Intel Alder Lake and VIdeo Rendering

supergafudo wrote on 10/29/2021, 4:03 AM

As you know new intel CPU uses a system with two core types, big and small, to save power. The windows scheduler direct the workload to the big or small cores in a way that would be very bad for video renders:

https://www.anandtech.com/show/16959/intel-innovation-alder-lake-november-4th/3

However, this way of doing things might be a bit of a concern, or at least it is for me. The demonstration that Intel performed was where a user was exporting video content in one application, and then moved to another to do image processing. When the user moved to the image processing application, the video editing threads were moved to the E-cores, allowing the image editor to use the P-cores as needed.

Now usually when I’m dealing with video exports, it’s the video throughput that is my limiting factor. I need the video to complete, regardless of what I’m doing in the interim. By defocusing the video export window, it now moves to the slower E-cores. If I want to keep it on the P-cores in this mode, I have to keep the window in focus and not do anything else. The way that this is described also means that if you use any software that’s fronted by a GUI, but spawns a background process to do the actual work, unless the background process gets focus (which it can never do in normal operation), then it will stay on the E-cores.

Comments

Reyfox wrote on 10/29/2021, 6:37 AM

and so far, power consumption has been extremely high.

eikira wrote on 10/29/2021, 8:31 PM

The only thing you may consider the new intel is the iGPU... its a little bit faster than in 11th generation. other than that nobody in their right mind would change if they really have an AMD 5900x or 5950x. As soon as dedicated Intel GPU comes out, i will try again vegas on my 5950x and if i get the same smooth h265 playback, i have no need for intel CPUs.

bvideo wrote on 10/29/2021, 8:42 PM

@supergafudo I wonder if that Windows setting to choose between favoring "background services" vs "programs" would have an effect on the scheduling you observe? (settings->system and security->advanced system settings->performance[settings] "Adjust for best performance of") And does Windows 10 even have the kind of scheduling you mention or is it all in hardware (thread director)? Also, how much CPU do the foreground (GUI-driven) processes have to use before fast cores are taken away from the background? In that article, "if the user changes the Windows Power Plan to high-performance, this behavior stops." That article doesn't really answer much.

Who's got a gen 12?

 

RogerS wrote on 10/29/2021, 9:16 PM

There's no way this would remain unaddressed as many people use Wintel for rendering and Intel would not want its newest hardware performing poorly in benchmarks or user reviews. So I'm not going to worry about it until this hits real world use.

Custom PC (2022) Intel i5-13600K with UHD 770 iGPU with latest driver, MSI z690 Tomahawk motherboard, 64GB Corsair DDR5 5200 ram, NVIDIA 2080 Super (8GB) with latest studio driver, 2TB Hynix P41 SSD, Windows 11 Pro 64 bit

Dell XPS 15 laptop (2017) 32GB ram, NVIDIA 1050 (4GB) with latest studio driver, Intel i7-7700HQ with Intel 630 iGPU (latest available driver), dual internal SSD (1TB; 1TB), Windows 10 64 bit

VEGAS Pro 19.651
VEGAS Pro 20.411
VEGAS Pro 21.208

Try the
VEGAS 4K "sample project" benchmark: https://forms.gle/ypyrrbUghEiaf2aC7
VEGAS Pro 20 "Ad" benchmark: https://forms.gle/eErJTR87K2bbJc4Q7

Hulk wrote on 10/29/2021, 10:56 PM

Maybe we should wait for reviews before making performance statements regarding Alder Lake? Keep in mind that Vegas is still heavily dependent on ST performance and Alder Lake will have the fastest Cores available with it's Golden Cove "P" cores. The reviews will tell the full story.

Former user wrote on 10/29/2021, 11:38 PM

The review embargo lifts on the day of sale, and they are likely to sell out very quickly, if you wait for reviews you won't have a CPU to buy. There should be reviews by people who legitimately bought them from retailers in Germany and elsewhere, but I haven't found any.

supergafudo wrote on 10/30/2021, 8:20 AM

@supergafudo I wonder if that Windows setting to choose between favoring "background services" vs "programs" would have an effect on the scheduling you observe? (settings->system and security->advanced system settings->performance[settings] "Adjust for best performance of") And does Windows 10 even have the kind of scheduling you mention or is it all in hardware (thread director)? Also, how much CPU do the foreground (GUI-driven) processes have to use before fast cores are taken away from the background? In that article, "if the user changes the Windows Power Plan to high-performance, this behavior stops." That article doesn't really answer much.

Who's got a gen 12?

 

Windows 10 do not have this new scheduler, is only for windows 11. At this time we do not know if Microsoft is going to let us to send workloads to the big cores at demand, but at the moment everything is automatic and this idea, that the user could manipulate the scheduler is not implemented in windows 11. They have designed the chip and the scheduler this way so the people that wants to buy alder lake for video editing should wait until the first reviews to see how negative is this effect in rendering times. Same for 3d rendering etc.. etc...

Howard-Vigorita wrote on 10/30/2021, 8:40 PM

I think the scheduler only kicks in for laptops with greater than 4 cores in low power or efficiency mode. I was glad not to see it active on my win11 11900k desktop set to the custom/performance power plan so I could tweak it myself in bios.

Hulk wrote on 10/31/2021, 8:16 AM

The new "Scheduler" to which you are referring is called the Thread Director and is a lower level version, which provides must better information regarding thread processing to the OS. Windows has been handling Big/Little cores in theory for many years with HT (or SMT with AMD CPU's). As with HT, there were some regressions in performance in the beginning but that has long since been worked out. It'll be the same with Alder Lake.

Let me provide an example of the Thread Director. Say you are running an application with 16 threads on the 12900K. Eight threads will be assigned to the Golden Cove cores but NO threads will be assigned to the GC HT cores. The remaining 8 threads will be assigned to the Gracemont cores, which makes sense because the Gracemont cores are much stronger than the HT Golden Cove threads from a compute point-of-view.

As we all know Vegas is still highly dependent on ST performance so Alder Lake on Vegas could be very promising...

supergafudo wrote on 10/31/2021, 10:38 AM

The new "Scheduler" to which you are referring is called the Thread Director and is a lower level version, which provides must better information regarding thread processing to the OS. Windows has been handling Big/Little cores in theory for many years with HT (or SMT with AMD CPU's). As with HT, there were some regressions in performance in the beginning but that has long since been worked out. It'll be the same with Alder Lake.

Let me provide an example of the Thread Director. Say you are running an application with 16 threads on the 12900K. Eight threads will be assigned to the Golden Cove cores but NO threads will be assigned to the GC HT cores. The remaining 8 threads will be assigned to the Gracemont cores, which makes sense because the Gracemont cores are much stronger than the HT Golden Cove threads from a compute point-of-view.

As we all know Vegas is still highly dependent on ST performance so Alder Lake on Vegas could be very promising...

The moment you will use your computer for any other task, windows will send the active threads to the Big Cores, letting the background render in the small ones, with a perfomance drop.

Hulk wrote on 10/31/2021, 11:06 AM

The new "Scheduler" to which you are referring is called the Thread Director and is a lower level version, which provides must better information regarding thread processing to the OS. Windows has been handling Big/Little cores in theory for many years with HT (or SMT with AMD CPU's). As with HT, there were some regressions in performance in the beginning but that has long since been worked out. It'll be the same with Alder Lake.

Let me provide an example of the Thread Director. Say you are running an application with 16 threads on the 12900K. Eight threads will be assigned to the Golden Cove cores but NO threads will be assigned to the GC HT cores. The remaining 8 threads will be assigned to the Gracemont cores, which makes sense because the Gracemont cores are much stronger than the HT Golden Cove threads from a compute point-of-view.

As we all know Vegas is still highly dependent on ST performance so Alder Lake on Vegas could be very promising...

The moment you will use your computer for any other task, windows will send the active threads to the Big Cores, letting the background render in the small ones, with a perfomance drop.

If that is the behavior and I assume you have tested it, then it's possible I may need the big cores on my foreground application. So that might be the correct behavior depending on the compute demands of the new foreground application. 4 (or 8) Skylake level Gracemont cores working on a background render while I have the Golden Coves to edit another project, work on a multitrack audio project, or play a game would again be the correct behavior.

If that other application is browsing the web then moving the GC's for that would be silly. We'll see if that's how ADL actually behaves soon enough. I'm not ready to shoot it down before official reviews are out.

Howard-Vigorita wrote on 11/4/2021, 11:49 AM

The more I hear about Alder Lake, the less impressive it sounds. Apparently Intel is co-developing it with a new Xeon code named Sapphire Rapids. Both are reported to share the same dye but the Alder Lake consumer cpu is reported to have large blocks physically disabled. Like avx512 for instance. While the Sapphire Rapids will keep it all plus an igpu option, direct hbm2 support, and up to 56 cores. Sounds like the one to get. Alder Lake is starting to look like a step sideways in comparison with at most a bump up in clock speed. But the support chipset could be a redeeming factor if it goes all pcie4.

It's also reported Intel is backing off a bit on the Alder Lake P/E core dichotomy because it can't figure out a way to give them different instruction sets, registers, or sizes. Starting to sound like they'll just be tagging cores based on thermals... inner cores running naturally hotter than border or corner cores while all are still time-shared off the same clock.

Reyfox wrote on 11/4/2021, 12:56 PM

Alder is good, but at a power premium which means you will have to also invest in high power cooling and power supply that has to be added to the cost. Is it the "bees knees"? No. Is it good. Yes. Would I upgrade? No. For the few percentage points of improvement, I can wait to see what AMD has coming next year that will fit into my AM4 socket.

Also, DDR5 costs way more and has higher latency than DDR4 for the moment.

I am waiting to see what Intel's graphics card do....

Former user wrote on 11/4/2021, 3:24 PM

The more I hear about Alder Lake, the less impressive it sounds. Apparently Intel is co-developing it with a new Xeon code named Sapphire Rapids. Both are reported to share the same dye but the Alder Lake consumer cpu is reported to have large blocks physically disabled. Like avx512 for instance. While the Sapphire Rapids will keep it all plus an igpu option, direct hbm2 support, and up to 56 cores.

The avx512 is still there and works when the E cores are disabled, and in normal mode it's disabled. I knew a lot of people were excited that they thought it was removed, believing the space it used on the silicon would be replaced with new technology, but no

Howard-Vigorita wrote on 11/4/2021, 4:06 PM

I've read that avx512 will still be there but moved out of the cores and disabled in Alder Lake. But it sounds like a course that can be easily reversed if they choose. In truth, however, I was a little disappointed by the small impact it contributes to ffmpeg performance with an 11900k when enabled in bios. Has a much more noticeable impact on prime95. And the bitcoin crowd thinks a whole lot of it too. So maybe ffmpeg just isn't optimized enough for it. I'm almost tempted to write something myself.

Former user wrote on 11/4/2021, 5:33 PM

. In truth, however, I was a little disappointed by the small impact it contributes to ffmpeg performance with an 11900k when enabled in bios.

And the advantage is only for software encoded X265 and who does that?

HEVC,VP9,AV1 , if I can't hardware encode 4K i'm not interested . Ofcourse it has it's uses where ever maximum efficiency and quality is required but for most hardware encoding is fine.

This is one of the articles about AVX-512

https://www.anandtech.com/show/17047/the-intel-12th-gen-core-i912900k-review-hybrid-performance-brings-hybrid-complexity/2

 

Former user wrote on 11/4/2021, 9:11 PM

Handbrake X264 and X265 encodes

supergafudo wrote on 11/5/2021, 5:34 AM

It looks that there are going to be some ways to bypass this:

https://www.anandtech.com/show/17047/the-intel-12th-gen-core-i912900k-review-hybrid-performance-brings-hybrid-complexity/3

Personally, I think this is a crazy way to do things, especially on a desktop. Intel tells me there are three ways to stop this behaviour:

Running dual monitors stops it

Changing Windows Power Plan from Balanced to High Performance stops it

There’s an option in the BIOS that, when enabled, means the Scroll Lock can be used to disable/park the E-cores, meaning nothing will be scheduled on them when the Scroll Lock is active.

Former user wrote on 11/5/2021, 7:07 AM

I've been reading that the IGPU is basically the same on 12900K as 11900K, but there is a difference
 

Multi-Format Codec Engines
  2

Intel® Quick Sync Video
     Yes

 

I don't know what that is, but on pro Nvidia cards, instead of having a single Nvenc encoder, they may have 2 or 3, and this increases the speed of encoding, I wonder if Intel have done the same here. Double Hardware encode speeds maybe or even independent encoders, so a screen recorder and a hardware encoder of a video editor never clash

Hulk wrote on 11/9/2021, 6:47 PM

As you know Alder Lake reviews are out and the video editing/encoding benchmarks are top of the charts. The 12900K often beats the 5950X mainly due to the fact that the Golden Coves in Alder Lake are currently the best performing single threaded cores available. As we all also know many apps, like Vegas Pro don't scale linearly with more cores so have those 8 Golden Coves really helps out with Vegas.

Also you will read reports of high power usage. That's true if you set the motherboard to unlimited or 241W or some other ridiculous value so you can show everybody your really high fps in game or some other benchmark. But the reality is that you can set the power limit to 125W or 150W and give up 5 or maybe 10% performance and have an efficient and cool running computer.

As for the Alder Lake UHD 770 GPU it's basically the Rocket Lake UHD 750 built on the new Intel 7 (10ESF) process with increase boost clocks.

VEGASPascal wrote on 11/10/2021, 2:22 AM

The VEGAS team is in good contact (biweekly) with the intel team and of cause we have got some Alder Lake test systems with DG2. We are not able to give you some performance numbers (drivers are limiting as laboratory version for testing) but we see some bigger performance improvements. The DG2 is interesting for encoding/decoding and Ai. The pressure for NVidia and Intel is high... Win11 and Alder Lake is a good step in the direction of M1 and Apple. We try our best to bring the efficiency with help of our strong partners (Microsoft, Intel, NVidia) into VEGAS.

Former user wrote on 1/5/2022, 6:37 PM

I've been reading that the IGPU is basically the same on 12900K as 11900K, but there is a difference
 

Multi-Format Codec Engines
  2

Intel® Quick Sync Video
     Yes

 

I don't know what that is, but on pro Nvidia cards, instead of having a single Nvenc encoder, they may have 2 or 3, and this increases the speed of encoding, I wonder if Intel have done the same here. Double Hardware encode speeds maybe or even independent encoders, so a screen recorder and a hardware encoder of a video editor never clash

Intel demonstrated HyperEncoding . Looks to be locked to Intel discreet GPU's . I don't think Vegas would benefit very much from this, unless doing multiple renders at once (parallel)

 

Hulk wrote on 1/5/2022, 9:42 PM

I'm hoping that ARC graphics cards will be readily available at launch. I like the idea that it will work with the iGPU in my 12700K.

Former user wrote on 1/5/2022, 10:23 PM

@Hulk Internet rumors have it that there's already an ARC product sitting at wholesaler warehouses but wasn't said which card it is. It's a nice bonus having the 'HyperEncoding' thing when buying intel cpu and gpu, AMD's bonus is very limited 'smart access memory' where the cpu can access GPU vram directly' but nobody is very impressed with it's performance increases.

Whereas HyperEncoding is using dual hardware encoders simultaneously the IGPU and GPU can simultaneously be used for processing too. They demonstrate a topaz AI upscaling product but it looks to use the same AI engine that Vegas uses, so there's 2 bonus's for having both the Intel CPU and GPU. Have to wait for pricing and performance figures

edit: ARC coming soon rumor is about Arc A370M, a laptop GPU