Best CPU for hqplayer

[MirinStereo]

1 hour ago, Jens_G said:

 Is it a matter of the base clock rate or what influence does the maximum turbo clock rate has here? For example, a (fictive) CPU with base clock rate at 2,8GHz and maximum turbo clock rate at 3,5GHz would fulfill the clock rate requirements or not?

What matters is the clock speed that is sustained on all cores utilized. For example, in the case of your fictive CPU the turbo core clock for all cores at load might be somewhere around 3.1 or 3.2GHz. The 3.5GHz is the turbo speed for a single core under load and thus not really relevalt for the application. To have the processor turbo to the maximum possible speed you need to make sure that it is adequately cooled.

As for the Xeon E5-2687W v3, I think it should work based on the similarity in specs to the 6950X. The 6950X should have higher IPC (newer manufacturing process and architecture) but the clock speed of the Xeon is higher which should compensate for that to an extent. Cache and core count is the same. However, I cannot be sure and this is merely a speculation. If you decide to try it please let the forum know if it works.

[Miska]

1 hour ago, ddetaey said:

In fact I doubt that even there is a CPU available today that can handle DSD512 poly-sinc-xtr (non -2s) all by itself , without CUDA support.

(even not the high ranked 

Intel Core i9-7980XE @ 2.60GHz

27,761

 

The i7-6950X I have can do it.

 

I think someone reported that it works on i9-7980XE. I'd like to hear if it works on other 79xxX models, but I'd assume it does based on 6950X.

 

And I believe someone reported it to work on AMD Threadripper 1950X. Now there's bunch of new second generation Threadrippers and would be nice to hear reports about those. But 2950X at least should be OK, since it is equivalent but newer to 1950X. I would really love to get my hands on 2990WX and see how it performs.

 

[bibo01]

I had a conversation with @Miska in the past about a possible benchmark that has a similar workload to his application.

If we find a benchmark that could be used on a HQPlayer system, people can employ it and we can keep it as reference. It would make life easier for many of us. 

The closest benchmarks in terms of workload similarity are:

POV-Ray benchmark (Miska has used it himself before for benchmarking);
C-ray;

NAMD Molecular Dynamics;

NPB  Fluid Dynamics (from the Linux-Bench suite).

 

Now it would be a question of researching which of the above benchmarks is easily available and usable (probably @seeteeyou is the best guy for this) then keep a log of different CPUs.

What do people think of this proposal?

[MirinStereo]

Benchmarking is an excellent idea. Great to know that there is software with comparable workload. I will happily contribute to the database once a specific benchmark is established.

 

These are some benchmark numbers for POV-Ray I found which include the 6950X and the 1950X which both seem to be capable of -xtr DSD512. As far as I know the 1920X does not work with DSD512 -xtr and hence the table shown should be taken with a grain of salt.

 

[elan120]

22 minutes ago, MirinStereo said:

As far as I know the 1920X does not work with DSD512 -xtr and hence the table shown should be taken with a grain of salt.

 

 

 

[4est]

12 minutes ago, elan120 said:

 

 

That is not playing standard xtr, but xtr-2s.

[elan120]

17 minutes ago, 4est said:

That is not playing standard xtr, but xtr-2s.

 

[bibo01]

1 hour ago, bibo01 said:

I had a conversation with @Miska in the past about a possible benchmark that has a similar workload to his application.

If we find a benchmark that could be used on a HQPlayer system, people can employ it and we can keep it as reference. It would make life easier for many of us. 

The closest benchmarks in terms of work load similarity are:

POV-Ray benchmark (Miska has used it himself before for benchmarking);
C-ray;

NAMD Molecular Dynamics;

NPB  Fluid Dynamics (from the Linux-Bench suite).

 

Now it would be a question of researching which of the above benchmarks is easily available and usable (probably @seeteeyou is the best guy for this) then keep a log of different CPUs.

What do people think of this proposal?

Downloads:
POV-ray http://www.povray.org/download/
C-ray https://openbenchmarking.org/test/pts/c-ray
NAMD Molecular Dynamics https://www.ks.uiuc.edu/Development/Download/download.cgi?PackageName=NAMD

NPB Fluid Dynamics https://www.nas.nasa.gov/publications/npb.html

 

With @Miska's help we can agree on one of them. I do not have direct experience with any of them.

I think that all of them are multi-platform, but I am not so sure about NPB (its page is not so clear). We also have to agree on a particular version, probably the latest.

Furthermore, NAMD has versions which include CUDA acceleration, a nice feature to evaluate in benchmarking too.
 

[rando]

POV-Ray wasn't exactly hard to find or complicated to run.  I'd question how relevant to HQPlayer performance tuning for the highest benchmark is more than how closely it synthesized workload. 

 

@Miska What would prevent adding a benchmark feature to HQPlayer using a built in test file, liability?

[Miska]

8 hours ago, rando said:

 

@Miska What would prevent adding a benchmark feature to HQPlayer using a built in test file, liability?

 

Different settings generate different load patterns, although they have a lot in common too. So I would need to define a work load that would be useful for benchmark.

 

I'll think about this, but likely it doesn't get implemented very soon.

 

[rando]

Ha, by no means was this a request.  Seemed like some discussion had been ongoing per system refinement for HQPlayer.  The natural conclusion would therefore be either an audible or work based internal benchmark.  Was simply asking if it had been considered, what limited your interest or ability to implement it.  

 

 

[n2it]

21 hours ago, ddetaey said:

In fact I doubt that even there is a CPU available today that can handle DSD512 poly-sinc-xtr (non -2s) all by itself , without CUDA support.

 

17 hours ago, MirinStereo said:

As far as I can work out the most important thing is to maximize cache. It is confirmed that the 6950X can do DSD512 poly-sync-xtr and there are some people that state their AMD Threadripper 1950X does the Job as well (others did not confirm).

 

13 hours ago, Miska said:

And I believe someone reported it to work on AMD Threadripper 1950X. Now there's bunch of new second generation Threadrippers and would be nice to hear reports about those. But 2950X at least should be OK, since it is equivalent but newer to 1950X. I would really love to get my hands on 2990WX and see how it performs.

 

 

Just to confirm, the Threadripper 1950x can do DSD512 poly-sinc-xtr.   The only thing I have not been able to do is 48kHz to 44.1x512 DSD.  96kHz to 44.1x512 DSD poly-sinc-xtr takes about 56% CPU non hyperthreaded (or 28% if HT) and the rest that I have tried take a little less.

 

I too would love a 2990WX (or even try a 2950x).  I feel like the 1950x is close on the 48kHz to 44.1x512 DSD poly-sinc-xtr.  But I really don't need one!

 

Maybe the better question is for the thread is what is the "Best GPU for HQplayer".  I was reading in the other HQplayer thread that the new Nvidia RTX2080 can do DSD512 poly-sinc-xtr.  If that is the case why spend the money on an expensive CPU if you really don't need the extra cores?  For certain things (e.g. gaming), I would have a much better system with a Ryzen 2700 with an Nvidia RTX 2080 - and it would have been (much) cheaper than my 1950x and Nvidia 1070ti.   (I didn't get the Threadripper for HQplayer - I got the Threadripper for the large number of PCIe lanes direct from the CPU and to be able to run a number of VMs ... so it was incremental to get the 1950x over the 1920x) 

 

Quite frankly - I can't tell the difference between the non-2s and the 2s.  My Nvidia 1070ti can easily run all the filters with all sample rates in 2s mode.  So it really makes me think about a GPU solution instead if I was building a new computer.

 

[bibo01]

4 hours ago, Miska said:

Different settings generate different load patterns, although they have a lot in common too. So I would need to define a work load that would be useful for benchmark.

 

I'll think about this, but likely it doesn't get implemented very soon.

If, as you say, it's not going to be implemented very soon, wouldn't be OK for you to define for now the most suitable benchmark for HQP among those I listed before and people can start ti run some test/comparison?!

[Miska]

31 minutes ago, bibo01 said:

If, as you say, it's not going to be implemented very soon, wouldn't be OK for you to define for now the most suitable benchmark for HQP among those I listed before and people can start ti run some test/comparison?!

 

You could check if some FP64 heavy benchmark puts CPUs in similar order like discussed here?

 

[Jens_G]

@n2ityou wrote: I was reading in the other HQplayer thread that the new Nvidia RTX2080 can do DSD512 poly-sinc-xtr. 

 

Can you plz give a link on that post? Thx

 

Was it the post of @semeniub? But he statet out there: „I've checked for you, and DSD64 to DSD512 with the poly-sinc-xtr filter works just fine with the RTX 2080 card.“

 

That is not what I‘m looking for. I want to upsample redbook 44.1kHz to DSD512 using poly-sinc-xtr...

[semeniub]

30 minutes ago, Jens_G said:

@n2ityou wrote: I was reading in the other HQplayer thread that the new Nvidia RTX2080 can do DSD512 poly-sinc-xtr. 

 

Can you plz give a link on that post? Thx

 

Was it the post of @semeniub? But he statet out there: „I've checked for you, and DSD64 to DSD512 with the poly-sinc-xtr filter works just fine with the RTX 2080 card.“

 

That is not what I‘m looking for. I want to upsample redbook 44.1kHz to DSD512 using poly-sinc-xtr...

Hello Jens_G,

 

This is what I wrote in another thread. Forgive me, I don't yet have the forum skills for cross-linking to posts in other threads, etc:

 

"I am a new member to CA, but have been a longtime user of HQPlayer.

 

I just wanted to let everyone know that a recent graphics card upgrade to a RTX 2080 has now allowed me (with my modest i7-6700k CPU) to use the full poly-sinc-xtr filter and upsample to DSD512 smoothly. This includes 44/16 files.

 

CPU load is around 65% (Roon + HQPlayer).

GPU load is around 75%.

 

The computer is running Windows 10, and the latest NVIDIA drivers are installed. HQPlayer is version 3.25.

 

I was previously using a GTX 1080 graphics card, and running HQPlayer like this was not even close to possible."

 

So yes, redbook 44.1kHz to DSD512 using poly-sinc-xtr is possible with the RTX 2080 card.

[Jens_G]

@semeniubThank you very much for that information. That gives me indeed some more flexibility options in designing my new Audio-PC ?? 

Would you be so kind to share the exact model and brand of your RTX 2080 with us?

[n2it]

@Jens_G also see these 2:

 

On 10/13/2018 at 12:21 PM, Miska said:

 

Oh yes, if you have 6700K then that is certainly a good option! It is not so different from the 7700K I have. And with the combination of 7700K and RTX2080 I can upsample RedBook to DSD512 using poly-sinc-xtr.

 

And as news, I just tried and first time I can run poly-sinc-xtr from 48k source to 44.1k x512 DSD, so single stage xtr with conversion between rate families! This is notable because number of DACs don't support 48k-base DSD rates.

 

 

On 10/16/2018 at 6:12 AM, eternaloptimist said:

Well, the ASUS DUAL RTX 2080 arrived in the post today and is now installed. Wow, this card is a beast!

I can upsample to DSD512 (to T+A DAC 8 DSD via NAA) with poly-sinc-xtr, poly-sinc-xtr-mp without a problem.  The CUDA offload is impressive. Task manager tells me the CPU (i6700K) is sitting stable at around 35%, the graphics card at 60 - 70%. With less intensive filter, the GPU usage sits at around 5 - 7%.

Am having a lot of fun experimenting! The poly-sinc-xtr-mp is remarkable - very dynamic!

 

 

[gdpr]

7 hours ago, n2it said:

Quite frankly - I can't tell the difference between the non-2s and the 2s.  My Nvidia 1070ti can easily run all the filters with all sample rates in 2s mode.  So it really makes me think about a GPU solution instead if I was building a new computer.

 

 

I am a little bit surprised by your statement that you cannot tell the difference between the non-2s and the 2s.

 

I can play poly-sinc-xtr-2s-512 without a problem, and I can play poly-sinc-xtr-256 (non-2s) as well.

Very difficult to identify exactly why, but I have a slight preference for the poly-sinc-xtr-256 (non-2s) - just a little bit more dynamic ???

I am sure I would not succeed succesfully doing a blind test?

 

Looking from the opposite side : if I am happy/satisfied with poly-sinc-xtr-256 (non-2s), why worry about DSD512?

I know we all want the best, but I am sure that in a couple of months, we will be all over DSD1024 .

 

Still, based on your comment, I am going to switch to poly-sinc-xtr-2s-512 for a week, and then switch back to poly-sinc-xtr-256 (non-2s).

It is only by 'missing' something, that one come up to one's personal 'best'.

 

Dirk

[bibo01]

11 hours ago, Miska said:

 

You could check if some FP64 heavy benchmark puts CPUs in similar order like discussed here?

 

I can check (and I will) which benchmark among POV-ray, C-ray, NAMD Molecular Dynamics and NPB Fluid Dynamics stresses FP64 more.

However, only someone with a range of PC systems (you?) can run a test and establish if it puts CPU in the "right" order for us.

Is it OK if, after some research, I am going to suggest you which of the 4 benchmarks to run? Deal?!

  

[bibo01]

15 hours ago, Miska said:

 

You could check if some FP64 heavy benchmark puts CPUs in similar order like discussed here?

 

 

4 hours ago, bibo01 said:

I can check (and I will) which benchmark among POV-ray, C-ray, NAMD Molecular Dynamics and NPB Fluid Dynamics stresses FP64 more.

However, only someone with a range of PC systems (you?) can run a test and establish if it puts CPU in the "right" order for us.

Is it OK if, after some research, I am going to suggest you which of the 4 benchmarks to run? Deal?!

  

All the 4 benchmarks run some form of floating point calculation. However, only NAMD Molecular Dynamics  https://www.ks.uiuc.edu/Development/Download/download.cgi?PackageName=NAMD

also takes into account CUDA acceleration. POV-ray, for example, doesn't and NPB is more geared towards parallel computation and data movement.  

[Miska]

5 hours ago, bibo01 said:

I can check (and I will) which benchmark among POV-ray, C-ray, NAMD Molecular Dynamics and NPB Fluid Dynamics stresses FP64 more.

However, only someone with a range of PC systems (you?) can run a test and establish if it puts CPU in the "right" order for us.

Is it OK if, after some research, I am going to suggest you which of the 4 benchmarks to run? Deal?!   

 

I was more thinking about utilizing one of the already existing benchmark results. If someone can find a list that already has CPUs in the correct order. IOW, CPUs that are known to be able to do xtr to DSD512 above the ones that are known not to. Other than that, it would be more of a collective effort, because I don't have for example any Threadrippers.

 

NAMD sounds like a potential candidate.

[gdpr]

In this topic, I see the following processors mentioned as being capable to support poly-sinc-xtr-512 (non-2s) without CUDA :

 

AMD Treadripper 1920x - 12 cores

AMD Treadripper 1950x - 16 cores

 

INTEL i7-6950X - 10 cores

INTEL i9-7980XE  - 18 cores

INTEL Xeon E5-2687W v3 - 10 cores

 

Probably I am missing a couple processors, but for now I believe it is reasonable to say you need more than 8 CPU cores to do poly-sinc-xtr-512 (non-2s).

 

Dirk

 

P.S. An AMD Ryzen 1700x is able to to support poly-sinc-xtr-lp-512-2s without CUDA.  It does not support poly-sinc-ext2-512.

AMD Ryzen 2700x does support  poly-sinc-ext2-512 without CUDA.  (It does not support poly-sinc-xtr-lp- 512 (non-2s)) , with only passive cooling in my Pinkfaun 2.16X streamer.

 

 

 

Dirk

 

[bibo01]

17 hours ago, Miska said:

I was more thinking about utilizing one of the already existing benchmark results. If someone can find a list that already has CPUs in the correct order. IOW, CPUs that are known to be able to do xtr to DSD512 above the ones that are known not to. Other than that, it would be more of a collective effort, because I don't have for example any Threadrippers.

 

NAMD sounds like a potential candidate.

I looked at NAMD site extensively, but I was not able to find a list of benchmark results, or at least what I found was referring to the number of nodes in parallel computing, not to a list of CPUs which is what we want.

The alternative would be to start testing ourselves and we compile a list of available CPUs. With just a dozen test users, it would be a rather comprehensive list already.

NAMD downloads (You will need a username and password to download software):

• Linux-x86_64-multicore (64-bit Intel/AMD single node)
• Linux-x86_64-multicore-CUDA (NVIDIA CUDA acceleration)
• MacOSX-x86_64 (Mac OS X for 64-bit Intel processors)
• MacOSX-x86_64-CUDA (NVIDIA CUDA acceleration)
• Win64 (Windows 7, 8, 10, etc.)
• Win64-CUDA (NVIDIA CUDA acceleration)
   

[Ralf11]

will it run on a GPU?