x265 Ryzen discussion [Archive] - Page 2

brumsky

5th March 2017, 15:25

While I must admit I don't have a solution to your color issue... I must say the loss of details specially on his shoulders is astoundingly awfully bad.

x265 really needs to massively improve in details retention in that specific area if it wants to be "taken seriously" compared to other HEVC encoders.

The loss of details is so ugly, it makes XviD (aka a 15 years old obsolete codec) look better in those regards. :eek:

I think if the coloring was correct then it wouldn't look as if as much detail was lost...

brumsky

5th March 2017, 15:32

Such output differences also make me wonder lately see this difference over here between AMD/NVIDIA and Intel outputs ;)

https://forum.doom9.org/showpost.php?p=1796861&postcount=207

though in your case the source is surely bt 709 and it seems you also flag it that way so it looks like a real chroma difference based on the encoder difference, overall pixel lose can be percepted as a color/lighting difference in your case 3rd Generation.

Most interesting is how his beard changes the perception of it becoming more brownish in your output then black

I'm running these tests on a Xeon E5-2683 v4, if that helps at all.

Any thoughts\ideas on how I can correct this? Is the encode bitstarved? I don't think so because increasing CRF from 21->20 is virtually imperceptible - which is about 31% more bitrate. Even increasing CRF from 21->19 doesn't make a note worthy difference.

Sagittaire

5th March 2017, 15:33

Ryzen 1800X does consume significantly more than advertised:

yes but efficacity is really good:
https://forum.doom9.org/showthread.php?p=1799462#post1799462

by far better than all 4C/8T intel CPU. In fact only i7-6900K and i7-6950K are better for fps/power efficacity for x264 encoding (CPU at 100%).

mandarinka

5th March 2017, 15:36

Ryzen 1800X does consume significantly more than advertised:

(...)

Completely wrong, as explained here (https://forum.doom9.org/showpost.php?p=1799497&postcount=4926):

If your numbers are based on wall power consumption, then they are inflated.
TDP (95 W) is only a rating for the CPU itself. When you meassure power draw on the 12V line for the CPU, then the power consumption you see is actually the draw of the CPU itself + losses of the VRM circuits (up to 20 %). If you measure power consumption of the whole PC at the wall, then you have another 15-20 % of power losses added on top, in addition to the inherent lack of precision in such scenario. Also your measured total consumption will include ramped-up CPU cooling fan (2-4 W, some coolers are rated for 0,7A/12V!) or water AIO pump (even more probably).

So when considering high 85% efficiency for both the PSU and VRM, 95W power consumed by CPU becomes 125 W delta measured, for illustration (95*1,15*1,15 = 125,6).

You can't just subtract idle power from load power at the wall, how many times do I have to repeat this?

According to your "math", i7-6950X would have "TDP" of 165 W.

CruNcher

5th March 2017, 16:13

Efficiency is ok but no compare to Intel Kaby Lake as a overall platform, they came close to Sandy Bridge but Intel is already 1 step ahead already of that what AMD released now ;)

Hype is never objective ;)

if you want to compare the efficiency you take a mobile platform and compare them and no desktop anyways ;)

In this case Kaby Lake Laptop vs Raven Ridge as a whole System.

The whole release here was targeted for Gamers entirely every Review shows that, there are some reviewers though that go deeper ;)

https://img.purch.com/o/aHR0cDovL21lZGlhLmJlc3RvZm1pY3JvLmNvbS9JL0kvNjU2NDQyL29yaWdpbmFsLzAyLVdhdHRhZ2UtQ3VydmVzLnBuZw==

Would have been nice to have the exact for HEVC and AVC Decoding/Encoding.

Though lot of Reviewer missing significant important data About Power Management setup and Timings, also it seems AMD has the Corepark issue that Intel fixed since some time now (Haswell) ;)

Though don't forget idle data is highly dependent on the GPU used in the review as well as the whole Mainboard target (Component Price Efficiency, useless overhead) ;)

LoRd_MuldeR

5th March 2017, 16:39

Even though a new CPU generation is interesting for the development of x265, I wonder if such a specific talk about hardware dependencies deserves a thread separate from a rather general and previously more software oriented thread?

@ Moderation: Split?

Split. Please use new thread for x265 Ryzen discussion.

Bloax

5th March 2017, 16:41

I would just like to say that it's looking very much like the 329 bux RZ 1700 is pretty much equal to the 489 bux Rz 1800x if you manually overclock it (why wouldn't you), so you should be basing your opinions on that rather than doing erroneous i7-7700k vs. RZ 1800x comparisions.

And for being the same price as the i7-7700k, the Rz 1700 sure blows it out the window for x264/5.

NikosD

5th March 2017, 16:56

Ryzen R7 1700 with its price and performance is the best CPU out there for everything but one.

It's best VFM (value for money) and can replace every CPU and platform for any workload but one.

It can be used as an everyday office/Internet desktop PC or a powerful content creation/productivity/rendering/video encoding workstation.

The only thing missing is 1080p gaming.

For anyone with a powerful GPU and a suitable 1440p or 4K monitor, the Ryzen R7 1700 is the best cpu to buy, again.

But not for 1080p, unfortunately.

That is the one and only weakness of the whole AMD platform/ecosystem.

I hope they find a solution soon.

CruNcher

5th March 2017, 16:56

Sadly Tomshardware made no direct Platform compare (you never really visualize that to hurt sales and make your industry sponsors unhappy, and of course we all need that AMD sells something) like that French Reviewer with his limited Data you have to take out older Review data of Intel Systems and compare yourself ;)

Also what we see here is only a Glimpse of the Efficiency with most a Nvidia Card tested which is in itself crazy the power of this Platform is entirely lying in it's very high efficiency Scalable Modularity in combination with AMD Hardware/Software ;)

And Developers now have to even faster progress on their Multithreading Efficiency ;)

We can hope that the Value will catch buyers so Software improvements get fast on they way, which will become the real value for everyone.

Bloax

5th March 2017, 17:16

Yeah, it's not a total beast in games due to having trouble hitting 4 Ghz (alongside the new Not-Hyperthreading causing performance issues seemingly due to mis-assignment of CPU threads), but this is a video encoding forum, and for that it absolutely wrecks. :)

I guess there's a shitstorm about it because someone thinks it's a terrible CPU or something? :D

Personally then I'm tempted to see if I can set up a capture card -> encoding machine -> vapoursynth -> knlmeans -> x264 streaming machine at some point, with the CPU to crunch that final step being a 1700.

CruNcher

5th March 2017, 17:20

iwod

5th March 2017, 17:29

Pretty much All games and programs on the PC platform are optimised for Intel. AMD, lacking the optimisation now and being completely new uArch, shows huge potential in its current form already.

Hopefully we see more software update that brings even more performance improvement.

NikosD

5th March 2017, 17:29

I guess there's a shitstorm about it because someone thinks it's a terrible CPU or something? :D

I think normal people, not Intel fanboys, are commenting its gaming weakness mainly because it's inconsistent with the rest of its performance and even DX12 multithreaded games are slower than DX11 version of that same game.

Those things shouldn't happen on the release date.

AMD should have already solved them with MS (Windows 10 scheduler) and/or the developers, months ago (and certainly before release time)

All the other reasons to complain can and will be solved with new BIOS/microcode and small patches to applications having issues (see winrar for example)

But gaming could be more difficult to resolve.

Let's go back to x265 now ;)

CruNcher

5th March 2017, 17:37

Did anyone test Ryzen even DX12 with a AMD Card ?

Most reviews i saw yet test with Nvidia and Nvidia Drivers on Win 10 :D

Any RX 480 based Vulkan/DX12 Review GCN based ?

or even better complete Platform compares (would most probably take average Reviewers to long on release needing clicks fast)

NikosD

5th March 2017, 17:40

Yes with R9 Nano and using Win 10 when you disable SMT it gets faster.

Using Win 7, RyZen is about 17% faster (!) in games and probably in general.

Win 7 doesn't suffer from SMT wrong scheduling.

Sagittaire

5th March 2017, 17:48

It will even more wreck for Video Encoding if what i said happens it's not only about Game (VR) Engines ;)

Thus why im absolutely not really interested in this Flagship but Raven Ridge at best with 6 Cores though only 4 will be realizable at first ;)

The G/S series is what im waiting for on AM4 with VEGA IP Internally and external :D

well most test use GTX 1080 for GPU at 700$, i7 7700K at 350$, certainely with top motherboard and RAM ... etc etc etc

If you use configuration between 1500 and 2000$, it's not to play at 1080p with screen at 200$ but more probaly with 1440p or 4K screen with Gsync. And in these condition, you are "GPU limited" and not "CPU limited", and all top CPU produce same fps.

If you want play in 1080p, you use GTX 1060 or RX 480, and with these GPU, all top CPU produce the same fps.

here test with RX 480 (~GTX 1060):
https://www.cowcotland.com/articles/2234-12/test-processeur-amd-ryzen-7-1800x.html
https://www.cowcotland.com/articles/2234-11/test-processeur-amd-ryzen-7-1800x.html

CruNcher

5th March 2017, 17:48

I think you will be disappointed in these "8 cores" of Zen given the lack of AVX units. When AMD refers to "FPU", they actually mean the vector unit. Benchmarks (https://forums.anandtech.com/threads/ryzen-strictly-technical.2500572/) in x265 show same performance at stock as 7700K. It might appear to be on-par with Intel EE SKUs, but the difference is that 1800X comes out of the factory at maximum clock speed, whereas the i7-EE can go to 4.5 GHz (Haswell) or 4.3 GHz (Broadwell), which will yield another 33% over the 3.0 GHz base clock.

BTW: If cost and power efficiency in threaded applications is your goal, E5-2699v4 (22x 2.2 GHz) is selling for $1750 on Amazon (https://www.amazon.com/INTEL-PROCESSOR-E5-2699V4-2-2GHZ-SMART/dp/B01DTYQH4G)and EBay.

Yes i very much expect Intels new Mainstream 6 Core that's coming as reaction faster then expected to destroy it alone at first.

Im pretty sure it will have AVX1/2 by default

Sagittaire

5th March 2017, 17:52

Using Win 7, RyZen is about 17% faster (!) in games and probably in general.

not for x264 or x265.

There are real problem for 7zip or winrar for exemple on Rysen. It's a specific problem on sub-memory system.

NikosD

5th March 2017, 17:54

I think you will be disappointed in these "8 cores" of Zen given the lack of AVX units. When AMD refers to "FPU", they actually mean the vector unit.

Legacy x87 and SSE instructions are very fast on RyZen due to the 128bit quad issue FPU (2 FADD + 2 FMUL units)

The AVX/AVX2 implementation is slower due to 128bit units vs 256 bit, but only for Haswell and onwards

Benchmarks (https://forums.anandtech.com/threads/ryzen-strictly-technical.2500572/) in x265 show same performance at stock as 7700K. It might appear to be on-par with Intel EE SKUs, but the difference is that 1800X comes out of the factory at maximum clock speed, whereas the i7-EE can go to 4.5 GHz (Haswell) or 4.3 GHz (Broadwell), which will yield another 33% over the 3.0 GHz base clock.

You compare apples with oranges.

The overclocked HEDT Intel's CPUs have already a higher TDP and real power consumption figures.
By overclocking them, you only make worse the power/perf ratio which is very good at stock clocks for both platforms.

And as we have already said, if you want to overclock a RyZen CPU then you buy 1700 (non X version)

BTW: If cost and power efficiency in threaded applications is your goal, E5-2699v4 (22x 2.2 GHz) is selling for $1750 on Amazon (https://www.amazon.com/INTEL-PROCESSOR-E5-2699V4-2-2GHZ-SMART/dp/B01DTYQH4G)and EBay.

Very expensive platform overall and not a good overall system with such a low clock.

Sagittaire

5th March 2017, 17:56

Using Win 7, RyZen is about 17% faster (!) in games

I wait this test!
where is the source, please?

NikosD

5th March 2017, 18:13

You are missing the point entirely. If you want threaded performance without sacrificing single-threading, the EE will go to 4.5 GHz (compared with 5 GHz on 7700K), whereas the 1800X can't go beyond 4.1 GHz (and neither can the 1700). E5-2699v4 is incredibly energy efficient, and it has a 3.6 GHz turbo frequency for single-threaded applications (3.3 GHz for four threads). Motherboards are available (https://www.newegg.com/Product/Product.aspx?Item=N82E16813182933)for under $400, which is far cheaper than four X370 boards, PSUs, etc. would cost.

What is the power consumption of HEDT@4.5 GHZ ?

Thermal requirements ?

Cost of CPU and platform ?

Are you seriously comparing a 330$ RyZen R7 1700 with anything else ?

You sure are missing the whole point entirely.

CruNcher

5th March 2017, 18:15

Intel is already reacting to this though with their Mainstream 6 Core which wont be limited instruction wise this time and surely competitively priced.

And especially overall more efficient compensating the 2 cores with the instruction efficiency and gaining the same IPC if not even slightly better.

The non IGPU Chip has in reality no chance Intel can always react fast.

Especially in a unoptimzed Software state the Zen 8 core will be crushed with pure benchmark based reviews by Intels 6 Cores this time, the same Game as previously nothing changed.

Only that the overall efficiency is not that huge apart anymore like it was with Sandy Bridge vs Bulldozer ;)

And AMDs CPUs can't be called heatpipes anymore ;)

AMD needs todo what they can best concentrate on the GPU and HSA, Intel could get ground their 2 using it's shrinking advantage and investing every time saved (AMD was running behind) into the GPU Core Developement.

NikosD

5th March 2017, 18:22

In order for i7 7700K to outperform 1800X in x265, you have to use settings like anandtech's review that are not multithreaded friendly and do not saturate the CPUs.

It's like you are doing that on purpose.

And I'm saying to you again, that overall a 1700 system is best buy compared to i7 7700K or any other CPU that Intel can offer.

Only 1080p gaming can justify a i7 7700K purchase.

Atak_Snajpera

5th March 2017, 18:24

i7-7700K costs $300 and outperforms 1800X in x265. It uses less power as well. Consumers that buy EE CPUs are not interested in energy efficiency, but want MT performance with no compromise in ST. For pure MT performance, as I noted before, Intel has many competitive offerings like the E5-2600v4 series (on sale) or the Xeon-D series for workloads not requiring memory bandwidth.

That's a lie.
http://i.imgsafe.org/c49a5accf4.png

Atak_Snajpera

5th March 2017, 18:37

@Stephen R. Savage

If you had 8C/16T you would already know that x265 CAN'T saturate 16T with 1080p source! I have E5-2690 (8C/16T) so I know what i'm saying. Basically you need two instances of x265 running at the same time to reach constant 100% cpu usage.

CruNcher

5th March 2017, 18:45

NikosD

5th March 2017, 18:48

Sorry but only testing with x265 is blatantly dumb

Sorry, but you have to say that to people testing x265 only.

I'm not that guy, that's why I'm talking about overall system performance from the beginning.

Are you here or are you dreaming ?

Do you even read my posts ?

I can justify people judging CPUs based on x265 performance, only because we are writing on this thread about x265 RyZen.

But I wouldn't call them dump people like you.

hajj_3

5th March 2017, 19:11

I wait this test!
where is the source, please?

https://forums.anandtech.com/threads/ryzen-strictly-technical.2500572/page-8#post-38775732

CruNcher

5th March 2017, 20:49

Sorry, but you have to say that to people testing x265 only.

I'm not that guy, that's why I'm talking about overall system performance from the beginning.

Are you here or are you dreaming ?

Do you even read my posts ?

I can justify people judging CPUs based on x265 performance, only because we are writing on this thread about x265 RyZen.

But I wouldn't call them dump people like you.

Here even the AMD Technical Marketing guy Robert Hallock and im pretty sure he speaks the truth he's more tech guy then marketing ;)

so we have a IPC difference of -6.8% to Kaby Lake confirmed by AMD themselves so im pretty sure the Mainstream 6 Core comming will hit that target they testing here 6/12 Threads overall more efficiently then Ryzen alone ;)

https://youtu.be/TBf0lwikXyU?t=496

Robert Hallock

https://www.youtube.com/watch?v=6Fy1_BEr1DE

Though OBS isn't very efficient on the Capture Side so many Cycles wasted their overall allready

Crazy in their Disclaimer they talk about a I7-6900K tested but in the Video they show a I7-7700K as result ?

Then once they say VBR result and then CBR result are they all MAD over @ AMD ;)

Remembers me about that Nvidia Intern that once upon a time forgot to mention the GTX 970 cutoff MC in the process of the cluster cuts ;)

So you overclock a I7-6900K and then call it I7-7700K ?

https://ark.intel.com/products/94196/Intel-Core-i7-6900K-Processor-20M-Cache-up-to-3_70-GHz
https://ark.intel.com//products/97129/Intel-Core-i7-7700K-Processor-8M-Cache-up-to-4_50-GHz

Thats how you create your results apart from the architectural enhancements just do it like Gamers do it bruteforce and expect the efficiency not to fall apart and create thread deadlocks in the process, very nice way testing the competitors frame drop issues indeed ;)

also overal that fast encoding setup is a real joke

When he talks about better quality :D

Its common wisdom in the streaming community that encoding video on the processor is the best way to go

Sorry but with that Setup it's blatantly nonsense vs ASIC and FPGA efficiency nowadays.

I hate Marketing especially when it comes from someone who should know better, lying in the faces of all who see that Video that's crazy.

Sagittaire

5th March 2017, 22:16

That's a lie.
http://i.imgsafe.org/c49a5accf4.png

well this test is strange. I have other that show opposite result:

http://www.kitguru.net/wp-content/uploads/2017/03/x265.png

and in this test i7-7700K 4C/8T is at 100% for CPU, and all other 8C/16T CPU are not at 100% of charge.

How in first test i7-7700K (CPU 100%) can be at 10% to R7 1800X (CPU 100%) ... and in second test i7-7700K (CPU 100%) can be at 34% to R7 1800X (CPU less than 100%)?

Moreover how i7 7700K at 4.2 Ghz (CPU at 100%) can be at only 34% to i7 6900K at 4.3 Ghz (CPU at 100%) with twice core/thread in multi instance encoding for x265?

In all test on the net, i7 7700K is certainely at 100% for CPU charge and you have better result for R7 1800X with less than 100% for CPU charge.

Sagittaire

5th March 2017, 23:16

and computer base produce really good preview too:

https://www.computerbase.de/2017-03/amd-ryzen-1800x-1700x-1700-test/3/#diagramm-handbrake

brumsky

5th March 2017, 23:22

Just found this review which runs hwbot's x265 benchmark tool.
http://www.overclockersclub.com/reviews/amd_ryzen_7_1800x_1700x_1700/8.htm

I downloaded and ran the benchmark on my Xeon E5-2683 v4 @ 2Ghz 16c/32t, I got 37.5fps. 1700x & 1800x at stock clocks beat my CPU... Once OC'd 1700-1800x smoke my score...

Sagittaire

5th March 2017, 23:23

and for game result is really interessing too:
https://www.computerbase.de/2017-03/amd-ryzen-1800x-1700x-1700-test/4/#abschnitt_benchmarks_in_uhd4k_bei_max_details

this test show that Rysen 7 is on par with all intel CPU for UHD even with Titan X. (GPU limited)

this test show that Rysen 7 well be on par with all intel CPU for FHD with GTX 1060 or RX 480. (GPU limited)

nevcairiel

5th March 2017, 23:41

If you end up GPU limited, its really no big surprise. The CPU isn't a crucial factor anymore. It only matters if you are CPU limited at least on one core, preferably more.

Sagittaire

6th March 2017, 00:08

If you end up GPU limited, its really no big surprise. The CPU isn't a crucial factor anymore. It only matters if you are CPU limited at least on one core, preferably more.

yes but if you use RX 480 or GTX 1060 (or less powerfull GPU) and you play on 1080p screen (or more) then i7 4770K, all Rysen R7 or i7 7700K will produce the same fps for game. And 95% gamers, at least, are in this configuration.

burfadel

6th March 2017, 04:16

There are three main things to know about Ryzen:

a bug in the Windows 10 scheduler such that often the SMT threads are used in favour of the Core threads, resulting in slowdown
--> Windows patch expected, or it may just be / already is? inclouded in the Windows Creators Update due next month
a driver for the CPU, allowing frequency and voltage adjustment etc every millisecond instead of 30 milliseconds, without reduction in power efficiency
--> coming soon (supposedly)
bios microcode and bios specific updates for RAM compatibility, allowing for faster RAM speeds
--> next month or so, encoding will benefit from faster RAM.

These three things together could have a signfiicant impact in certain workloads.

Atak_Snajpera

6th March 2017, 10:32

it is funny that older scheduler in win7 works better with newer cpu ;)

Sagittaire

6th March 2017, 11:13

it is funny that older scheduler in win7 works better with newer cpu ;)

It's just little bug in Win10 bor Ryzen.

Complete explication here:
http://www.hardware.fr/articles/956-24/retour-smt-mode-high-performance.html

You can solve SMT problem just with little modification in base register (+7% in game, just like SMTOff).

MS could make patch really quickly for that.

burfadel

6th March 2017, 13:11

That's a different thing, separate from the scheduler issue. The scheduler issue will be fixed in a Windows update, the performance gained through performance mode will be in a CPU driver.

AzraelNewtype

7th March 2017, 09:34

well this test is strange. I have other that show opposite result:

http://www.kitguru.net/wp-content/uploads/2017/03/x265.png

and in this test i7-7700K 4C/8T is at 100% for CPU, and all other 8C/16T CPU are not at 100% of charge.

How in first test i7-7700K (CPU 100%) can be at 10% to R7 1800X (CPU 100%) ... and in second test i7-7700K (CPU 100%) can be at 34% to R7 1800X (CPU less than 100%)?

Moreover how i7 7700K at 4.2 Ghz (CPU at 100%) can be at only 34% to i7 6900K at 4.3 Ghz (CPU at 100%) with twice core/thread in multi instance encoding for x265?

In all test on the net, i7 7700K is certainely at 100% for CPU charge and you have better result for R7 1800X with less than 100% for CPU charge.

Those results aren't opposite though? The youtube screencap was showing encoding time (lower is better), yours is showing fps (higher is better). In both cases, the Ryzen chip is showing up faster than the 7700K, which is what Atak_Snajpera was arguing. The 6800K falls between them in both too, but there's some disparity about the relative difference between them, particularly when it comes to the overclocks, but only because one of them is overclocked to different speeds between the two tests. It's not affecting the overall ranking of the four chips in common.

simonhowson

8th March 2017, 14:34

Motenai Yoda

8th March 2017, 19:11

I'm hoping there will be an analysis of Ryzen and x265 like this classic analysis of how Intel Nehalem offered dramatic speed improvements for x264.

https://web-beta.archive.org/web/20090115091348/http://x264dev.multimedia.cx

But Nehalem has a couple of stuff over penryn
and RyZen actually doesn't have nothing more, but an smt performance drop which nor intel's, nor older amd's cpus show.
First of all, the Nehalem has a much faster SSE unit than the Penryn. A huge number of SSE operations have had their throughput doubled:

All shuffle instructions
All basic math instructions (add, subtract, bitmath)
Many more complex math instructions (sign, absolute value, average, compare)
All unpack/pack instructions

These changes are hard to take advantage of: they naturally sped up a large number of functions, especially the Hadamard transform (which by definition is just a massive series of adds, subtracts, and unpacks). That is, lots of stuff got faster, but theres no obvious way (as far as Ive found so far) to leverage this for even more of an increase.

The cacheline split problem is basically gone: the penalty is now a mere 2 clocks instead of 12 for a cacheline-split load. This, combined with the SSE speed improvements, made it worthwhile to make SSE2 versions of width-8 SAD functions, despite the fact that this requires more instructions than the MMX versions. This also meant that all cacheline functions throughout x264 were no longer useful, and had to be disabled. One of the great benefits of this is not only making SAD faster, but that every function that made heavy use of unaligned loads got faster, even those with cacheline optimizations, but especially those without. The biggest examples, as per the graph earlier, are bipred and pixel_avg (qpel). To give an idea of the magnitude of this improvement, luma motion compensation for a 1616 block took 150 cycles on Penryn without cacheline split optimization, 111 cycles with, and takes 62 cycles on the Nehalem.

Intel has finally come through on their promise to make float-ops-on-SSE-registers-containing-integers have a speed penalty. So, we removed a few %defines throughout the code that converted integer ops into equivalent, but shorter, floating point instructions.

Sagittaire

8th March 2017, 19:39

But Nehalem has a couple of stuff over penryn
and RyZen actually doesn't have nothing more, but an smt performance drop which nor intel's, nor older amd's cpus show.

actually, the problem for x265 and rysen is that x265 is really unable to saturate 8C/16T CPU with default setting and 1080p source for exemple.

If you want saturate 8C/16T, you must make 4K encoding with x265. There are in fact really impressive potential speed improvement with x265. With CPU charge at 100%, x265 benchmark should be exactly like x264 benchmark, with same relative difference between CPU:

http://www.hardware.fr/getgraphimg.php?id=446&n=9

http://www.hardware.fr/getgraphimg.php?id=446&n=10

In this test, at CPU charge at 100%, R7 1800X and i7 6900K should be at ~12 fps and i7 6950K at ~13 fps.

burfadel

9th March 2017, 04:56

actually, the problem for x265 and rysen is that x265 is really unable to saturate 8C/16T CPU with default setting and 1080p source for exemple.

If you want saturate 8C/16T, you must make 4K encoding with x265. There are in fact really impressive potential speed improvement with x265. With CPU charge at 100%, x265 benchmark should be exactly like x264 benchmark, with same relative difference between CPU:

http://www.hardware.fr/getgraphimg.php?id=446&n=9

http://www.hardware.fr/getgraphimg.php?id=446&n=10

In this test, at CPU charge at 100%, R7 1800X and i7 6900K should be at ~12 fps and i7 6950K at ~13 fps.

Has anyone shown the per-core usage on the 8-core, 16-thread processors? I think it comes down to distribution of threads. Also keep in mind x265 is more complex than x264, and even through parallelisation it still has to wait for the data to complete in the relevant threads before the encode can proceed.

I assume by that graph that there are some actions that are delaying the progression of the encode, it would be good to know what they are then focus on improving their speed through parallelisation or code improvement.

There is also a bug in Windows 10 scheduler allegedly, once fixed the performance should improve a bit under Ryzen.

huhn

9th March 2017, 08:27

you should check out OC ram encoding with ryzen. it is possible already with none OC ryzen.

this is a test that shows remarkable gains for OC ram using on old skylake.
http://www.techspot.com/article/1171-ddr4-4000-mhz-performance/page2.html

OC ram is currently hard to use on ryzen and more tests are needed to confirm benchmarks like this.

broadwell E is using quad channel already so there should be a huge diminishing return for OC ram.

NikosD

9th March 2017, 08:30

According to this table, RyZen has obviously slower AVX2 integer calculations than Broadwell and Skylake/Kabylake.
The table has been created using the internal Instruction Latency and Throughput tool of AIDA64.

RyZen has half speed of Skylake/Kabylake's integer AVX2 addition and quarter speed of Skylake/Kabylake's integer AVX2 multiplication

Broadwell's AVX2 integer speed is between RyZen and Skylake/Kabylake

https://s12.postimg.org/789c14ict/CPU_chart_v16.png

The rest of Instruction Latency & Throughput table is here:

RyZen
http://users.atw.hu/instlatx64/AuthenticAMD0800F11_K17_Zen_InstLatX64.txt

Kabylake
http://users.atw.hu/instlatx64/GenuineIntel00906E9_Kabylake_InstLatX64.txt

We need an ASM developer of x265 to give us his feedback regarding most critical groups of Instructions Mix optimized in AVX2 to compare the architectures of RyZen vs Intel.

simonhowson

9th March 2017, 11:39

According to this table, RyZen has obviously slower AVX2 integer calculations than Broadwell and Skylake/Kabylake.
The table has been created using the internal Instruction Latency and Throughput tool of AIDA64.

RyZen has half speed of Skylake/Kabylake's integer AVX2 addition and quarter speed of Skylake/Kabylake's integer AVX2 multiplication
.
Yeah these were always going to be weaknesses, but I wonder if there are some parts of the Zen architecture that that may be better and can enable optimisations that won't work on recent Intel CPUs? I mean beyond the SMT implementation that seems to be very good.

LigH

9th March 2017, 11:49

Many people already complained about RyZen's AVX2 implementation being slower than top intel implementations; but I assume that it will still be a remarkable speed-up in comparison to not using AVX2 routines (in x265, specifically)? I am just reminded of Phenom-II implementing SSE3 so slow that x264/x265 won't consider using it...

ShogoXT

9th March 2017, 13:47

Keep in mind currently the higher memory multipliers are broken and the only to get stable beyond 2933 is to have a motherboard with a external clock generator. Currently only available on the ASRock taichi , fatality professional, Asus cross hair 6, and gigabyte aorus gaming k7. You also need single rank ram with Samsung b dies.

I bought my parts today, but went cheap. Ryzen 1700, gigabyte ab350 gaming 3, and gskill aegis 16gb kit 2x8gb 3000. I will probably oc to 3.8 GHz and ram to 2933 and leave it.

I'm have it setup by Saturday or Sunday. Let me know if you want me to test something.

Atak_Snajpera

9th March 2017, 13:56