Jump to content
IGNORED

Best CPU for hqplayer


sbenyo

Recommended Posts

8 minutes ago, bobflood said:

I would not do anything at all until later this year when Apple brings out the M1X Mini and Intel Generation 11 CPUs become available. Reading the specs on these gives me hope that they might each be game changers for HQP use.

 

There will always be something new coming. New Ryzen chips are coming too...

 

8 minutes ago, bobflood said:

BTW, I have Roon on my old machine (i5 8400) which I have upgraded to Win 11/64 under the Beta Insiders Program and it is amazingly stable. I use my M1 for HQP and can do ASDM7EC/ Poly Sinc long Gauss/48K Rate Family and that is all I need so I am happy.

 

 I have problems when I run Roon and HQP on the same machine. not separate machines...

No electron left behind.

Link to comment
41 minutes ago, bobflood said:

I would not do anything at all until later this year when Apple brings out the M1X Mini and Intel Generation 11 CPUs become available. Reading the specs on these gives me hope that they might each be game changers for HQP use.

 

BTW, I have Roon on my old machine (i5 8400) which I have upgraded to Win 11/64 under the Beta Insiders Program and it is amazingly stable. I use my M1 for HQP and can do ASDM7EC/ Poly Sinc long Gauss/48K Rate Family and that is all I need so I am happy.

Edit: I meant to say 12th generation Intel this fall or winter, not 11th which is out now. And, I forgot to add that I can do ASDM7EC to DSD 256 on the M1 and this is from any PCM from Tidal or Qobuz. I have no local library so I have no DSD content.

Link to comment

im running dsd256 asdm7ec, gauss-long on a 9900k. stock bios settings for a long time but now ive been experimenting with underclocking/undervolting. build is w/ hdplex and passive heatsink

 

stock it runs 4.7ghz boost at ~1.35v, load temps 60~65c

 

i dropped the clock down to 4.2ghz at 1.1v, load temps dropped >10c and still runs dsd256 7ec great, actually i think it even sounds a bit better now.

 

anyone tried similar ?

Link to comment

Gauss-long is not really CPU demanding filter even when used to up sampling cross family to DSD256 with EC modulator.

I have similar passive HDPlex setup with lower spec i7-9700. I also underclock my CPU by set CPU to max 4GHz and offset CPU voltage by -0.06V 

Upsampling from 16/44 to DSD256x44.1 poly-sinc-gauss-xla EC5 will only have max CPU temp at about 62 deg C (everage is about 55-56).

Doing from 24/48 to DSD256x44.1 poly-sinc-gauss-xla EC5 will be a bit heavier and cause max cpu temp to reach about 66 deg C (about 60-62 average).

RoonROCK: NUC8i5 Akasa case > HQPlayer OS in HDPlex: H3V3 case - i7-12700K > Oratek TOFU Raspberry Pi CM4  - Jussi's NAA OS > Holo Spring 3 L2 DAC with Preamp > DIY EL84PP amp > Snell Type E/III speakers

Ubiquiti EdgerouterX SFP / Sonore Optical Module / Linear Power Supply for all

Link to comment

Intel Alder Lake with E-Cores and P-Cores should provide good performance with the Intel Thread Director. Intel names Windows 11 as the first optimized operating system.

4-1080.3bc2162c.jpg

 

14-1080.1c9c735a.jpg

Source: https://www.computerbase.de/2021-08/hot-chips-33-intel-alder-lake-steht-und-faellt-mit-dem-thread-director/

 

@MiskaIs there a customized HQPlayer for this, e.g. for multicore DSP?

Link to comment

Here's interesting read on Alder Lake architecture:

https://www.anandtech.com/show/16881/a-deep-dive-into-intels-alder-lake-microarchitectures

 

I'm most curious about cross-core communication latency and bandwidth, if they have improved it to level of Apple's M1, it would mean one could run EC modulators at DSD512 rates due to work split. I'm just not very optimistic this would be the case.

 

But other interesting thing is the thread director that loads real cores first before going to SMT (like HQPlayer internally already does):

https://www.anandtech.com/show/16881/a-deep-dive-into-intels-alder-lake-microarchitectures/2

 

Big question is how much OS ends up fighting back. Like Windows' tendency to bundle high load threads on a single core and move threads a lot between cores (increases latency). While macOS behavior is total opposite.

 

Understandable but a hit in itself is loss of AVX-512:

https://www.anandtech.com/show/16881/a-deep-dive-into-intels-alder-lake-microarchitectures/5

On the positive side it means that AVX2 will land on Atom and making it possibly feasible to do EC modulators on Atom.

 

Signalyst - Developer of HQPlayer

Pulse & Fidelity - Software Defined Amplifiers

Link to comment
On 8/24/2021 at 8:20 PM, Miska said:

I'm most curious about cross-core communication latency and bandwidth, if they have improved it to level of Apple's M1, it would mean one could run EC modulators at DSD512 rates due to work split. I'm just not very optimistic this would be the case.

Interesting article, apparently the bandwidth will be improved but latency might be higher too

 

On 8/24/2021 at 8:20 PM, Miska said:

But other interesting thing is the thread director that loads real cores first before going to SMT (like HQPlayer internally already does)

That sounds promising, actual cores doing the job instead of creating SMT chatter, I didn't really get the Threadripper comment, does AMD already doing this?

 

On 8/24/2021 at 8:20 PM, Miska said:

Big question is how much OS ends up fighting back. Like Windows' tendency to bundle high load threads on a single core and move threads a lot between cores (increases latency)

It is clear it will take some time for new Linux kernels to interact with the task scheduler, now I don't really know nor I would ask you what you do on your code but if you are scheduling yourself (which I'm almost sure you do) you have the 30 micro second windows to distribute  loads which of course polling and moving tasks on cores that often creates overhead as well, this will be part of the embedded controller

 

On 8/24/2021 at 8:20 PM, Miska said:

Understandable but a hit in itself is loss of AVX-512:

https://www.anandtech.com/show/16881/a-deep-dive-into-intels-alder-lake-microarchitectures/5

On the positive side it means that AVX2 will land on Atom and making it possibly feasible to do EC modulators on Atom.

I guessed the operations you do during the conversion process (filters, modulators and noise shapers) be largely served by AVX2, unless you were experimenting with 512 as well (split between your Intel / AMD code?).

 

all in all it is good news, new "real" architectural change from Intel, for once I'm glad to see them moving, also the silicon shortage should not affect "much" Intel's production meaning availability will be better.

 

thanks for posting, I love Anand (and sometimes Ian although his writing can get boring sometimes)

Link to comment
3 hours ago, luisma said:

It is clear it will take some time for new Linux kernels to interact with the task scheduler, now I don't really know nor I would ask you what you do on your code but if you are scheduling yourself (which I'm almost sure you do) you have the 30 micro second windows to distribute  loads which of course polling and moving tasks on cores that often creates overhead as well, this will be part of the embedded controller

 

HQPlayer pins certain types of workloads to certain CPU cores. As long as the OS is not doing something stupid to mess up the workload distribution, things are fine. You certainly don't want to have an EC modulator randomly thrown to a low performance core as it would ruin the realtime processing deadlines. Sometimes OS schedulers think that high load threads are just some bulk processing where the execution timelines are not important and for example decide to put all of them on threads of a single core (Windows).

 

3 hours ago, luisma said:

I guessed the operations you do during the conversion process (filters, modulators and noise shapers) be largely served by AVX2, unless you were experimenting with 512 as well (split between your Intel / AMD code?).

 

On the Intel code, there's support for AVX-512 as well (and also SSE4.2 for old CPUs). However, many Intel CPUs limit clock frequencies when AVX-512 is in use, so it doesn't have quite as much benefit as it could. But if Atom line also gains support for AVX2, then I can generally drop support for anything else and maybe go for a single build. On Windows, the AMD optimized build is a bit difficult due to various technical reasons. While on Apple it is not applicable in first place and in future it may be possible to move to a single build there too, if Apple drops Intel CPUs out and moves completely on their own CPUs.

 

At the moment, making a release takes a lot of time and effort just due to number of builds.

 

Signalyst - Developer of HQPlayer

Pulse & Fidelity - Software Defined Amplifiers

Link to comment
  • 2 weeks later...
  • 2 weeks later...

I am about to move my i7 9700 cpu to HQ Player duties (no GPU). Any suggestions on how to set up? I have overclocked in the past, should I consider? I have not made my mind up on cooling yet, either a noctua fan or water, but noise won't be an issue as the machine is away from the listening room.

Link to comment
On 8/15/2021 at 9:47 PM, Miska said:

For Embedded, latest high clock rate Intel and AMD CPUs are good, using the respective builds. 


I‘m using an i9-9900k, which has its base  lock frequency at 3.6Ghz, with the standard overclocking setting in the BIOS bringing it up to 5.0GHz. (I run HQPlayer Embedded, booting the latest image with HQPlayer OS from an USB stick, and am still amazed about how much better it sounds than HQPlayer Desktop on Windows 10Pro with exactly the same settings…)

 

I don‘t have any performance issues with DSD256 and 7EC. If I see correctly, of all the CPUs available up to now the i7-7740X had the highest base clock at 4.30GHz, the i9-10900K can be brought up to a max clock of 5.30GHz, and the current i5-11600k is at a base clock rate of 3.9GHz while the i9-11900K can be brought up to 5.2GHz.  @Miska, is there any benefit from using one of these over my i9-9900K, either theoretical or in terms of achievable soundquality, when using HQPlayer Embedded?

Link to comment
1 hour ago, Bertel said:

I run HQPlayer Embedded, booting the latest image with HQPlayer OS from an USB stick

I think I may have misunderstood before. I currently have HQ Player Desktop, but with embedded you actually don't need to install it. Is that right? So I could have a pc with linux or windows, but I can boot from the memory stick with embedded on it irrespective of the installed OS? Thanks.

Link to comment
On 9/20/2021 at 10:07 AM, Bertel said:

I don‘t have any performance issues with DSD256 and 7EC. If I see correctly, of all the CPUs available up to now the i7-7740X had the highest base clock at 4.30GHz, the i9-10900K can be brought up to a max clock of 5.30GHz, and the current i5-11600k is at a base clock rate of 3.9GHz while the i9-11900K can be brought up to 5.2GHz.  @Miska, is there any benefit from using one of these over my i9-9900K, either theoretical or in terms of achievable soundquality, when using HQPlayer Embedded?

 

Only real advantage of the 10900K over 9900K is two more CPU cores, but I'd say in practice with HQPlayer use difference between the two is small. I don't have any 11900K's, but number of cores is back down to 8, but OTOH, you gain support for AVX-512. Since I have not tested, I don't know how 11900K fares. And I may skip over it and go directly to 12th gen which seems to have bigger architectural changes.

 

Signalyst - Developer of HQPlayer

Pulse & Fidelity - Software Defined Amplifiers

Link to comment
1 hour ago, Miska said:

And I may skip over it and go directly to 12th gen which seems to have bigger architectural changes.

 

I do it like this, too. The next audio PC has i9-12900K on board. I am particularly interested in the Intel Thread Director Windows 11. But AVX512 is not included. 

Link to comment
13 minutes ago, StreamFidelity said:

But AVX512 is not included. 

 

If someone has 11th gen CPU, I'd be interested to hear whether AVX-512 affects clock speeds. So far in the past, like also my Xeon W-2245 CPU, using AVX-512 has meant that CPU is locked to the base clock speed, which has limited the performance boost you'd otherwise get from AVX-512 - since Turbo Boost has fallen off. If you can get both Turbo clocks and AVX-512 simultaneously, it would have a pretty huge performance benefit.

 

I'm just not horribly eager to get 11th gen CPU just to test that one out...

 

So far 11th gen has been the only regular Core series with AVX-512, which has otherwise been feature of bigger Xeon CPUs which don't have a built-in GPU. It likely takes up quite a bit of space on the chip and produces plenty of heat which is why it may not be on 12th gen again that has also new bigger GPU. So I'd guess it's a tradeoff between having more capable on-chip GPU (or GPU at all) vs having AVX-512.

 

Signalyst - Developer of HQPlayer

Pulse & Fidelity - Software Defined Amplifiers

Link to comment
40 minutes ago, Bertel said:

Hmm, I had read that on Z590 mainboards the high clock speed was kept with AVX-512 (e.g. here: https://www.hwcooling.net/en/intel-avx-512-tested-in-x265-how-to-enable-it-and-does-it-help/), but haven‘t tried myself yet…

 

It shouldn't depend on the mainboard / chipset (of course they could still be overclocking), only on the CPU model and it's AVX-512 implementation and it's impact on the thermal envelope. That's why I'm curious if 11th gen Core implementation of AVX-512 manages to support it without limiting clock rates. Theoretically it should give almost 2x boost on many DSP operations.

 

Based on that page, it will certainly test your cooling, but if cooling is not limiting, you could get some performance gain with HQPlayer. Certainly there is still some gain even on my Xeon W-2245 despite clock rate limitations.

 

However, it is important to note what is said there:

Quote

The slower version of AVX-512 in mainstream processors does not actually differ that much from the full-speed version in Intel server CPUs. Server CPUs have an extra 512-bit FMA unit, so they can execute FMA instructions at twice the performance of AVX2, while the client version uses only the existing 256-bit FMA units (which already regular Skylake processors have) and has half the performance in these operations.

But this difference probably only applies to floating-point FMA operations, which are important, for example, for scientific calculations.

 

This specifically applies to HQPlayer where there's heavy reliance on those FMA units. Unlike the video encoding case the article is about. So based on above, you are unlikely to see notable performance gains with AVX-512 on 11th gen Core CPUs.

 

Signalyst - Developer of HQPlayer

Pulse & Fidelity - Software Defined Amplifiers

Link to comment
5 minutes ago, kelvinwsy said:

Hi I am on Gigabyte Z590 Aorus Ultra with the i11990k. What does it take to do this test..I see in the  Bios  av512 default is Off ??

Presently I am on 4.2 ghz 1st 4 coresvabd 4.1 the remainder

 

If you enable the AVX-512 feature, generic HQPlayer builds will use it. You can compare HQPlayer performance with/without it and also check out what clock speeds you have while doing so.

 

Signalyst - Developer of HQPlayer

Pulse & Fidelity - Software Defined Amplifiers

Link to comment

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now



×
×
  • Create New...