in the test had a 65W TDP.

Here the test of 105W mode:

https://www.phoronix.com/review/amd-9600x-9700x-105w/6

Which means 9800X3D and 9700X at the same TDP are roughly the same for the tasks that do not require gobbles of L3 cache.

Also this:

[Sagittaire]

Originally Posted by birdie

It's not far better, it just has a much higher TDP.

The 9700X in the test had a 65W TDP.

Here the test of 105W mode:

https://www.phoronix.com/review/amd-9600x-9700x-105w/6

Which means 9800X3D and 9700X at the same TDP are roughly the same for the tasks that do not require gobbles of L3 cache.

Also this:

Well in my previous post, I don't speak about old x264 codec but recent codec like SVT-AV1.

And you speak about Phoronix: Phoronix make test for 9700X (at 105W) and 9800X3D for SVT-AV1 for 1080p in mono scession encoding and 9800X3D is simply the best CPU in area. 3D V-Cache seem have enormous impact for modern codec.

[birdie]

Originally Posted by Sagittaire

Well in my previous post, I don't speak about old x264 codec but recent codec like SVT-AV1.

And you speak about Phoronix: Phoronix make test for 9700X (at 105W) and 9800X3D for SVT-AV1 for 1080p in mono scession encoding and 9800X3D is simply the best CPU in area. 3D V-Cache seem have enormous impact for modern codec.

Enormous? 9% doesn't look like anything "enormous".

The price difference on the other hand? Yeah, absolutely, more than 50%. Are you ready to pay 50% more to get 9% faster encoding? I am not.

[Z2697]

9% more fps is enormous, and of course 50% on price is more enormous.
Those things aside, svt-av1 preset 13? I'll give a big big question mark here, seriously? Preset 13? What?

Now I can understand where "9700X vs 14700K +42% handbrake performance" in AMD's Zen5 publication comes from LOL.

[Sagittaire]

Originally Posted by birdie

Enormous? 9% doesn't look like anything "enormous".

The price difference on the other hand? Yeah, absolutely, more than 50%. Are you ready to pay 50% more to get 9% faster encoding? I am not.

1) Well it's not me that speak about phoronix ... ;-)

2) It's not me that say 3D V-Cache have no impact for encoding ... ;-)

3) 9950X3D will be certainely the best CPU in area for multipart encoding or ABR ladder encoding scenario (full charge CPU at 100% like for my benchmark)

4) It's possible to have more than 9%:

[Z2697]

For aomenc / libaom-av1, speed 8 / cpu-used 8 (whatever they are called) realtime is also just the fastest, lowest quality preset you have in that encoder.

All those things aside, the both results you posted do show one thing: V-Cache parts of Zen4 tends to have lower encoding speed compared to non-V-Cache parts, but for Zen5 things are different, even quite a lot in specific cases.

Specific cases, being the keyword here. When I see something like 13600K being faster than 14900K, I'd think the result is highly unreliable.

[Sagittaire]

Originally Posted by Z2697

For aomenc / libaom-av1, speed 8 / cpu-used 8 (whatever they are called) realtime is also just the fastest, lowest quality preset you have in that encoder.

Well it's Phoronix bench and not mine. In my bench I use veryslow preser for x264, slow preset for x265, complexity 4 for SVT-AV1 and fast preset for VVCEnc. The objective is to have comparable and usable speed for H264, H265, H266 and AV1 and make quality comparison.

All those things aside, the both results you posted do show one thing: V-Cache parts of Zen4 tends to have lower encoding speed compared to non-V-Cache parts, but for Zen5 things are different, even quite a lot in specific cases.

In fact no, when you make fine analyse. 7700X is 105W CPU with really higher max clock than 7800X3D at 65W. And 7800X3D produce same speed than 7700X. Thus 3D V-cache work really well for encoding with Zen 4 too. In fact, if you want make comparison for Zen 4, you must compare 7700(no X) at 65W and 7800X3D at 65W too.

Specific cases, being the keyword here. When I see something like 13600K being faster than 14900K, I'd think the result is highly unreliable.

Here it's really complex to analyse. In fact 13xxx and 14xxx Intel generation have Pcore and Ecore. Ecore are really usefull if CPU charge is at 100%. Anyway in this bench, Intel CPU are not at 100% for CPU charge. And in this case unoptimized encoder will use Pcore and Ecore to make encoding. Ecore are efficient but really slower than Pcore. You have 16 Ecore in 14900K and only 8 Ecore in 13600K. And Phornix bench is Linux specialised. My analyse is than Linux is unable to direct the encoder to the fastest core (Pcore) and will use Pcore and Ecore indifferently. Windows does not have this problem for example and will choose Pcore as a priority if the CPU is not at saturated load. And in this case, you can have 13600K faster than 14900K for Linux OS.


My benchmark have not this problem (For Linux or Windows) because you sature all CPU at least with 32 threads and more if I want (I can saturate at 100% AMD Ryzen Threadripper PRO 7995WX with 192 threads for all codecs). I believe this is the only codec benchmark that can do this with H264, H265, H266 and SVT-AV1.

[Sagittaire]

Someone can make this bench with 14900K?