Best CPU for hqplayer

[eboy]

@Miska I wonder why gpu offload increases at 192khz and 352.8khz pcm2dsd playback?

 

pcm2dsd settings:

 

poly-sinc-ext2 + ASDM7ECv2 + 192/24/2 -> DSD256

poly-sinc-ext2 + ASDM7ECv2 + 352.8/24/2 -> DSD256
 

[Miska]

4 hours ago, eboy said:

@Miska I wonder why gpu offload increases at 192khz and 352.8khz pcm2dsd playback?

 

A bit simplified, higher the source rate, more demanding the filter becomes to process.

 

[eboy]

@Miskathe gpu offload is smaller on ubuntu 22.04 with same settings compared with windows 10

 

poly-sinc-ext2 + ASDM7ECv2 + 192/24/2 -> DSD256

poly-sinc-ext2 + ASDM7ECv2 + 352.8/24/2 -> DSD256

[Miska]

1 hour ago, eboy said:

@Miskathe gpu offload is smaller on ubuntu 22.04 with same settings compared with windows 10

 

It is likely just due to different way to display loads. I doubt there's any notable difference. You should use nvidia-smi on both platforms to compare the figures. It has pretty different load figures than what Windows Task Manager shows.

 

[hpsxrb]

To evaluate the capacity of a CPU for HQPlayer, is there any kind of test that gives us an indicative value and preference that is more common in CPU reviews?
If so, which one or which ones?
Thank you very much.

[Miska]

For benchmarking purposes, HQPlayer has null output feature and reports processing time of the run. But this is not very commonly used though...

 

[semente]

What would be the best option for HQPlayer: a 2013 12-core Xeon or a 2017 6-core i7?

[Miska]

2 hours ago, semente said:

What would be the best option for HQPlayer: a 2013 12-core Xeon or a 2017 6-core i7?

 

Does the Xeon in question have AVX2 support? If not, it is ruled out immediately.

 

[semente]

21 minutes ago, Miska said:

 

Does the Xeon in question have AVX2 support? If not, it is ruled out immediately.

 

The specs sheet mentions AVX-512, not sure if that meets the requirements:

 

https://www.intel.co.uk/content/www/uk/en/products/sku/193749/intel-xeon-w3235-processor-19-25m-cache-3-30-ghz/specifications.html

[Miska]

1 hour ago, semente said:

The specs sheet mentions AVX-512, not sure if that meets the requirements:

 

https://www.intel.co.uk/content/www/uk/en/products/sku/193749/intel-xeon-w3235-processor-19-25m-cache-3-30-ghz/specifications.html

 

That Xeon model is from 2019, not from 2013. If something has AVX-512, it also has AVX2.

 

[semente]

34 minutes ago, Miska said:

 

That Xeon model is from 2019, not from 2013. If something has AVX-512, it also has AVX2.

 

Oops, looks like I have mixed the Xeon model. 😂

[jacobacci]

I am not looking for the most powerful CPU. My requirements are as follows:

- Computer needs to output to a Weiss INT202 via Firewire. This means a TI Chipset based Firewire card and an Intel motherboard running Windows

- Computer silent, preferably passive cooling

- The attached DAC is a dCS Delius, capable of 24/192kHz PCM, no DSD

The goal is to convert everything (including DSD up to DSD512) to 192kHz PCM, basically replacing the dCS Purcell upsampler. So no demanding upsampling to DSD512 this time. Should I ever venture into upsampling to DSD, I would limit it to DSD256 and not the most demanding filters.

From the Weiss INT202 connection to the DAC is via double AES.

 

My question whether there is a sweetspot for CPU for this usage scenario?

Thanks for any suggestions.

 

[Miska]

4 hours ago, jacobacci said:

I am not looking for the most powerful CPU. My requirements are as follows:

- Computer needs to output to a Weiss INT202 via Firewire. This means a TI Chipset based Firewire card and an Intel motherboard running Windows

 

If I guess the chip they use correctly, it could also work under Linux, for example with HQPlayer OS or NAA OS. Or Ubuntu, at least with my custom kernel.

 

4 hours ago, jacobacci said:

- Computer silent, preferably passive cooling

- The attached DAC is a dCS Delius, capable of 24/192kHz PCM, no DSD

The goal is to convert everything (including DSD up to DSD512) to 192kHz PCM, basically replacing the dCS Purcell upsampler. So no demanding upsampling to DSD512 this time. Should I ever venture into upsampling to DSD, I would limit it to DSD256 and not the most demanding filters.

From the Weiss INT202 connection to the DAC is via double AES.

 

My question whether there is a sweetspot for CPU for this usage scenario?

Thanks for any suggestions.

 

Conversion from DSD512 to 192k PCM at best quality is pretty heavy too. For example my Ryzen 7 5800X cannot do it with best quality settings. So I would look for something as powerful as possible that you can still passive cool.

 

[eboy]

 

@Miska I tested Hqplayer with Ryzen 5 5500U (laptop) on Windows 11 and Ubuntu 22.04.

 

Windows 11: Hqplayer 4.19.3 (avx2) (ASIO)
Ubuntu 22.04: Hqplayer 4.19.3-96avx2 (ALSA)

 

poly-sinc-ext2 + ASDM7ECv2 + 192/24/2 -> DSD256

 

Playback in Windows 11 did not work, the sound was choppy and glitchy. Playback in Ubuntu 22.04 worked just fine.

 

The laptop dont have a Nvidia GPU so no GPU offload.

 

Average CPU Windows 11 (typeperf \Processor(_Total)\% Processor Time): 38-45%
Average CPU Ubuntu 22.04 (top): 42-44%

 

Is Windows 11 more demanding than Ubuntu 22.04 or is Ryzen working better on Linux?
 

[MJ1409]

Some observations after installing a 13900k into my existing rig. I’m still tweaking and playing around…

 

One good aspect is that with a bit of undervolting, temps top out around 70c with 192/24 using ASDM7ECv2 @ 1024x48 (with cuda offload).

 

In saying this though, there is something wired going on with the cpu load allocation and there are still dropouts. For some weird reason a random e-core is being straddled with a high % of the load???

I did notice this line in the log…. Not sure if it is meant to say performance=8??

 

This testing was undertaken with the following:

• Ubuntu Server 22.04 LTS (5.15.0-52-generic)
• hqplayerd_4.33.0-145avx2_amd64.deb
• Tried with multicore auto & 1

As next steps I will try with Jussi’s custom kernel to see if it will make a difference…. 
 

any other suggestions are welcome….

 

 

[El Guapo]

8 minutes ago, MJ1409 said:

any other suggestions are welcome….

This syndrome looks pretty similar to the first time I installed 12900k. I guess the 5.15 kernel didn’t recognize the CPU microcode / architecture. Best way is to turn off all e-core and wait for newer kernel version.

[StreamFidelity]

5 hours ago, MJ1409 said:

For some weird reason a random e-core is being straddled with a high % of the load???

 

The same with me (HQPlayer 4 Embedded 4.32.5, i9-13900KF without CUDA). 

 

I have only activated two E-cores (17,18) and the problem is solved for now.

[Miska]

On 10/27/2022 at 11:02 AM, eboy said:

Is Windows 11 more demanding than Ubuntu 22.04 or is Ryzen working better on Linux?

 

Windows is more demanding, and so far Ryzen has been better on Linux overall. Windows has had some issues scheduling efficiently on Ryzen CCX layout.

 

P.S. Especially with AMD CPUs, I recommend using my custom kernel for better performance. At least for now.

[Miska]

On 10/27/2022 at 12:34 PM, El Guapo said:

This syndrome looks pretty similar to the first time I installed 12900k. I guess the 5.15 kernel didn’t recognize the CPU microcode / architecture. Best way is to turn off all e-core and wait for newer kernel version.

 

Likely yes. @StreamFidelity @MJ1409 How does output of "ls /sys/devices" look like for the 13900K?

 

It is also possible that support is there, but someone decided to change how things are presented...

[StreamFidelity]

19 hours ago, Miska said:

How does output of "ls /sys/devices" look like for the 13900K?

This is wat I got:

 

root@hqplayer:~# ls /sys/devices
LNXSYSTM:00  cpu          intel_pt     msr          platform     software     tracepoint   virtual
breakpoint   intel_bts    kprobe       pci0000:00   pnp0         system       uprobe

 

[Miska]

3 hours ago, StreamFidelity said:

This is wat I got:

 

root@hqplayer:~# ls /sys/devices
LNXSYSTM:00  cpu          intel_pt     msr          platform     software     tracepoint   virtual
breakpoint   intel_bts    kprobe       pci0000:00   pnp0         system       uprobe

 

 

OK, looks like there have been at least some changes. It may also need some upcoming kernel updates as well.

 

I will need to get a 13th Gen CPU and look into this. I just don't have the money right now, so it will have to wait a little.

 

[chipvn]

@jacobacci I have just completed upgrading my modified passive case HDPlex H3V3 HQPlayer Server from running Z390M / i7-9700 (TDP 65W) to Z690M / i7-12700K (TDP 125W). I installed Ubutu Desktop 22.04 LTS with 5.15.0-52-lowlatency Kernel. HQPlayer is HQPlayer Desktop 4.19.3 avx2 version.

 

With the i7-9700K, I could only barely upsample everything to DSD512 with ASDM7ECv2 and filter combo gauss-long / gauss-hires-lp. However, I need to let all core running at 4.4GHz and CPU temperature reached 81oC after about 1hr (with heavily undervolt and PL1 68W / PL2 89W)

 

With i7-12700K, after set P-Core to max at 3.9GHz, undervolt, PL1 62W / PL2 69W, surprisingly, I could do all to DSD512 with ASDM7EC2v2 and even heavier filter combo gauss-xla / gauss-long continuously up 2-2.5hrs but CPU temperature was only max at 68oC (average just about 63-65oC).

 

@MiskaI could not imagine this result at all: DSD512/ASDM7ECv2/poly-sinc-gaus-xla with max CPU clock speed of just 3.9GHz at just 62W PL1 / 69W PL2 and still can control CPU temperature below 68oC on a small passive case like the HDPlex H3V3. Thanks for the great work.

 

[chipvn]

Unbelievable, even max CPU clock speed (P-Core) set at 3.8GHz / PL1 60W / PL2 66W, the i7-12700K is still able to do DSD512/ASDM7ECv2 with gauss-xla for 1x and gauss-long for Nx. CPU max temperature look like could be lower further down to only 66oC in my passive case. 😅 👍

 

By the way, my Mainboard is the MSI MAG Z690M Mortar WIFI DDR5. I use the 2 x 8GB lowest spec DDR5 4800CL40 RAM from Crucial (got them at only $70 under something like stocking clearing promotion)

[sbenyo]

20 hours ago, chipvn said:

With i7-12700K, after set P-Core to max at 3.9GHz, undervolt, PL1 62W / PL2 69W, surprisingly, I could do all to DSD512 with ASDM7EC2v2 and even heavier filter combo gauss-xla / gauss-long continuously up 2-2.5hrs but CPU temperature was only max at 68oC (average just about 63-65oC).

I wonder why my AMD 1920x running at 4.0GHz, with similar specs to the 12700K, is not capable of doing any of the EC modulators at DSD512 (everything works fine at DSD256). I know it's older model but I was hoping it could do the lower ones.

 

12700K:

12 Cores, 20 Threads @3.6GHz, Alder Lake.
Release date: Q4 2021.

 

1920x:

12 Cores, 24 Threads @3.5GHz, Zen.
Release date: Q3 2017.

[chipvn]

I believe there were significant changed in CPU capability between generation especially within the pass 1-2 years for both Intel and AMD. There are many more thing to consider not only number of cores or core's clock speed.