Rendering 2x faster when restricting to P-cores for Alder Lake...

Comments

john_dennis wrote on 9/17/2022, 9:47 AM

@bitman

Did you reboot the machine before each render?

My main system:
Motherboard: Asus X99-AII
CPU: Intel i7-6850K
GPU: Sapphire Radeon RX480-8GB
RAM: Corsair Dominator (4 x 4 GB) DDR4 2400
Disk O/S & Programs: Intel SSD 750 (400 GB)
Disk Active Projects: 1TB & 2TB WD BLACK SN750 NVMe Internal PCI Express 3.0 x4 Solid State Drives
Disk Other: WD Ultrastar/Hitachi Hard Drives: WDBBUR0080BNC-WRSN, HGST HUH728080ALE600, 724040ALE640, HDS3020BLA642
Case: LIAN LI PC-90 Black Aluminum ATX Full Tower Case
CPU cooling: Corsair Hydro series H115i
Power supply: SeaSonic SS-750KM3 750W 80 PLUS GOLD Certified Full Modular Active PFC Power Supply
Drive Bay: Kingwin KF-256-BK 2.5" and 3.5" Trayless Hot Swap Rack with USB 3
Sound card: Crystal Sound 3 on motherboard. Recording done on another system.
Primary Monitor: Asus ProArt PA248q (24" 1920 x 1200)
O/S: Windows 10 Pro 22H2, Build 19045.2130

Camera: Sony RX10 Model IV

https://www.youtube.com/user/thedennischannel

bitman wrote on 9/17/2022, 10:35 AM

 

Did you reboot the machine before each render?

@john_dennis No not needed, it does not make a difference, sometimes I do reboot, most of the time I close Vegas, check if no process is hanging, and start Vegas again. It seems that Vegas 20 (and 19) is quite good at cleaning up their process footprint, this was not always so in older versions.

I have installed a trial version of "Process Lasso" to experiment, but their affinity setting behaves just the same as my own attempts, affinity is also ignored.

APPS: VIDEO: VP 365 (20), VP 19 post (latest build -643), (uninstalled VP 12,13,14,15,16 Suite,17, VP18 post), Vegasaur, a lot of NEWBLUE plugins, Mercalli 6.0, Respeedr, Vasco Da Gamma 15 HDpro XXL, Boris Continuum 2022.5, Davinci Resolve Studio 18, SOUND: RX 10 advanced Audio Editor, Sound Forge Audio Clean Lab 3, Sound Forge Audio Studio 16, Sound Forge Pro 14, Spectral Layers Pro 8, Audacity, FOTO: Zoner, DXO, Luminar, Topaz...

  • OS: Windows 11 Pro 64, version 22H2
  • CPU: i9-12900K with Cooler: Noctua NH-D15s
  • RAM: DDR5 Corsair 64GB (5600-40 Vengeance)
  • Graphics card: ASUS GeForce RTX 3090 TUF OC GAMING (24GB) 
  • Monitor: LG 38 inch ultra-wide (21x9) - Resolution: 3840x1600
  • C-drive: Corsair MP600 PRO XT NVMe SSD 4TB (PCIe Gen. 4)
  • Video drives: Samsung NVMe SSD 2TB (980 pro and 970 EVO plus) each 2TB
  • Mass Data storage & Backup: WD gold 6TB + WD Yellow 4TB
  • MOBO: Gigabyte Z690 AORUS MASTER
  • PSU: Corsair HX1500i, Case: Fractal Design Define 7 (PCGH edition)
  • Misc: Logitech G915, Evoluent Vertical Mouse, shuttlePROv2

 

 

pierre-k wrote on 9/17/2022, 12:33 PM

Considering how expensive this processor is, the results are worrying. This raises more questions.....

1. Are the results the same in win10?

2. What are the results with processors that have more P cores than 8? For example the old 10 core intel and 16 core AMD? Theoretically, their performance should be higher, because then it seems that E cores are unnecessary for rendering.

3. Is it possible to give the same test of playing the timeline benchmark from Vegas? And change in the Gpu I/O tab and turn E cores off and on?

bitman wrote on 9/17/2022, 1:12 PM

I did a quick retest with dynamic RAM preview max = 6531MB (=10%) instead of zero as before

This shaved off a few seconds. I also disabled my mem XMP profile, this is not worth it for the 1s gain I had.

test set: "classic" 4K Sample Project to 1080p (25P) -

60s NVENC: 8 P cores (16 Threads) + 8 E-cores (8 Threads)

61s NVENC: 8 P cores (16 Threads) + 8 E-cores (8 Threads) Vegas Affinity set to P cores

38s NVENC: 8 P cores (16 Threads) + 8 E-cores parked by Gigabyte DRM fix tool

 

59s QSV: 8 P cores (16 Threads) + 8 E-cores (8 Threads)

58s QSV: 8 P cores (16 Threads) + 8 E-cores (8 Threads) Vegas Affinity set to P cores

37s QSV: 8 P cores (16 Threads) + 8 E-cores parked by Gigabyte DRM fix tool

 

59s MC: 8 P cores (16 Threads) + 8 E-cores (8 Threads)

63s MC: 8 P cores (16 Threads) + 8 E-cores (8 Threads) Vegas Affinity set to P cores

44s MC: 8 P cores (16 Threads) + 8 E-cores parked by Gigabyte DRM fix tool

APPS: VIDEO: VP 365 (20), VP 19 post (latest build -643), (uninstalled VP 12,13,14,15,16 Suite,17, VP18 post), Vegasaur, a lot of NEWBLUE plugins, Mercalli 6.0, Respeedr, Vasco Da Gamma 15 HDpro XXL, Boris Continuum 2022.5, Davinci Resolve Studio 18, SOUND: RX 10 advanced Audio Editor, Sound Forge Audio Clean Lab 3, Sound Forge Audio Studio 16, Sound Forge Pro 14, Spectral Layers Pro 8, Audacity, FOTO: Zoner, DXO, Luminar, Topaz...

  • OS: Windows 11 Pro 64, version 22H2
  • CPU: i9-12900K with Cooler: Noctua NH-D15s
  • RAM: DDR5 Corsair 64GB (5600-40 Vengeance)
  • Graphics card: ASUS GeForce RTX 3090 TUF OC GAMING (24GB) 
  • Monitor: LG 38 inch ultra-wide (21x9) - Resolution: 3840x1600
  • C-drive: Corsair MP600 PRO XT NVMe SSD 4TB (PCIe Gen. 4)
  • Video drives: Samsung NVMe SSD 2TB (980 pro and 970 EVO plus) each 2TB
  • Mass Data storage & Backup: WD gold 6TB + WD Yellow 4TB
  • MOBO: Gigabyte Z690 AORUS MASTER
  • PSU: Corsair HX1500i, Case: Fractal Design Define 7 (PCGH edition)
  • Misc: Logitech G915, Evoluent Vertical Mouse, shuttlePROv2

 

 

Howard-Vigorita wrote on 9/17/2022, 1:36 PM

@pierre-k Only troubling if you have a processor whose cores all have the same clock rate independent of their thermals and you cannot implement a work-around. Closest I can come with the 11900k is to assign lower per-core assignment priority to interior cores and go with liquid cooling. Which works fine till I run out of cores and they all get saturated. Running interior cores at a lower clock looks pretty good. Core Temps look just like the utilization charts... if better manual tuning was possible, bumping up the e-cores a little more might further improvement but it looks pretty darn good as it is.

Btw, getting ready to post my results to the Sample Project and noticed my fresh install of Vegas set my DRP to 0. Setting it to my usual 5% of 32gb ram got me slightly better numbers:

Sample Project 24/48 FHD (1080p 25p) render: Nvenc: 0:49; Qsv: 0:48; MC: 0:52

I find the narrow spread between MainConcept and hardware render curious. I have Vegas set to use the Nvidia as gpu and the Intel Xe for decoding. Task Manager shows that the MainConcept render is heavily assisted by the Nvidia 3060 doing substantial vram buffering. Seems different. Will have to look into that.

bitman wrote on 9/17/2022, 1:49 PM

Considering how expensive this processor is, the results are worrying. This raises more questions.....

1. Are the results the same in win10?

2. What are the results with processors that have more P cores than 8? For example the old 10 core intel and 16 core AMD? Theoretically, their performance should be higher, because then it seems that E cores are unnecessary for rendering.

3. Is it possible to give the same test of playing the timeline benchmark from Vegas? And change in the Gpu I/O tab and turn E cores off and on?

@pierre-k

1. I cannot test in windows 10, anyway, the schedular should be better tuned in windows 11 according to MS for the hybrid processors. No improvements will come to 10 I assume. I already found some registry settings in Windows 11 for tuning multimedia, game, sound etc. scheduling. Not sure if Vegas falls in a category or is recognized.

2. The old processor core types are not as powerful as the P-core golden cove architecture, P-cores have several improvements and a different architecture, is a 10 nm x86 core die, higher top speed, more cash etc...

The Idea of having hybrid processor is not that crazy, both in function and marketing:

function: it makes sense to move background threads to "background cores" (aka efficient E-core), and Apple also uses hybrid processors (if I am not mistaken).

marketing: E cores are small in size and you can squeeze a lot in which makes for the uninformed buyer more cores = better, regardless of real performance per core, something AMD likes to play out vs Intel

3. I had like 17.5 frames

Last changed by bitman on 9/17/2022, 1:50 PM, changed a total of 1 times.

APPS: VIDEO: VP 365 (20), VP 19 post (latest build -643), (uninstalled VP 12,13,14,15,16 Suite,17, VP18 post), Vegasaur, a lot of NEWBLUE plugins, Mercalli 6.0, Respeedr, Vasco Da Gamma 15 HDpro XXL, Boris Continuum 2022.5, Davinci Resolve Studio 18, SOUND: RX 10 advanced Audio Editor, Sound Forge Audio Clean Lab 3, Sound Forge Audio Studio 16, Sound Forge Pro 14, Spectral Layers Pro 8, Audacity, FOTO: Zoner, DXO, Luminar, Topaz...

  • OS: Windows 11 Pro 64, version 22H2
  • CPU: i9-12900K with Cooler: Noctua NH-D15s
  • RAM: DDR5 Corsair 64GB (5600-40 Vengeance)
  • Graphics card: ASUS GeForce RTX 3090 TUF OC GAMING (24GB) 
  • Monitor: LG 38 inch ultra-wide (21x9) - Resolution: 3840x1600
  • C-drive: Corsair MP600 PRO XT NVMe SSD 4TB (PCIe Gen. 4)
  • Video drives: Samsung NVMe SSD 2TB (980 pro and 970 EVO plus) each 2TB
  • Mass Data storage & Backup: WD gold 6TB + WD Yellow 4TB
  • MOBO: Gigabyte Z690 AORUS MASTER
  • PSU: Corsair HX1500i, Case: Fractal Design Define 7 (PCGH edition)
  • Misc: Logitech G915, Evoluent Vertical Mouse, shuttlePROv2

 

 

todd-b wrote on 9/17/2022, 9:02 PM

 

 

Sample Project 24/48 FHD (1080p 25p) render: Nvenc: 0:49; Qsv: 0:48; MC: 0:52

I find the narrow spread between MainConcept and hardware render curious.

@Howard-Vigorita Constrained by GPU. You have enough available CPU to make GPU encoding unnecessary at 1080P. Only the last 7 seconds of the project, a transcode, will greatly benefit GPU encoding over CPU encoding. In Resolve as an example, almost all projects I CPU encode, because the bottleneck is always GPU processing. Why use an inferior encoder for no benefit. This won't be true for transcoding.

john_dennis wrote on 10/1/2022, 1:14 AM

Techno-philosophical question of the day.

If one is parking e-cores to improve Vegas Pro performance, why not just buy an i7-12700K?

My main system:
Motherboard: Asus X99-AII
CPU: Intel i7-6850K
GPU: Sapphire Radeon RX480-8GB
RAM: Corsair Dominator (4 x 4 GB) DDR4 2400
Disk O/S & Programs: Intel SSD 750 (400 GB)
Disk Active Projects: 1TB & 2TB WD BLACK SN750 NVMe Internal PCI Express 3.0 x4 Solid State Drives
Disk Other: WD Ultrastar/Hitachi Hard Drives: WDBBUR0080BNC-WRSN, HGST HUH728080ALE600, 724040ALE640, HDS3020BLA642
Case: LIAN LI PC-90 Black Aluminum ATX Full Tower Case
CPU cooling: Corsair Hydro series H115i
Power supply: SeaSonic SS-750KM3 750W 80 PLUS GOLD Certified Full Modular Active PFC Power Supply
Drive Bay: Kingwin KF-256-BK 2.5" and 3.5" Trayless Hot Swap Rack with USB 3
Sound card: Crystal Sound 3 on motherboard. Recording done on another system.
Primary Monitor: Asus ProArt PA248q (24" 1920 x 1200)
O/S: Windows 10 Pro 22H2, Build 19045.2130

Camera: Sony RX10 Model IV

https://www.youtube.com/user/thedennischannel

bitman wrote on 10/1/2022, 2:01 AM

Techno-philosophical question of the day.

If one is parking e-cores to improve Vegas Pro performance, why not just buy an i7-12700K?

@john_dennis Indeed, but Vegas Pro is not the only NLE I use, I also use my PC for photo applications, sound fix and gaming. And talking about Vegas pro, parking my e-cores is a sure way (=guaranteed and reproducible) to almost render 2x faster. And strangely enough sometimes without disabling the e-cores I get the same fast rendering; this happens mostly whenever I run something in the background - such as windows own task manager to monitor CPU core usage (as mentioned before in this post) ...

I could be wrong, but I think the Vegas team should be working together with Intel and Microsoft to get to the bottom of this, it may be "just" a matter of cloaking the render process as something requiring the P (performance) cores rather than shoving some threads to the E-cores at random. It is not like that the P-cores are starved from Vegas, otherwise when running extra programs would slow down the rendering, but it does not, on the contrary, strangely enough it speeds up the rendering.

By the way, last week I submitted an official ticket on this matter at Magix support:

Ticket#2022092117004289

Here some arguments (apart from Vegas Pro!) on i9-12900K vs i7-12700K

13.13% faster CPU speed?
8 x 3.2GHz & 8 x 2.4GHzvs8 x 3.6GHz & 4 x 2.7GHz

4 more CPU threads?
24vs20

2MB bigger L2 cache?
14MBvs12MB

19.94% higher PassMark result?
41317vs34449

5MB bigger L3 cache?
30MBvs25MB

22.5% higher multi-core Geekbench 5 result?
17279vs14105

20.13% higher Cinebench R20 (multi) result?
10511vs8750

APPS: VIDEO: VP 365 (20), VP 19 post (latest build -643), (uninstalled VP 12,13,14,15,16 Suite,17, VP18 post), Vegasaur, a lot of NEWBLUE plugins, Mercalli 6.0, Respeedr, Vasco Da Gamma 15 HDpro XXL, Boris Continuum 2022.5, Davinci Resolve Studio 18, SOUND: RX 10 advanced Audio Editor, Sound Forge Audio Clean Lab 3, Sound Forge Audio Studio 16, Sound Forge Pro 14, Spectral Layers Pro 8, Audacity, FOTO: Zoner, DXO, Luminar, Topaz...

  • OS: Windows 11 Pro 64, version 22H2
  • CPU: i9-12900K with Cooler: Noctua NH-D15s
  • RAM: DDR5 Corsair 64GB (5600-40 Vengeance)
  • Graphics card: ASUS GeForce RTX 3090 TUF OC GAMING (24GB) 
  • Monitor: LG 38 inch ultra-wide (21x9) - Resolution: 3840x1600
  • C-drive: Corsair MP600 PRO XT NVMe SSD 4TB (PCIe Gen. 4)
  • Video drives: Samsung NVMe SSD 2TB (980 pro and 970 EVO plus) each 2TB
  • Mass Data storage & Backup: WD gold 6TB + WD Yellow 4TB
  • MOBO: Gigabyte Z690 AORUS MASTER
  • PSU: Corsair HX1500i, Case: Fractal Design Define 7 (PCGH edition)
  • Misc: Logitech G915, Evoluent Vertical Mouse, shuttlePROv2

 

 

BruceUSA wrote on 10/1/2022, 8:25 PM

I am curious what CPU Multi Core points can you get on Cinebench R23 on your 12900K at stock speed? I just did a rendering 20 pass to complete the test on my 12700K @5.2 all P core, 5.3 @ 6 core and turbo boost 3 cores @5.4. The rendering get me 24,525 points. CPU draw maximum power of 270 watts.

Last changed by BruceUSA on 10/1/2022, 8:49 PM, changed a total of 3 times.

Intel i7 12700k @5.2Ghz all P Cores, 5.3@ 6 Core, Turbo boost 3 Cores @5.4Ghz. 4.1Ghz All E Cores.                                          

MSI Z690 MPG Edge DDR5 Wifi                                                     

TEAMGROUP T-Force Delta RGB 32GB DDR5 -6200                     

Samsung 980 Pro x4 Nvme .M2 1tb Pcie Gen 4                                     

ASRock RX 6900XT Phantom 16GB                                                        

PSU Eva Supernova G2 1300w                                                     

Black Ice GTX 480mm radiator top mount push/pull                    

MCP35X dual pump w/ dual pump housing.                                

Corsair RGB water block. RGB Fan thru out                           

Phanteks Enthoo full tower

Windows 11 Pro

bitman wrote on 10/21/2022, 4:02 AM

It has been a month since I submitted Ticket#2022092117004289 on the issue, I had zero response from magix/vegas customer support. As Vegas claims that their development priority will focus more on stability and performance, I am baffled by their silence on a clear performance issue with the latest Intel hardware (E-core P-core hybrid CPU's).

APPS: VIDEO: VP 365 (20), VP 19 post (latest build -643), (uninstalled VP 12,13,14,15,16 Suite,17, VP18 post), Vegasaur, a lot of NEWBLUE plugins, Mercalli 6.0, Respeedr, Vasco Da Gamma 15 HDpro XXL, Boris Continuum 2022.5, Davinci Resolve Studio 18, SOUND: RX 10 advanced Audio Editor, Sound Forge Audio Clean Lab 3, Sound Forge Audio Studio 16, Sound Forge Pro 14, Spectral Layers Pro 8, Audacity, FOTO: Zoner, DXO, Luminar, Topaz...

  • OS: Windows 11 Pro 64, version 22H2
  • CPU: i9-12900K with Cooler: Noctua NH-D15s
  • RAM: DDR5 Corsair 64GB (5600-40 Vengeance)
  • Graphics card: ASUS GeForce RTX 3090 TUF OC GAMING (24GB) 
  • Monitor: LG 38 inch ultra-wide (21x9) - Resolution: 3840x1600
  • C-drive: Corsair MP600 PRO XT NVMe SSD 4TB (PCIe Gen. 4)
  • Video drives: Samsung NVMe SSD 2TB (980 pro and 970 EVO plus) each 2TB
  • Mass Data storage & Backup: WD gold 6TB + WD Yellow 4TB
  • MOBO: Gigabyte Z690 AORUS MASTER
  • PSU: Corsair HX1500i, Case: Fractal Design Define 7 (PCGH edition)
  • Misc: Logitech G915, Evoluent Vertical Mouse, shuttlePROv2