Best CPU for hqplayer

[soundnerdie]

2 minutes ago, dericchan1 said:

I am a noob.

 

can you please advise what these commands are? Do you add them to your config.txt?

 

VDD_CORE_Acurrent(7)=6.76893900A

VDD_CORE_V volt(15)=1.0046900V

EXT5V_V volt(24)=5.14292000V (27W power supply)

Sorry, I didn't explain the numbers. They were generated using the command

vcgencmd pmic_read_adc

These numbers are for reference only.

[roderickvd]

On 2/2/2024 at 6:52 PM, dericchan1 said:

Soooo,

 

oc to 3000hz will not boot up for me

 

oc to 2900hz will boot but not stable, the os froze on me occasionally and hqplayer sometimes just exit on its own…

 

oc to 2800hz seems to be stable but stutters at dsd 256x48 ultra light modulator.

 

Does changing the CPU governor from ondemand to performance or even schedutil improve the stutter? I've had succes with that on lower-power SBCs doing upsampling duties. 

[dericchan1]

Just now, roderickvd said:

 

Does changing the CPU governor from ondemand to performance or even schedutil improve the stutter? I've had succes with that on lower-power SBCs doing upsampling duties. 

Force_turbo=1 helps abit but not much

[roderickvd]

Not the same thing though...

[dericchan1]

1 minute ago, roderickvd said:

Not the same thing though...

I have not tried that

[soundnerdie]

The information in these links appears to say that the force_turbo=1 setting "disables" the CPU governor.

 

https://grimore.org/lakka/overclocking_and_performance_considerations

 

https://forums.raspberrypi.com/viewtopic.php?t=297478

 

 

[dericchan1]

51 minutes ago, soundnerdie said:

The information in these links appears to say that the force_turbo=1 setting "disables" the CPU governor.

 

https://grimore.org/lakka/overclocking_and_performance_considerations

 

https://forums.raspberrypi.com/viewtopic.php?t=297478

 

 

Not sure about that but force_turbo=1 seem to help a bit for me

[roderickvd]

Your screenshot says the governor is ondemand, which is why I was asking. 

[soundnerdie]

1 hour ago, roderickvd said:

Your screenshot says the governor is ondemand, which is why I was asking. 

Well spotted

[Miska]

force_turbo is enough. Governor doesn't really matter since it would keep clocks at max anyway under such load.

 

15 hours ago, roderickvd said:

oc to 2800hz seems to be stable but stutters at dsd 256x48 ultra light modulator

 

Mine boots at 3 GHz, but it occasionally freezes. 2.9 GHz is stable.

 

Keep the output at fixed 44.1k x256?

 

[Mops911]

I apologize if this was asked, but not sure how to search for this topic in this large thread ;-)

 

I run a i9-12900KS without GPU to upsample all to DSD256 with 7EC-super and poly-shrt-mp. This can be done on 14w with plenty of headroom.

 

I know HQPlayer has all sophisticated core allocation, but why disable hyperthreading for HQP for Pcores?  If doing so, it seems the overhead for splitting the work goes down and the 8 threads are not used more than 75%...

 

Why or Why not?

[Miska]

2 hours ago, Mops911 said:

I know HQPlayer has all sophisticated core allocation, but why disable hyperthreading for HQP for Pcores?  If doing so, it seems the overhead for splitting the work goes down and the 8 threads are not used more than 75%...

 

Why or Why not?

 

HyperThreading should be left enabled. It helps both HQPlayer and the OS. Since it halves the total OS context switch overhead. And HQPlayer knows how to efficiently utilize those hardware threads.

 

[Miska]

Some new machine build report... Since my previous development machine with Ryzen 7 5800X started crashing, I had to build a new one... I wanted a small one, since my office is so crowded with equipment. So I went for Mini-ITX, but didn't want to trade any power for that.

 

Case: Fractal Design Define Nano 7

PSU: Seasonic Vertex PX-850 (ATX 3.0 and PCIe 5.0 compliant)

Motherboard: ASRock Z790 PG-ITX/TB4

CPU: Intel i9-14900K

RAM: 64 GB kit of G.Skill Ripjaws S5 DDR5-6400 CL32

CPU cooler: be quiet! Silent Loop 2 280mm AIO

GPU: PNY RTX A4500

 

I had to make some creative modifications to fit AIO fans between front panel and AIO radiator, I used two ASUS ROG Strix 120mm fans. Bigger case would make this easier build regarding AIO.

 

This can do DSD1024 with poly-sinc-gauss-xla + ASDM7EC-super, stock, without overclocking. Power draw from mains is ~280W while doing this.

 

[danielB]

6 minutes ago, Miska said:

Some new machine build report... Since my previous development machine with Ryzen 7 5800X started crashing, I had to build a new one... I wanted a small one, since my office is so crowded with equipment. So I went for Mini-ITX, but didn't want to trade any power for that.

 

Case: Fractal Design Define Nano 7

PSU: Seasonic Vertex PX-850 (ATX 3.0 and PCIe 5.0 compliant)

Motherboard: ASRock Z790 PG-ITX/TB4

CPU: Intel i9-14900K

RAM: 64 GB kit of G.Skill Ripjaws S5 DDR5-6400 CL32

CPU cooler: be quiet! Silent Loop 2 280mm AIO

GPU: PNY RTX A4500

 

I had to make some creative modifications to fit AIO fans between front panel and AIO radiator, I used two ASUS ROG Strix 120mm fans. Bigger case would make this easier build regarding AIO.

 

This can do DSD1024 with poly-sinc-gauss-xla + ASDM7EC-super, stock, without overclocking. Power draw from mains is ~280W while doing this.

 

Hi Miska,

testing your own software is satisfying?:)))

On 7950x3d and rtx 4090,moved physx on cpu,and 3% oc on cpu same filters,cpu 10% @ 5.25-27.

Power draw 260-270w. Can t figure out why it prefers 1st ccd with 128mb cache and 5.25Ghz instead of 2nd ccd with 5.7-5.9 Ghz

 

 

[Miska]

1 minute ago, danielB said:

testing your own software is satisfying?:)))

 

Nice to test new hardware always! But I do so much testing every day, that it can be tiring at times... 😅

 

1 minute ago, danielB said:

On 7950x3d and rtx 4090,moved physx on cpu,and 3% oc on cpu same filters,cpu 10% @ 5.25-27.

Power draw 260-270w. Can t figure out why it prefers 1st ccd with 128mb cache and 5.25Ghz instead of 2nd ccd with 5.7-5.9 Ghz

 

How's the per core loads? IOW, how's the load for modulator cores?

 

[danielB]

3 minutes ago, Miska said:

 

Nice to test new hardware always! But I do so much testing every day, that it can be tiring at times... 😅

 

 

How's the per core loads? IOW, how's the load for modulator cores?

 

Good question.If you let me know how i can do this...,i m only using task manager...,resource monitor?sorry,me kinda noob

[Zauurx]

1 hour ago, Miska said:

Some new machine build report... Since my previous development machine with Ryzen 7 5800X started crashing, I had to build a new one... I wanted a small one, since my office is so crowded with equipment. So I went for Mini-ITX, but didn't want to trade any power for that.

 

Case: Fractal Design Define Nano 7

PSU: Seasonic Vertex PX-850 (ATX 3.0 and PCIe 5.0 compliant)

Motherboard: ASRock Z790 PG-ITX/TB4

CPU: Intel i9-14900K

RAM: 64 GB kit of G.Skill Ripjaws S5 DDR5-6400 CL32

CPU cooler: be quiet! Silent Loop 2 280mm AIO

GPU: PNY RTX A4500

....

 

I'll make a note of the configuration once I've moved and changed home. 

I'll be installing a machine in an isolated area... but with a fanless server, you can already take advantage of DSD512.

 

 

Highest measured interrupt to process latency (µs):   144,70
Average measured interrupt to process latency (µs):   3,278846

Highest measured interrupt to DPC latency (µs):       113,30
Average measured interrupt to DPC latency (µs):       0,680750

 

[firedog]

On 2/2/2024 at 3:43 AM, soundnerdie said:

Nice. May consider this one although a bit overkill

 

https://store.minisforum.com/products/minisforum-ms-01

Hi-

That seems to have a cooling fan on the bottom. Any idea of the db level of the fan?

[Miska]

3 hours ago, danielB said:

Good question.If you let me know how i can do this...,i m only using task manager...,resource monitor?sorry,me kinda noob

 

When you switch to Performance tab, the one that has CPU load graph, right-click on the CPU load graph and select "logical cores". This way you can see per-core loads. Not as detailed as what "htop" under Linux can show. but enough for most cases.

 

Note that the Task Manager shows GPU loads incorrectly when CUDA is in use. For GPU loads, use nvidia-smi from command prompt. I usually use 5 second update interval with "-l 5" argument.

 

[danielB]

21 minutes ago, Miska said:

 

When you switch to Performance tab, the one that has CPU load graph, right-click on the CPU load graph and select "logical cores". This way you can see per-core loads. Not as detailed as what "htop" under Linux can show. but enough for most cases.

 

Note that the Task Manager shows GPU loads incorrectly when CUDA is in use. For GPU loads, use nvidia-smi from command prompt. I usually use 5 second update interval with "-l 5" argument.

 

Looks like is using mostly 6 cores from the 2nd ccd

[danielB]

39 minutes ago, Miska said:

 

When you switch to Performance tab, the one that has CPU load graph, right-click on the CPU load graph and select "logical cores". This way you can see per-core loads. Not as detailed as what "htop" under Linux can show. but enough for most cases.

 

Note that the Task Manager shows GPU loads incorrectly when CUDA is in use. For GPU loads, use nvidia-smi from command prompt. I usually use 5 second update interval with "-l 5" argument.

 

My guess is that is reading  the clock from the first ccd and run same limit on the 2nd

[Miska]

7 hours ago, danielB said:

My guess is that is reading  the clock from the first ccd and run same limit on the 2nd

 

Yeah, it could have different clocks, but the Task Manager is not flexible enought to show the details. Please see earlier posts, there are some other tools to show per-core clocks.

 

By default, HQPlayer attempts to split work so that each channel is on it's own CCD.

 

[LoryWiv]

8 hours ago, Miska said:

Note that the Task Manager shows GPU loads incorrectly when CUDA is in use. For GPU loads, use nvidia-smi from command prompt. I usually use 5 second update interval with "-l 5" argument.

 

For those of us who are command-prompt challenged, is a small utility like HWMonitor more accurate in determining GPU loads while CUDA offload is running?

[Miska]

2 minutes ago, LoryWiv said:

For those of us who are command-prompt challenged, is a small utility like HWMonitor more accurate in determining GPU loads while CUDA offload is running?

 

I don't know... I just trust the official Nvidia utility. It is pretty simple, open command prompt and run the utility. Or open Notepad and copy-paste following:

@echo off
nvidia-smi -l 5

And then File -> Save As. Select "All files" as file filter, as save it as "gpu-load.bat" on your desktop. Now when you double-click on that one, you should get the GPU loads visible...

 

[soundnerdie]

1 hour ago, Miska said:

 

Yeah, it could have different clocks, but the Task Manager is not flexible enought to show the details. Please see earlier posts, there are some other tools to show per-core clocks.

 

By default, HQPlayer attempts to split work so that each channel is on it's own CCD.

 

This tool might help

 

https://www.hwinfo.com/