Best CPU for hqplayer

[elan120]

15 minutes ago, ted_b said:

So no dc-ATX converter, just the direct ATX 500W power supply?  Is this a better solution than the 300W/dc-ATX converter combo for ATX and another psu for the cpu?

I have tried using Keces P8/20V to HDPlex 800W DC-ATX, and from DC-ATX to both EPS and ATX route before settled with two Sean Jacobs DC3 power supplies to power EPS and ATX.  The improvement with this setup is very noticeable.

 

IIRC, I7-6700K will consume up to 110W, which the Paul Hynes SR7EH with rating at 6A won't work.

 

I am planning on doing test and compare the upcoming Taiko DC-ATX module feeding by a unregulated power supply with my current setup when it is available, and there is a strong possibility this setup will again, surpass what I current have.

[Bass Concussion Racing]

Alright, in light of the general "don't go fanless or you'll be restricted by thermals" trend, and considered that I'm not willing to wait for next-gen processors (there'll always be some "next" after all) please let me put it in a shorter, "yay or nay" fashion...

 

Do any of you gentlemen think that a decent dampened case, fairly good PSU and CPU cooling, an i9-10900K and GeForce GTX 1050 Ti graphics for CUDA offloading, Ubuntu machine will be quiet enough not to bother my listening sessions on open-back headphones @ 1 meter / 3 feet along with a couple of lightweight networking programs running at the same time?

 

I'd be fine playing FLAC > DSD256, poly-sinc-ext2, ASDM7 + convo EQ and occasional crossfeed, whatever buffer helps on my S.M.S.L M200 (AK4497EQ) DAC.

 

Thanks you in advance for your help!

[Miska]

15 minutes ago, Bass Concussion Racing said:

i9-10900K and GeForce GTX 1050 Ti graphics for CUDA offloading

 

I don't think this is good combination for CUDA offloading. Likely the CPU just ends up waiting for the GPU. For just display output, this GPU is fine, but then again you could also use the built-in GPU of 10900K.

 

18 minutes ago, Bass Concussion Racing said:

Do any of you gentlemen think that a decent dampened case, fairly good PSU and CPU cooling, an i9-10900K and GeForce GTX 1050 Ti graphics for CUDA offloading, Ubuntu machine will be quiet enough not to bother my listening sessions on open-back headphones @ 1 meter / 3 feet along with a couple of lightweight networking programs running at the same time?

 

Likely it is quiet enough yes, at least similar one is fine for me. In my case using Fractal Design Define case (has factory installed sound proofing). Either Noctua NH-D15S or BeQuiet Dark Rock Pro 4 CPU cooler.

 

[Bass Concussion Racing]

13 minutes ago, Miska said:

I don't think this is good combination for CUDA offloading.

 

Thanks a lot, Miska!

For HQPlayer itself, most of all of course... I'm in love with its sound quality which I could A/B tell apart from any other players even though my stack is no S-grade one! Unblelievable.

 

So you're saying that there'd be no real advantage over the built-in graphics with that... Ha!

I'm no gamer so I don't need fancy graphics, but is there any graphics card in the ~150€ price range you might suggest which would be able to offload EQ and crossfeed, that goes well with any 10900K-similar price/power CPU?

I can't think of anyone better than you for such suggestions... Thank you!

[Miska]

4 minutes ago, Bass Concussion Racing said:

So you're saying that there'd be no real advantage over the built-in graphics with that... Ha!

I'm no gamer so I don't need fancy graphics, but is there any graphics card in the ~150€ price range you might suggest which would be able to offload EQ and crossfeed, that goes well with any 10900K-similar price/power CPU?

 

For CUDA offloading to make sense, you need GPU that is faster than the CPU. So with that CPU, I'd go minimum with something like RTX2080. Preferably RTX3080 or RTX3090.

 

You can only offload EQ and crossfeed if they are convolution based. The cross-feed in HQPlayer matrix is not convolution based and it is so light weight to process that there's no point in offloading it.

 

[Bass Concussion Racing]

2 minutes ago, Miska said:

The cross-feed in HQPlayer matrix is not convolution based

Uh. Silly me... Of course, otherwise it would be shown in covolution, not pipeline setup!  =)

And indeed crossfeed is really lightweight.

 

So you're telling me that I'd need to get a GPU which will be twice as pricier... Yikes!

Forgive my ignorance, but it looks like it makes a lot more sense to up the CPU-game then.

I'm eyeing some AMD Ryzen 9 (3900X and 5900X) which appear to cost just 100$ more but sport far better performance than Intel i9-10900K or X-series i7s.

Would that be a solid choice? Would that mean that I'd have to up the cooling-game as well, or forget about getting a quiet unit?

Thanks a lot once more. and excuse my n00biness again!

[Miska]

6 minutes ago, Bass Concussion Racing said:

Uh. Silly me... Of course, otherwise it would be shown in covolution, not pipeline setup!  =)

And indeed crossfeed is really lightweight.

 

No, convolution dialog is just mutually exclusive way to do simple convolution things. For more complex convolution things you use pipeline setup, and provide the necessary filter .wav files there.

 

8 minutes ago, Bass Concussion Racing said:

So you're telling me that I'd need to get a GPU which will be twice as pricier... Yikes!

 

You don't have to, get started with just the CPU first and see later if you need a GPU.

 

GPUs are very expensive and very hard to get in first place these days.

 

9 minutes ago, Bass Concussion Racing said:

I'm eyeing some AMD Ryzen 9 (3900X and 5900X) which appear to cost just 100$ more but sport far better performance than Intel i9-10900K or X-series i7s.

 

Since you are planning to run Ubuntu, AMD CPUs are fine. But if you do get one, go with 5000-series. For HQPlayer on Windows, AMD is not a good choice. But for Ubuntu there's a special AMD optimized build. I'm using Ryzen 7 5800X.

 

[Bass Concussion Racing]

AMD-optimized Ubuntu on an AMD Ryzen 9 5900X... Sounds just great.

I'll center my choices around that, get a decently dampened case, good CPU and PSU cooling and consider myself a happy camper, I believe.

Thanks a lot once more! You've been extremely helpful, Jus... Er... Miska!

[luisma]

17 hours ago, ted_b said:

As some of you know, I am a longtime HQplayer user

We know, since the 1st alpha or beta release IIRC

 

17 hours ago, ted_b said:

Any opinions or ideas are welcome. 

Can only speak of my own experience.

 

1. Started with a NUC 4 years ago, connected to my DAC (the DAC model for this comments is irrelevant), core i7 NUC, good results, noisy because of the fan onboard. Powered the NUC with Uptone's JS2

2. Added an NAA, cheap PC older generation Intel, fanless celeron, industrial type passive PC. Powered by the JS2, boost in quality.

3. Built a new server, passive HDPlex chassis with Ryzen 5600X, I can do 7EC with no problems on the passive and I'm happy with it, that said I won't do passive again. Connected DAC directly to the server USB 3.2 port, nice improvement over older hardware (mainly I think because of 7EC). Server powered by DCATX HDPlex adapter with a Dell power supply connected.

4. Added the cheap celeron industrial PC as NAA, low EMI, low processing power, powered by the JS2. Improvement over connecting the DAC directly to the PC, don't know if because EMI or just because NAA with its buffering protocol and such and just handling the DAC is better.

 

All I can say is I do prefer the NAA over other configs. Waiting on my PH SR7 to try it with the passive server but it might take a while.

 

The NAA can be purchased with no memory for $80, memory you could get Apacer if you are into that from Ebay for $25 - $32 8Gb and you can do ramroot too, in the BIOS you could configure every single parameter like disabling unused USB ports, Serial connections, onboard network or onboard adapters, Intel Sleep states and such.

 

IME the NAA is so far preferred.

 

Good luck Ted

 

 

[ted_b]

Luisma (and others), thanks so much for the replies and opinions.  I wish I had better luck with NAAs lately.  My Holo Spring (May on the way) KTE just sounds richer and better colors with a Windows 10 ASIO driver (i.e direct from my Hynes-powered Startech card) than it does via my NAA (Lmitche's NUCi7 with either AL or HQP OS).  I might try putting Windows-something on that NAA and try again.  I had a Windows NAA several years ago when I had the Caps V1 stuff (Zuma as my server, and smaller atom NAA) and it sounded fine, but much later built the i7-6700k server (with lots of Larry's help cleaning it up) and went direct, and that sound is still my benchmark.  Go figure!.  I'd LOVE if I could get that sound via an NAA, as it would isolate and simplify.

[dean70]

Will the RTX3060 with 12GB memory be suitable to run CUDA offload Sinc-L DSD256 from PCM sources with i7-10700 CPU?

[Chodi]

So not to blow up this thread but I have ASDM7EC running at DSD128 no problem using a amd 2600x overclocked to 4k. I am enclosing a screenshot. I am using this on a Windows computer with an Ares II dac. It will not do 256 without skipping but this is a lowly cpu. Thinking about getting a Ryzen 7 5800x to try and see if it will do 256. Maybe I'm missing something in all these discussions about problems? Maybe I'm just blessed.

[Rune]

@ChodiI do not think there is any CPU that does DSD256 ASDM7EC, Sinc-L without GPU. Both my i9-9900KS or Ryzen 5950X  only does that combination with GPU offload and my 5950X can only do ADSM7EC under linux.

[4est]

On 4/16/2021 at 7:30 AM, Miska said:

GPUs are very expensive and very hard to get in first place these days.

 

And this is an understatement!

[Chodi]

3 minutes ago, Rune said:

@ChodiI do not think there is any CPU that does DSD256 ASDM7EC, Sinc-L without GPU. Both my i9-9900KS or Ryzen 5950X  only does that combination with GPU offload and my 5950X can only do ADSM7EC under linux.

I understand your point and you are correct regarding sync L. I did not even go near that filter knowing that current cpu's can't handle that in DSD. I used sync M a  more gentle approach. I do have a Nvidia 1060 gpu that allows for Cuda offload. I can do massive upsampling in pcm with any filter including sync L. I was surprised that my little cpu could easily handle DSD 128 using sync M. Going beyond that in DSD is a not possible. Frankly, the pcm sounds so good I may stick with that as I can use any filter combination and modulator in pcm no sweat. 

[jamesg11]

For a reference point ... macmini M1 will handle ~90% of the filters, though I’ve still got a few to try.

 

macmini M1, 16GB pcm24-192 to ASDM7EC 256 48k into RME direct-dsd mode, 100 ms, no convolution -

 

will not do sinc L, polysinc xtr lp, polysinc xtr mp, polysinc hb

will do ext, ext 2, sinc S, sinc M, closed form, fast, 16M, etc ...

 

Miska will no doubt see a causal pattern here?

 

[Bass Concussion Racing]

14 hours ago, jamesg11 said:

macmini M1, 16GB pcm24-192 to ASDM7EC 256 48k into RME direct-dsd mode, 100 ms, no convolution -

That's quite a feat!

What about upsampling to DSD, same sinc-M settings?

Also, how noisy does the Mini get while working that out? Will it bother me while listening to open-backs @ 1 meter / 3 feet?

 

You're really making me consider getting an M1 Mini instead of building a Ryzen 9 5900X-based box now, since I could double or even triple-boot macOS along with Windows and Linux... The price would be more or less the same, incredibly enough when you think it's Apple.

[jamesg11]

9 hours ago, Bass Concussion Racing said:

That's quite a feat!

What about upsampling to DSD, same sinc-M settings?

Also, how noisy does the Mini get while working that out? Will it bother me while listening to open-backs @ 1 meter / 3 feet?

 

You're really making me consider getting an M1 Mini instead of building a Ryzen 9 5900X-based box now, since I could double or even triple-boot macOS along with Windows and Linux... The price would be more or less the same, incredibly enough when you think it's Apple.

Those tests were from pcm to dsf256. I have the mini in another room, but it’s quiet enough & only slightly warm. I can point a SPL meter at it ...

 

Note that I’m yet to try convolution, though It’ll be just 2-channel.

[Rune]

On 4/18/2021 at 2:10 PM, Bass Concussion Racing said:

I could double or even triple-boot macOS along with Windows and Linux...

You cannot boot into windows or Linux on the Mac Mini M1. It is OS X only.

[Account Closed]

15 hours ago, jamesg11 said:

Those tests were from pcm to dsf256. I have the mini in another room, but it’s quiet enough & only slightly warm. I can point a SPL meter at it ...

 

Note that I’m yet to try convolution, though It’ll be just 2-channel.

Thanks for all your testing. I have a Mac Mini M1 on order this AM!

[Bass Concussion Racing]

3 hours ago, Rune said:

You cannot boot into windows or Linux on the Mac Mini M1. It is OS X only.

These stubborn guys at Corellum could, and provide a way to install a custom-kernel Ubuntu.

Can't say about Windows... I don't really care much, honestly. I guess it's just a matter of time before Apple makes it possible.

 

The Mini option is tempting: sure easier to "build" than a custom desktop machine, and probably less in need of maintanance.

Still I wonder how it'd compare to Ryzen 9 5900X.

[Miska]

2 minutes ago, Bass Concussion Racing said:

These stubborn guys at Corellum could, and provide a way to install a custom-kernel Ubuntu.

Can't say about Windows... I don't really care much, honestly. I guess it's just a matter of time before Apple makes it possible.

 

The Mini option is tempting: sure easier to "build" than a custom desktop machine, and probably less in need of maintanance.

Still I wonder how it'd compare to Ryzen 9 5900X.

 

At least Ryzen 7 5800X is more than 2x faster per CPU core... And this comparing to the high performance cores of M1.

 

[Bass Concussion Racing]

4 minutes ago, Miska said:

At least Ryzen 7 5800X is more than 2x faster per CPU core... And this comparing to the high performance cores of M1.

Whoa.

And I guess a 9 would be even faster... Even a 3900X.

It looks like they have a very different approach, the M1 focusing on single-thread performance, but yeah... I was just being lazy and dreaming of a ready-made machine I could mess with without breaking a sweat.

Off I go embrark on a custom build! I'm sure it'll also be more rewarding in the end.

Thanks a lot once more for the help, Miska! And for oh yes HQPlayer!

[Miska]

2 minutes ago, Bass Concussion Racing said:

It looks like they have a very different approach, the M1 focusing on single-thread performance, but yeah... I was just being lazy and dreaming of a ready-made machine I could mess with without breaking a sweat.

 

I wouldn't say so. It takes two cores to do same thing on M1 than 5800X does with one core. And load of the single 5800X core is lower than either of the two M1 cores.

 

What M1 excels at is multi-core performance - it has much much lower cross-core communication penalty than Intel or AMD CPUs. If Intel or AMD CPU would have similar low penalty, we could be doing DSD512 with EC modulators...

 

[jamesg11]

Miska, re M1 - here’s a go figure result & maybe something to sort?

 

Noted before that M1 won’t handle dsf256 files (& oddly, if I attempt this, HQP just hangs & I need to ‘force quit’ ... I can’t just hit stop, or anything, & go to another album).

 

But, I stumbled on my one & only dff256 album & it plays flawlessly at ASDM7EC 256 wide 48k.

 

(Is it trivial to convert dsf to dff?)