Does CPU cache relate to sound quality?

[elcorso]

The question is, for all that say that having more cache/processing power is not important, have they tried it?, or is just a technical opinion?.

 

"The proof of the pudding is in the eating".

 

I believe it's better to ask a good music player or DAC designer

 

As I said at the beginning I'm far away to give scientific opinions about computers, but please try HQ Player upsampling (plus 'heavy filters') 16/44 to DSD256 with a Mac Mini 2010 and a Mac Pro 2015 (I own both) and then we can talk...!

 

Happy listening,

 

Roch

[NAIDIVER]

"The proof of the pudding is in the eating".

 

I believe it's better to ask a good music player or DAC designer

 

As I said at the beginning I'm far away to give scientific opinions about computers, but please try HQ Player upsampling (plus 'heavy filters') 16/44 to DSD256 with a Mac Mini 2010 and a Mac Pro 2015 (I own both) and then we can talk...!

 

Happy listening,

 

Roch

 

The heat is on!!

[mansr]

"The proof of the pudding is in the eating".

 

I believe it's better to ask a good music player or DAC designer

 

As I said at the beginning I'm far away to give scientific opinions about computers, but please try HQ Player upsampling (plus 'heavy filters') 16/44 to DSD256 with a Mac Mini 2010 and a Mac Pro 2015 (I own both) and then we can talk...!

 

No matter now intensive your processing, there is a point where the computer is fast enough to perform it reliably. Adding cache (or anything else) beyond that will not automatically bring any improvement. It can just as easily make matters worse if you have an analogue interference problem.

[yincr]

I agree with that, but the interesting part is, what is that point?, how fast is enough?.

[rwwjr44]

No matter now intensive your processing, there is a point where the computer is fast enough to perform it reliably. Adding cache (or anything else) beyond that will not automatically bring any improvement. It can just as easily make matters worse if you have an analogue interference problem.

 

Yes. As with most things in computers, striking the proper balance is most important. Arbitrarily decreasing latency times and increasing cache may upset that balance and make matters worse. What is the proper balance? Well... now you get into computer engineering and the actual components used to build your computer.

 

Chances are decreasing the latency and increasing cache to your computer installation's limits won't do any harm because the designers probably built in some "fudge" factors.

[mansr]

Yes. As with most things in computers, striking the proper balance is most important. Arbitrarily decreasing latency times and increasing cache may upset that balance and make matters worse. What is the proper balance? Well... now you get into computer engineering and the actual components used to build your computer.

 

Chances are decreasing the latency and increasing cache to your computer installation's limits won't do any harm because the designers probably built in some "fudge" factors.

 

Latency in an audio application matters only for things like digital synthesizers used as live instruments. There you need the latency to be no more than 5 ms, preferably less. For pure playback of recorded music, 100 ms plenty low enough, and even a modest computer has no trouble keeping up with that if it is at all fast enough to do the required processing. Nobody cares if it takes one tenth of a second from pressing play until music comes from the speakers.

 

If heavy signal processing is involved, more computing power is required, and a larger cache can play a part in providing that, but latency has little to do with this. Such processing is always done in blocks; while a finished block is playing, the next is being processed. As long as a block can be processed in less time than it plays for, everything is fine. You can put more or less buffering between this processing and the DAC, which increases or decreases the latency. If the buffer is too small, you do run the risk of it depleting if there's even a tiny glitch in the preceding DSP stage while a larger buffer will mask an occasional late block as long as the average processing time per block is smaller than the playback time.

 

What's important to understand is that the amount of latency in the playback chain does not affect how it sounds, only when it sounds. Pursuit of low latency in music playback is wasting effort that would be better spent on preventing electrical noise from the computer entering the analogue side.

[yincr]

I hate arguing about something that nobody will try to prove, in a way, that there won't be any doubt about the correct answer. But, I can tell something about my own experience. I think that the amount of cache, is more related with the processing power/CPU, so, probably i3 PCs will have less cache than a i7 PC, and that's something, I guess, you can not choose, because it is part of the design. In my case, I use HQPlayer with a NAA, and I have tried different computers in my setup, and a close friend has tried different computers than mine, in a different setup, both are very resolving systems. And we have found, that fast computers sound better than slow ones, even if the latter has no drop outs. As someone said in this thread, balance is important, in my setup, the powerful computer is the one that runs HQPlayer, and the optimised (trying to avoid different kind of noise) but also powerful, is the NAA, that is a LPSU/fanless NUC i7. So, I guess there is a correlation between latency, cache size and processing power, but it is not practical nor worthwhile, to find out this correlation, but in my case, It was worthwhile try different escenarios using common sense, and choose the one, I thought that sounds better.

[rrstesiak]

All:

The original post had the following requirements:

 

Someone told me that more cpu cache is better. That's why i7 ( 8mb cache ) sounds better than i3 and i5 ( 6mb cache )

 

While another theory is lower cache sounds better.

 

What is right and what is wrong?

 

At first I dismissed this post as a bit esoteric and pretty much irrelevant as far as audio playback on computers, but I do have the following rather detailed story of sorts to tell which may surprise some readers.

 

1. First of all, system requirements need to be stated. The OP indicated that CPU cache has a direct correlation to the perceived, but as of yet immeasurable quality, of played back music.

2. Further statements in the chain were made that the larger the cache/faster the cpu, the better the sound.

3. To narrow the scope, I would like to limit this discussion to the PLAYBACK of audio only. However, the BEST possible playback.

 

It is in my experience, with current technology, the best sonics are obtained by carefully specifying hand picked components that can handle audio playback requiring the LEAST power possible without adversely affecting playback.

 

In my possession I have two Linux computers, both running MPD and FLAC files out to a Bryston BDA-1 DAC via M2TECH HiFACE TWO USB/SPDIF converter via BNC cable.

In my direct experience with my lengthy and thorough recent project titled My Custom Designed Music Server for under $500 that matches $3,000 Servers

The minimal system far surpasses the powerful, workstation class system in audio playback.

And I mean *FAR* surpasses. While I do not go into details of why the more powerful system was less capable at producing the best audio in that article, I can list some main reasons here:

 

1. Power Supply: The Workstation requires 750Watts; in a traditional ATX power supply. I chose an excellent one which only utilizes the fan under moderate or greater load. In the case of mpd audio playback, the fan never spun up. However, these power supplies by design are not spec'd for audio; therefore they are "noisy".

2. I chose the now dated but still very capable quad core Intel Ivy Bridge 3770K i7 CPU, utilizing 22nm tech, 1024MB L2 cache (256MB per core), and 8MB L3 cache. TDP is 77 Watts base frequency; non-overclocked.

2a. To run this CPU at core and overclocked frequencies with logevity and reliabilty in mind, I had to resort to Liquid Cooling - introducing another fan and associated but silent radiator and hosing. Of critical note, a fan of some sort is nearly mandatory for this class CPU even today. There may be some exotic (expensive) other cooling methods including custom cases, but that would blow my budget and was not necessary for my scientific work.

3. To run this CPU, as well as PCI Gen3 slots, and many other motherboard features, a significant amount of additional discrete electrical components are required... which equals noise..unwanted in audio.

 

I could probably list a few more but these should suffice.

 

In stark contrast, my lower power Intel Braswell Embedded Linux server far surpasses the workstation in audio playback. The main reasons are as follows:

 

1. Power Supply: embedded botherboard uses a simple 19v DC in, with an external laptop-like "brick" power supply rated at 60 Watts. While this falls short of a linear power supply, even it outperforms the workstation classes' high efficiency ATX power supply in audio. Further, it removes the power supply from the internals of the server. I have performed measurements in the past and EMF effects span about 12 inches from their source. EMF = bad for audio.

2. CPU is Intel Braswell N3150, utilizing 2MB L2 cache and 14nm tech, with 4 cores and TDP of a mere 6 watts! Even while playing back 192/24 flac files, cpu utilization is only on average around 12% on ONE core.

3. As the board is an embedded solution with minimal power requirements and a mini-itx form factor, reducing the amount of discrete components and circuitry to the minimum required; resulting in not only the desirable effect of improving audio playback performance, but yielding a much more aesthetically pleasing music rack component.

 

In conclusion, I have listed a few key technical reasons WHY the lower power system prevails, but at a very high level I can simply state the workstation class solution for audio fell short of the performance of my NAD 516 entry-level audiophile CD player, but my low power system exceeds the performance of this CD player. I have written many articles on other forums detailing this comparison.

 

Just providing a different point of view in the hopes this makes folks think twice.

(it is obvious the embedded solution has a SMALLER cpu cache but sounds BETTER)

 

In closing, another very significant fact is I am far from being alone in these concepts. In fact, Bryston's BDP series of music servers costing $3,000 follow very similar logic. I am certain if I researched, many other high-end audio companies also use this architecture.

 

Respectfully,

 

Ron

 

ps. to be brutally honest, I can not think of a single reason how one could possibly actually HEAR and be able to reliably tell in a blind study, given two identically configured PC's expect for CPU cache, which computer has a larger cpu cache in music playback; all else bring equal. The law of diminishing returns is at it's greatest in this thought experiment backed up with my experimentation that in my case yielded a smaller cache is "better"; however, I again feel the many other system parameters far outweighed the effect of the cpu cache.

[yincr]

I agree with your point of view, that the best solution would be, enough power for the music playback using the least power possible, I guess, that is a universal goal for everything. But, there is no just one better solution for everybody, because everybody has different needs, if you would like to use HQPlayer/Closed Form/DSD7/4xDSD or BugHead, I do not think, that they would run well in your system. That's why, if I buy a Computer, I would like to buy a powerful one, that will give enough flexibility to try different players.

[audiventory]

The question is, for all that say that having more cache/processing power is not important, have they tried it?, or is just a technical opinion?.

 

Technical opinion usually based on knowledges how computer and DSP work inside.

 

These knowledges based on theoretical stuff and practical experience.

 

Before I don't meet "magical" behaviour of tecnical object without technical reasons.

 

Except printer, that almost stable won't work, when need urgently print document

[yincr]

jaja, if everything had to be explained just by technical opinions, this site wouldn't even exist. Let people who have had any experience, tell us what they think, I think that is part of being here reading in this forum. I am an Electrical Engineer and Software Engineer, but also admirer of Einstein, "Not everything that counts can be counted, and not everything that can be counted counts.". I also think that if the hobby is about listening, I should be using my ears to judge. Of course, technical knowledge and common sense, should be used to try to avoid being cheated on. I stop her, I am not trying to convince anybody, I just tried to share my experience, the OP asked about cache size and sound performance, and in my experience, faster processor usually are sold with more cache, so, I do not know how to prove anything about cache and sound, but in my context, faster PC sounds better than slow ones (both fairly optimised regarding all kind of noise).

[mansr]

Technical opinion usually based on knowledges how computer and DSP work inside.

 

These knowledges based on theoretical stuff and practical experience.

 

Before I don't meet "magical" behaviour of tecnical object without technical reasons.

 

Except printer, that almost stable won't work, when need urgently print document

 

http://theoatmeal.com/comics/printers

[audiventory]

I am an Electrical Engineer and Software Engineer, but also admirer of Einstein, "Not everything that counts can be counted, and not everything that can be counted counts.". I also think that if the hobby is about listening, I should be using my ears to judge.

 

I agree. Need keep open mind.

 

But same sequence of digits must sound same on same apparatus.

Cache size don’t impact to content.

Cache size can impact to bursts in audio stream. After some size value (when no bursts) cache size don’t impact to audio stream bursts for same system.

 

Of course, if apparatus work properly without bugs.

 

I carefully trust to ears impressions. My ears is not stable instrument (mood, health, air pressure, temperature,…). And for me hard detect difference 2 dB of noise level in audible, and especially in inaudible range.

 

As example, spectrum analysis allow do it 24/365 with precision that allow measurement instrument.

Independent factors mentioned above.

Also it can repeat everybody who can do it.

 

 

faster processor usually are sold with more cache, so, I do not know how to prove anything about cache and sound, but in my context, faster PC sounds better than slow ones (both fairly optimised regarding all kind of noise).

 

Are you want say that cache size impact to noise level?

Could you describe (if possibly in details) how you compare noise for both cases (used computer configuration, apparatus, time between tests, repeatability,…)?

I suppose, reason not in cache size. There are many variants.

 

 

Why I Believe Printers Were Sent From Hell To Make Us Miserable - The Oatmeal

 

Ha-ha-ha!

[panosm]

Guys, easy now:) If i believe that sounds better to me (or you) then that is what matters:) if you believe that a smaller solution sounds the same, go for it:)

[mansr]

Are you want say that cache size impact to noise level?

Could you describe (if possibly in details) how you compare noise for both cases (used computer configuration, apparatus, time between tests, repeatability,…)?

I suppose, reason not in cache size. There are many variants.

 

I once had a computer, a DEC Alpha from 1996, where electrical noise from various system activities made it into the analogue side of the onboard sound card. If I ran a test specifically designed to cause lots of cache misses, a distinctive hiss emanated from the speakers. The SCSI bus had a different noise signature. I plugged in whatever PCI sound card $10 got you at the time, and the noise mostly vanished. In this particular system, a larger cache probably would have made some degree of improvement, but it would still have been terrible.

 

A modern PC is a completely different story. The noise levels are much lower to begin with, and with a decent external DAC, very little will get through.

[audiventory]

Right developed internal DAC also good defended by noise.

 

As rule all enough pro ADC and DAC (both internal and external) have same level of noise about -120 dB.

 

I don't stumble about audio interface that give noise floor significantly better.

 

Each manufacturer test own hardware on diferent systmes (including different cache). However, noise level almost same for all.