Best CPU for hqplayer

[Yviena]

Hmm in my case with  HQplayer 4.8.1 Multicore DSP ticked on, off, or grayed my 5900x has problems with lots of drop-outs with EC7 DSD256, seems i still need to disable corepinning, and manually assign threads/cores.

[Miska]

Intel has just announced bunch of new CPUs. Which look interesting, like the H-series, which should offer nice performance with lower TDP.

 

Desktop CPUs feature 5.3 GHz turbo and 4.8 GHz all-core turbo clocks, support for DDR4-3200 RAM and 20 PCIe 4.0 lanes (4 lanes more than before). Likely the claimed 19% IPC performance improvement will also show up in HQPlayer cases.

 

https://hothardware.com/news/intel-announces-four-new-processor-families-at-ces2021

 

[Miska]

I got the new Ryzen 7 5800X machine up and running after changing the motherboard. So now it has

• Ryzen 7 5800X CPU
• ASUS ROG STRIX B550-E GAMING motherboard
• G.Skill Trident Z Neo DDR4-3800 CL16 RAM

G.Skill would also have DDR4-3800 CL14 RAM, but it wasn't available.

 

ASDM7EC just about works:

[StreamFidelity]

47 minutes ago, Miska said:

ASDM7EC just about works:

 

What does that mean? Are there any dropouts? Purely from the technical specs, it should work:

CPU cores 8
Threads 16
Base Clock 3.8GHz
Max Boost Clock Up to 4.7GHz

[mikel]

I am wondering.....do folks here percieve an audible difference even between DSD128 and DSD256 using the prefferred ASDM7EC?

What I mean by that is blind testing.

Or is the interest more around solving the technical challenge and push current boundaries?

Since I can only do DSD128 I do not really have a reference point or way to compare.

 

[Rune]

1 hour ago, Miska said:

ASDM7EC just about works:

That was also my findings with 5950X, But only with Embedded.

Windows Desktop version was not 100% stable with ASDM7EC DSD256

[mikel]

To add to above post... my goal is also to have as little as possible energy consumption.

I dont like the idea of they power hungry systems.

So if DSD 128 is "good enough" I am happy.

 

[Rune]

36 minutes ago, mikel said:

So if DSD 128 is "good enough" I am happy.

 

If you do not require DSD256EC Then you have more choices and less problems.

Then you could use a 65W TDP CPU. Think I read somewhere that Ryzen 5800 non X will be annonced very soon. Or maybe one of the new AMD APU with zen 3 cores.

Or a Mac Mini with M1 CPU

[Miska]

10 hours ago, StreamFidelity said:

What does that mean? Are there any dropouts? Purely from the technical specs, it should work:

 

No dropouts, but the two cores are loaded to about 96%, so pretty high without too much margin.

 

9 hours ago, Rune said:

That was also my findings with 5950X, But only with Embedded.

Windows Desktop version was not 100% stable with ASDM7EC DSD256

 

This is HQPlayer Desktop on Ubuntu 20.04. One thing to note is that in order to make most of the motherboard hardware work, I needed to use the Canonical's "hardware enablement" kernel. So in other words install "linux-lowlatency-hwe-20.04" meta-package. This brings in their latest short-term kernel, which is currently version 5.8.

 

Also note that you need very latest motherboard BIOS that you can flash to certain motherboard models from USB memory stick without CPU or RAM installed. Because the motherboards have been shipped from factory with a BIOS that doesn't understand about 5000-series CPUs.

 

8 hours ago, Rune said:

Or a Mac Mini with M1 CPU

 

That already gives DSD256... ;)

 

[Yviena]

1 hour ago, Miska said:

 

No dropouts, but the two cores are loaded to about 96%, so pretty high without too much margin.

 

 

This is HQPlayer Desktop on Ubuntu 20.04. One thing to note is that in order to make most of the motherboard hardware work, I needed to use the Canonical's "hardware enablement" kernel. So in other words install "linux-lowlatency-hwe-20.04" meta-package. This brings in their latest short-term kernel, which is currently version 5.8.

 

Also note that you need very latest motherboard BIOS that you can flash to certain motherboard models from USB memory stick without CPU or RAM installed. Because the motherboards have been shipped from factory with a BIOS that doesn't understand about 5000-series CPUs.

 

 

That already gives DSD256... ;)

 

 

Is there actually any reason why Intel seems to have a lower load with EC7 256, benchmarks seem to indicate that IPC/performance is actually higher around 5%ish with the new 5xxx series.

[luisma]

17 hours ago, Miska said:

ASDM7EC just about works:

Thanks for posting, I was expecting more, means the 3950X or 3900X won't do, kind of disappointing, Intel still king

 

[luisma]

16 hours ago, StreamFidelity said:

What does that mean? Are there any dropouts? Purely from the technical specs, it should work

Hi @StreamFidelityIt means it is "just there" you'll be better off with Intel (based on the reports from others including yours), IIRC you were running Intel right?

[luisma]

15 hours ago, mikel said:

I am wondering.....do folks here percieve an audible difference

Good question, I think I do but I don't trust my ears on that one, it could be completely imagined or subjective

[luisma]

14 hours ago, Rune said:

If you do not require DSD256EC Then you have more choices and less problems.

I have the entry level 5600X on an HDPlex H5 passive cooled and does DSD128 just fine, my initial post here

Very thankful to Jussi for posting as I am on a waiting list for the 5800X, I read about the new Intel coming and 4.8/5.3, IPC 19% increase apparently but don't know if that would be enough for HQP to do stable DSD256 at 7EC (EDIT: meant DSD512 ASDM5EC or 7EC and DSD256 7EC with "loooooong filters")

https://www.anandtech.com/show/16390/intel-previews-11th-gen-core-rocket-lake-core-i911900k-and-z590-coming-q1

 

 

[luisma]

4 hours ago, Yviena said:

Is there actually any reason why Intel seems to have a lower load with EC7 256

I think that aside of the "higher clock" intracore communication is faster and more optimized EDIT:on Intel, I can even dare to go a little further and say that possibly the kernel (at least in the Linux case) will work "better" with Intel than AMD, to my knowledge full kernel support is delayed with AMD for many features, temps, and such, possibly also core task distribution.

 

Just my opinion, I could be wrong

[asdf1000]

2 hours ago, luisma said:

Intel still king

 

Yep, single core speed is (and always has been) king spec, for EC modulators at DSD256. There's no getting around this.

 

 

[Miska]

9 hours ago, Yviena said:

Is there actually any reason why Intel seems to have a lower load with EC7 256, benchmarks seem to indicate that IPC/performance is actually higher around 5%ish with the new 5xxx series.

 

There could be many reasons, but it is very difficult to say why. Same way it is hard to say why excatly Apple's M1 performs so well with modulator workload split to twice as many cores, while Intel/AMD suffers on same setup. Many of the things that matter are chipmakers biggest secrets.

 

Caches could be different (for example algorithms used for caching), instructions per clock efficiency could be different for the instructions HQPlayer uses, practical cache and memory bandwidths could be different, cross core communication latency and bandwidth could be different...

 

Benchmarks are always representative of performance only for the particular benchmark scope. They are not really applicable for HQPlayer use case for example. So to have applicable figures you really need to benchmark with HQPlayer itself.

 

[rando]

@Miska if this is found to be off topic.  I will gladly remove it.  

 

On 1/11/2021 at 7:21 PM, Miska said:

Intel has just announced bunch of new CPUs. Which look interesting, like the H-series, which should offer nice performance with lower TDP.

 

Per the WSJ

 

Quote

Intel Corp. ousted its chief executive in a surprise move that pivots the semiconductor giant closer to its engineering roots after a period of technology missteps, market-share losses and pressure from a hedge fund.

 

Intel on Wednesday said CEO Bob Swan would be succeeded by VMware Inc. chief Pat Gelsinger effective Feb. 15. 

 

This further casts interest on the release of their discrete GPU and it's integration within a very tumultuous market as well as their own CPU offerings.  

[Marco]

11 hours ago, Miska said:

 

There could be many reasons, but it is very difficult to say why. Same way it is hard to say why excatly Apple's M1 performs so well with modulator workload split to twice as many cores, while Intel/AMD suffers on same setup. Many of the things that matter are chipmakers biggest secrets.

 

Caches could be different (for example algorithms used for caching), instructions per clock efficiency could be different for the instructions HQPlayer uses, practical cache and memory bandwidths could be different, cross core communication latency and bandwidth could be different...

 

Benchmarks are always representative of performance only for the particular benchmark scope. They are not really applicable for HQPlayer use case for example. So to have applicable figures you really need to benchmark with HQPlayer itself.

 

Now what could come in handy is a single executable to run as a benchmark to test e.g. the EC filter/modulators (or just 1 for dsd64,128,256 and 512) and output a score. If this would be indicative of the performance of the actual load when playing music, it would be really usefull in exchanging scores cross platform. For instance dsd64 could score  480%, dsd128 220%, dsd256 101% (just above real speed) and dsd512 45%. One could easily test for best scores instead of black and white "it works"/ "doesn't work" as well as be able to exchange scores for comparison (be it with same hardware or not).

[luisma]

16 hours ago, Marco said:

Now what could come in handy is a single executable to run as a benchmark to test e.g. the EC filter/modulators (or just 1 for dsd64,128,256 and 512)

Hi Marco, at the risk of being flagged as pretentious, your idea good but in practice there are many variables at play like source rate, memory used, OS and kernel used, filter used and such, unless you keep the system, content played and such as static which is impractical you can't provide a good reference.

PC benchmarks run for a specific CPU architecture with specific variables and instructions sets, some benchmarks will "blend" certain tests. I think it is very complex and even if you create a benchmark for let's say Redbook content if the music has for example high dynamic range will stress the modulator even more than audio with quiet passages. So you could define fix / static variables for testing but the result will be meaningless when you subject these systems to common usage rendering the benchmark not very useful.

Going back to PC benchmarks (on which I'm not expert either) IIRC there were even specific benchmarks created for specific games because you can''t create a "universal" benchmark for all games since each game will stress the system differently.

 

But what do I know I could be wrong and perhaps this could be possible.

 

Just my opinion

 

[Zauurx]

Yes, I thing the same.
 can, with pleasure, listen to flac files (44/16) played in DSD128 ASDM7EC with a little Tiny Lenovo i5-3470T(2 cores - 4 logicals).
I am using Win10 LTSC which I optimized by "hand" + A0 + Fidelizer + ProcessLasso (42 processes with Roon and HQP in progress).
I had a convolution too.
I did the test yesterday. Telling me that ProcessLasso was not that important. I stopped it.
I had dropouts every 5 sec.
So everything counts .. with a limited CPU (it's different with a 9900K)
Of course, I can not listen  from DSD or HiRes with these settings, but from Redbook .. no problem with 85% CPU load (98% for HQP).

[Marco]

1 hour ago, luisma said:

Hi Marco, at the risk of being flagged as pretentious, your idea good but in practice there are many variables at play like source rate, memory used, OS and kernel used, filter used and such, unless you keep the system, content played and such as static which is impractical you can't provide a good reference.

PC benchmarks run for a specific CPU architecture with specific variables and instructions sets, some benchmarks will "blend" certain tests. I think it is very complex and even if you create a benchmark for let's say Redbook content if the music has for example high dynamic range will stress the modulator even more than audio with quiet passages. So you could define fix / static variables for testing but the result will be meaningless when you subject these systems to common usage rendering the benchmark not very useful.

Going back to PC benchmarks (on which I'm not expert either) IIRC there were even specific benchmarks created for specific games because you can''t create a "universal" benchmark for all games since each game will stress the system differently.

 

Right. Certain things should be set static (Red Book, Modulator and Filter, choose or fix to highest demanding/most popular or generic one), the rest would be of help actually: having more or different memory yet getting a higher score in the same system means it's easier to see what various hardware and software settings do. And vice versa.

 

I'm looking for general scores, so not interested in all possible filter and modulator combinations, just the real demanding popular one. A true benchmark. Just Red Book to X-speed DSD, EC & Cuda (Closed Form maybe?) enabled or not. This to keep the test program easiest to keep up to date and compact. Maybe it's just as easy to add all settings of HQPlayer, or that otherwise as you say, it is of no use at all. Could very well be.

And yes, the proof is in the pudding, iow you need the real application (embedded/desktop/..) to know for sure.

 

But what is really in the dark now, is how much a certain software setting or hardware difference changes calculation speed. We've seen back and forth discussions about greying out and (de)selecting certain settings in order to get at the optimum performance for a given system.

 

In the end, Jussi is the only one with the insight to evaluate this as doable and of a test program's validity for the main programs . I would like to know if it can be done and if it's something to consider.

Above all, Jussi, I can easily imagine you're busy enough as it is, but I also think it would get lots of customers on a much clearer upgrade path, be that for general GPU's, memory, CPU's or settings.

You might even already have a similar environment at home, to measure what certain changes do.

 

It could be build off of HQ Pro, as one has all the options, can write without selecting a real output. Generating a score depending on the (minimum?) achieved speed would be just the indication one needs to know how a system could be optimised, as well as what one shouldn't do. This of course, with a disclaimer that results doesn't make the author responsible.

 

That's it in a nutshell.

 

Of course, one can always install a trial version of HQPlayer and see if it works, but it really doesn't generate detailed of possible bottlenecks. I have no problem keeping an Excel sheet for this.

 

I'll leave it at that.

[luisma]

35 minutes ago, Marco said:

In the end, Jussi is the only one with the insight to evaluate this as doable and of a test program's validity for the main programs

Very true

35 minutes ago, Marco said:

(Closed Form maybe?)

Maybe not, from what I remember from empirical testing on AMD 5600X closed form seems to be a little "better" distributing load across cores, also poli xtr, you will be cheating yourself getting a reference score from these filters (I certainly did)

 

37 minutes ago, Marco said:

I would like to know if it can be done and if it's something to consider.

Good question Marco

[rando]

To recap an old story. 

 

Sole programmer of HQP is reticent, to put it politely, where the topic of benchmarks for this program are concerned.  I'd suggest there are more worthy ideas to pursue which stand to actually bear fruit.  The only benchmark is SOUND QUALITY!  

 

Your personal experience is the platform upon which it is adjudged.  How hard your computer is working might not even make the difference you think it will.  How well the entire system in use was paired and prepared does.  

[Miska]

On 1/14/2021 at 10:17 PM, Marco said:

Now what could come in handy is a single executable to run as a benchmark to test e.g. the EC filter/modulators (or just 1 for dsd64,128,256 and 512) and output a score. If this would be indicative of the performance of the actual load when playing music, it would be really usefull in exchanging scores cross platform. For instance dsd64 could score  480%, dsd128 220%, dsd256 101% (just above real speed) and dsd512 45%. One could easily test for best scores instead of black and white "it works"/ "doesn't work" as well as be able to exchange scores for comparison (be it with same hardware or not).

 

You can already do this with HQPlayer itself by running it to null output. This doesn't produce any output, but performs everything else. And as usual at the end HQPlayer reports how long it took to process the file. This way the playback output doesn't slow down the processing and it proceeds at full steam instead.