AMD Ryzen

With AMD Ryzen now released & people mentioning that the extra cores would help with editing, I was wonder if VRD would take advantage of the extra cores Ryzen brings or not ?
 

jmc

Active member
With AMD Ryzen now released & people mentioning that the extra cores would help with editing, I was wonder if VRD would take advantage of the extra cores Ryzen brings or not ?
https://www.pcper.com/reviews/Processors/AMD-Ryzen-7-1800X-Review-Now-and-Zen

I would expect it to use all 8 cores as it does in the Handbrake test with x264 etc.

What I wonder is if it will be twice as fast with VRD/X264 vs H264 like my 6 core Sandybridge-E (2012).
Would REALLY hate to lose that.

jmc
 

Dan203

Senior Developer
Staff member
It actually does hyperhtreading, so 16 threads. AFAIK even x264 can use that many threads effectively.
 

hydra3333

Member
Although I read somewhere today that SMT threads are 4x slower ... Fud ? Sounds plausible, go figure.
http://investorshub.advfn.com/boards/read_msg.aspx?message_id=129235033
How Soon Can We Expect Gaming Performance Improvements?
One particularly interesting recent discovery revealed a bug that’s significantly detrimental to Ryzen’s performance in Microsoft’s Windows 10 scheduler. The scheduler does not appropriately recognize Ryzen’s cache size and cannot distinguish physical cores from SMT threads. This in turn is causing it to often incorrectly schedule tasks in the much slower — approximately 4 times slower — SMT threads rather than primary physical core threads.

This is why we’ve seen performance dramatically improve in some games after disabling SMT. We’ve seen this clearly demonstrated in games optimized for heavy hyperthreading usage like the Total War series where disabling SMT improves performance by as much as 17%. Apart from impending game patches for the Total War series and games based on the Nitrous engine like Ashes of The Singularity, there are a number of additional near-term solutions that AMD and its board-partners are working on.
 

Dan203

Senior Developer
Staff member
I know a guy who built one of these. I had him run a test for me. Using Hanbrake we both encoded the same two 5 minute files, one 720p and one 1080i, to H.265. His was actually slower then my i7-7700k. Only about 1fps on the 1080i, but about 24fps on the 720p. The only thing he noted was that his machine was not running at 100% CPU during either enocde, so he ran the test again encoding both files simultaneously and was able to shave a minute off the total encode time. But even then he was still about 2 minutes slower then my i7. (~9 minutes vs ~7 minutes for me)
 

jmc

Active member
I know a guy who built one of these. I had him run a test for me. Using Hanbrake we both encoded the same two 5 minute files, one 720p and one 1080i, to H.265. His was actually slower then my i7-7700k. Only about 1fps on the 1080i, but about 24fps on the 720p. The only thing he noted was that his machine was not running at 100% CPU during either enocde, so he ran the test again encoding both files simultaneously and was able to shave a minute off the total encode time. But even then he was still about 2 minutes slower then my i7. (~9 minutes vs ~7 minutes for me)
Would love to see the VRD-X264 vs H264 encode results... is x264 twice as fast?

Thanks,
jmc
 
Anyway to run comparison tests with VRD ?

I know a guy who built one of these. I had him run a test for me. Using Hanbrake we both encoded the same two 5 minute files, one 720p and one 1080i, to H.265. His was actually slower then my i7-7700k. Only about 1fps on the 1080i, but about 24fps on the 720p. The only thing he noted was that his machine was not running at 100% CPU during either enocde, so he ran the test again encoding both files simultaneously and was able to shave a minute off the total encode time. But even then he was still about 2 minutes slower then my i7. (~9 minutes vs ~7 minutes for me)
 

Dan203

Senior Developer
Staff member
Unfortunately VRD doesn't support x265 yet.

FYI I just ran the same 720p test using NVEncode on an NVidia GTX1050 GPU and it got 590fps! That's nearly 8x faster then CPU x265. Insane! (yes we're working on adding NVEncode to v6 as well)
 
I meant in VRD usage based what's in the program now. For example, I compress using H.264 WTV format to 720P at 50%. I was trying to figure if VRD would benefit from the additional cores that Ryzen provides.
 

Dan203

Senior Developer
Staff member
The encoder we use in our consumer version, the one made by MainConcept, can not use more then 4 cores so even on a 4 core machine with hyperhtreading it doesn't run at 100%. The Pro product includes x264 which can use more cores and easily maxes out my machine. However it still can't max all 16 threads of the Ryzen. So unless you run multiple versions to do simultaneous encodes it's not going to do you much good.
 

jmc

Active member
The encoder we use in our consumer version, the one made by MainConcept, can not use more then 4 cores so even on a 4 core machine with hyperhtreading it doesn't run at 100%. The Pro product includes x264 which can use more cores and easily maxes out my machine. However it still can't max all 16 threads of the Ryzen. So unless you run multiple versions to do simultaneous encodes it's not going to do you much good.
Ok, I understand the above to be saying that ("it" means **X264**) can NOT max out a Ryzen 8/8HT cpu.
(I know that H264 will not)

So do you feel that the Skylake-X 8core/8HT consumer chip to come out later this year (I HOPE)
would likely be limited the same as the 8core/8HT Ryzen? (that it is a X264 issue and not a Ryzen issue)

That I would be back to running two VRD compressions at once even with X264?

With my 6core/6HT I stopped doing two compressions when I got X264.< twice as fast as H264.

Can not get anyone to respond here with any 8/8HT core X264/H264 test results.

Would be nice to save $500 if Ryzen is as good as Intel 8 cores but unless I can get real test results
I'll have to play it safe and go with the coming Skylake-X 8core.

Heh, maybe the Ryzen competition will make Intel come down off the crazy $1700 10 core chip price.

Thanks,
jmc
 
Last edited:

Dan203

Senior Developer
Staff member
I don't actually have one of these Ryzen CPUs and I may have overstated my relationship to the guy who does by referring to him as a "friend". He's a guy I know from another forum. I asked about running more tests using x264 but he hasn't responded, so not sure if I'll ever get those.

On the Plus side DanR is thinking about getting a Ryzen system, so if he does then we'll be able to run tests in house and report numbers to you.
 

jmc

Active member
As an aside, it appears as if your end-to-end performance for long encodes may be affected by your luck, eg whether windows and/or x264 jags "primary" threads or "hyperthreads" in any particular run (if this is true, and it is said to apply to both amd and intel) ?

https://wccftech.com/amd-ryzen-performance-negatively-affected-windows-10-scheduler-bug/

Do you feel lucky today ?
https://www.youtube.com/watch?v=8Xjr2hnOHiM&t=107
In my testing here I found that if you use the basic H264.MP4 profile then
it is faster if you turn OFF Hyperthreads. (I'm remembering... 13% faster)

If you crank up the quality settings as in selecting the L3 Preset in "Advanced"
then you want to keep HTs ON - faster.

With my 6 core/6HT 3930 Sandybridge @4.3 GHz anyway.

jmc
 
Last edited:

jmc

Active member
On the Plus side DanR is thinking about getting a Ryzen system, so if he does then we'll be able to run tests in house and report numbers to you.
Sounds good, fingers crossed.

Pretty sure it is going to be a while before Skylake-X 8core comes out.

My 6 core cpu is in it's 6th year and overclocked the whole way.
Just hope it hangs in there till I replace it.

Thanks!
jmc
 

Otter

Member
All of the above is interesting, but leaves out the most important aspect - what Profile and settings are being used by each program. Without some way to exactly duplicate the same settings in each encoder, you are comparing apples and oranges. I use VRD for all my editing and encode things that only need medium quality SD. The standard VRD profiles encode quickly on my 4.2ghz 8 core FX-8370 and give decent quality. At least on my processor, VRD/H264 does use all 8 cores equally with 60%-80% being used on each while encoding a single file. As said, the only way to hit 100% is to run 2 encodes simultaneously. I am planning to purchase a 1800X soon, but am waiting to see some opinions on the various X370 motherboards and let the BIOs shakeout before I choose one.

If I am creating 720p or 1080 files with lots of action and motion and seeking the best quality, I always Fast Recode and use x264 for the final recode. I have far more experience tweeking and experimenting with the x264. The results I get have smoother fast action, sharper detail for things like titles and less artifacting at edge thresholds. Especially in areas of the same color like blue skies or green walls, VRD gives distinct bands of the shading rather than a smooth gradient. The x264 gives a more natural transition with no noticeable banding. The flip side is that these tweeks I use in x264 to give better quality also give a much longer encode time.

So.. apples and oranges... your actual mileage may vary... but VRD/H264 can use 8 cores
 

Dan203

Senior Developer
Staff member
The test I had my friend run was using the same source file and the same "profile" in Handbrake, so it was apples to apples.

FYI I've got x265 and Intel QS HEVC running in debug version of VRD here so I now know how they compare in VRD. Since we don't use the full decoding chain, like Handbrake, our QS is a little slower but it's still pretty fast. I'm working on adding a NVEnc encoder right now. We'll see how that compares, but in MediaCoder I can get 750fps using it on my GTX1050. :D
 

hydra3333

Member
FYI I've got x265 and Intel QS HEVC running in debug version of VRD here so I now know how they compare in VRD ... it's still pretty fast. I'm working on adding a NVEnc encoder right now. We'll see how that compares, but in MediaCoder I can get 750fps using it on my GTX1050. :D
Heart stops with excitement and defibrillator applied.
 

jmc

Active member
The test I had my friend run was using the same source file and the same "profile" in Handbrake, so it was apples to apples.

FYI I've got x265 and Intel QS HEVC running in debug version of VRD here so I now know how they compare in VRD. Since we don't use the full decoding chain, like Handbrake, our QS is a little slower but it's still pretty fast. I'm working on adding a NVEnc encoder right now. We'll see how that compares, but in MediaCoder I can get 750fps using it on my GTX1050. :D
Seem to be seeing more and more references to VRD-6 (and HEVC)....GREAT!

Fingers crossed that the X264 and 2:3 pull down deinterlace probem will be fixed there.
Heh, been ordering from Amazon UK instead of Amazon US to get around that problem.

The 750fps made my jaw drop till I realized that was GPU not CPU result.

Is the CPU still the "go to" for quality or can you no longer see a difference?
(I would LOVE 750fps!)

Thanks,
jmc
 
Top Bottom