rocky500 Posted January 26, 2019 Posted January 26, 2019 (edited) 9 minutes ago, Ittaku said: He's never said that. What he said is it CAN be done on a CPU. It's not just about doing millions of taps, it's the filter algorithm itself, and the WTA in the Chord equipment is proprietary and secret, and you can't reproduce that with existing software upscaling solutions. What you can do, is many more taps... If the Chord stuff sounds better, it proves that it's not just a matter of how many taps. I don't think we are reading the same thread here. Something like this. I don't want to go back and read He said the WTA filters are nothing special and marketing fluf, HQ player already does 2 million taps or something. CPU is better than FPGA at this So why would not be any comparisons showing this? Edited January 26, 2019 by rocky500
Ittaku Posted January 26, 2019 Posted January 26, 2019 Just now, rocky500 said: I don't think we are reading the same thread here. Something like this. I don't want to go back and read He said the WTA filters are nothing special and marketing fluf, HQ player already does 2 million taps or something. CPU is better than FPGA So why would not be any comparisons showing this? All I could say is if someone were to steal the WTA algorithm and make a plug in for it on a software media player in CPU, there'd be hell to pay...
Chill3 Posted January 26, 2019 Posted January 26, 2019 2 hours ago, davewantsmoore said: LOL Chili, what are you doing?! Man you need to Chill, seriously you are now so self absorbed on a irrelevant technical point that your concerned on who “ likes” a post
davewantsmoore Posted January 26, 2019 Posted January 26, 2019 1 hour ago, Chill3 said: Man you need to Chill, seriously you are now so self absorbed on a irrelevant technical point that your concerned on who “ likes” a post Why do people keep posting back to argue if it's so irrelevant?! "Concerned" is perhaps a bit dramatic... I'm not particularly, except for anyone trying to follow along at home, and actually learn something. I just wonder by what logic you can like Bills post. You don't know he's correct .... so (!?) you're picking sides, I guess.
Chill3 Posted January 26, 2019 Posted January 26, 2019 5 minutes ago, davewantsmoore said: Why do people keep posting back to argue if it's so irrelevant?! "Concerned" is perhaps a bit dramatic... I'm not particularly, except for anyone trying to follow along at home, and actually learn something. I just wonder by what logic you can like Bills post. You don't know he's correct .... so (!?) you're picking sides, I guess. Just liked the post, made me smile
davewantsmoore Posted January 26, 2019 Posted January 26, 2019 1 hour ago, rocky500 said: @davewantsmoore I am trying to understand where you are coming from in this thread. Can you show me somewhere that upscaling with a CPU sounds better than the m scaler when people have tried both? Let's walk it back some..... There's an article which is posted on the Chord website.... explaining how the M Scaler works. Chord say that degree to which you approximate the "sinc" function (sinc theoretically extends to infinity) has a big impact on sound quality. The article then goes on to suggest, that if an accurate representation of sinc is important ..... that you could use a computer, and devote as much time and/or processing power to computing the sinc function as you wanted ..... and this would be "better" (as in a better approximation of the sinc function) than the M Scaler. According to Chord, a better approximation of the sinc function sounds better. It would not matter whether you did the processing in "real time" (eg. what accuracy you could compute in ~600ms) .... or whether you devoted super-large amounts of time to computing sinc (eg. wait 10 hours for your audio). The computer can both do a "better" job at computing sinc in a limited time ..... and it can compute sinc for an "unlimited" time (as long as you want), which the DSP cannot do (easily) due to its architecture. What *I* came in to say..... was that if the audio quality continues to increase as they increase the approximation of sinc ...... then Chord should (and could) use a faster calculator to improve the result. There are a number of big and small caveats as to why they potentially didn't take this path ..... but overall the logic is there. There's been a few nonsensical responses to my comment. One that RW wouldn't agree with the article. Sure, he might not want people to hear the message "don't by an M Scaler, cos DIY" ...... but the article only kinda said that in a round about way (while illustrating the point that better sinc = better sound). RW isn't going to disagree with that (compute better sinc), as it's a core premise about how the M Scaler works. The other one is that a computer (as in an x86 computer like what is made by Intel/AMD) doesn't have as much processing power. This is so wrong that it would be hilarious... if we didn't have it presented as straight faced fact ..... causing anyone casually dredging up this thread from an internet search to conclude the answer is "no one knows" (although they would be stupid to consider this website actually gives them a correct answer). That's sad. 1 hour ago, rocky500 said: Can you show me somewhere that upscaling with a CPU sounds better than the m scaler when people have tried both? No... I'm not going into this.... as is was never part of my original comment. This is a straw-man invented by others. Chord say that closer approximation of sinc sound better. Faster calculators can more closely approximate sinc. That's it..... those are the facts. Nothing controversial. If you wanted to actually best the M Scaler with a computer .... then you would need to either use a very very long instance of sinc.... or you would need to experiment with how to truncate your "short" sinc.... and work on the bandlimiting filters, etc. One player which has done all this work is XXhighend. They also optimise their work to their own DAC (although you can use your own DAC). There should be no astonishment, that XXhighend ask you to use an extremely high-end Intel x86 CPU .....as the extra processing power can be put to good use.
davewantsmoore Posted January 26, 2019 Posted January 26, 2019 29 minutes ago, Chill3 said: made me smile Sure, ,hence my "LOL".
davewantsmoore Posted January 26, 2019 Posted January 26, 2019 2 hours ago, Ittaku said: He's never said that. What he said is it CAN be done on a CPU. It's not just about doing millions of taps, it's the filter algorithm itself, and the WTA in the Chord equipment is proprietary and secret, and you can't reproduce that with existing software upscaling solutions. What you can do, is many more taps... If the Chord stuff sounds better, it proves that it's not just a matter of how many taps. What I would have liked to do, about 20 pages ago.... is unpack some of the things here - and how they relate to what is in the article Chord posted.... and how it relates to the "WTA algorithm". The algorithm is about "what happens when sinc doesn't actually extend to infinity". You have to truncate it - and how you do this is important. If you did use a computer in the impractical way inferred in the article (use heaps of time/cpu to get a more accurate representation), then that "WTA algorithm" isn't (as) important any more. So, using a computer (in the impractical way inferred) doesn't rely on it. .... but as I mentioned in my last post, this is mainly academic interest only. The real comment was "Why don't Chord use more taps? Will Chord use more taps?" ('cos 1 million taps is "meh" .... even though their marketeers tell us otherwise).
Chill3 Posted January 26, 2019 Posted January 26, 2019 Sorry to be so direct but 1 hour ago, davewantsmoore said: Let's walk it back some..... There's an article which is posted on the Chord website.... explaining how the M Scaler works. Chord say that degree to which you approximate the "sinc" function (sinc theoretically extends to infinity) has a big impact on sound quality. The article then goes on to suggest, that if an accurate representation of sinc is important ..... that you could use a computer, and devote as much time and/or processing power to computing the sinc function as you wanted ..... and this would be "better" (as in a better approximation of the sinc function) than the M Scaler. According to Chord, a better approximation of the sinc function sounds better. It would not matter whether you did the processing in "real time" (eg. what accuracy you could compute in ~600ms) .... or whether you devoted super-large amounts of time to computing sinc (eg. wait 10 hours for your audio). The computer can both do a "better" job at computing sinc in a limited time ..... and it can compute sinc for an "unlimited" time (as long as you want), which the DSP cannot do (easily) due to its architecture. What *I* came in to say..... was that if the audio quality continues to increase as they increase the approximation of sinc ...... then Chord should (and could) use a faster calculator to improve the result. There are a number of big and small caveats as to why they potentially didn't take this path ..... but overall the logic is there. There's been a few nonsensical responses to my comment. One that RW wouldn't agree with the article. Sure, he might not want people to hear the message "don't by an M Scaler, cos DIY" ...... but the article only kinda said that in a round about way (while illustrating the point that better sinc = better sound). RW isn't going to disagree with that (compute better sinc), as it's a core premise about how the M Scaler works. The other one is that a computer (as in an x86 computer like what is made by Intel/AMD) doesn't have as much processing power. This is so wrong that it would be hilarious... if we didn't have it presented as straight faced fact ..... causing anyone casually dredging up this thread from an internet search to conclude the answer is "no one knows" (although they would be stupid to consider this website actually gives them a correct answer). That's sad. No... I'm not going into this.... as is was never part of my original comment. This is a straw-man invented by others. Chord say that closer approximation of sinc sound better. Faster calculators can more closely approximate sinc. That's it..... those are the facts. Nothing controversial. If you wanted to actually best the M Scaler with a computer .... then you would need to either use a very very long instance of sinc.... or you would need to experiment with how to truncate your "short" sinc.... and work on the bandlimiting filters, etc. One player which has done all this work is XXhighend. They also optimise their work to their own DAC (although you can use your own DAC). There should be no astonishment, that XXhighend ask you to use an extremely high-end Intel x86 CPU .....as the extra processing power can be put to good use. Ok, I encourage you to apply for a role work Chord as your clearly smarter than there current staff even better make your own competive product 1
Guest Eggcup The Daft Posted January 26, 2019 Posted January 26, 2019 (edited) 2 hours ago, Chill3 said: Ok, I encourage you to apply for a role work Chord as your clearly smarter than there current staff even better make your own competive product Clearly Chord's current staff understand exactly what @davewantsmoore is saying, because, er, they said it themselves. The m-scaler is built the way it is, because that is what they deem best overall after taking into account all aspects of building their products and not just this one. As for a different digital playback system design, how do you know davewantsmoore isn't actually using an upscaling application on an Intel PC? There are a few ways to do that out there... Edited January 26, 2019 by Eggcup The Daft
Ittaku Posted January 26, 2019 Posted January 26, 2019 For what it's worth, I used to upscale on my PC at extreme levels but once I upgraded my DAC, it sounded better at the native sample rate. Whatever my DAC does in terms of upscaling and filtering is better than what I could achieve on my (very powerful) PC.
Guest Posted January 26, 2019 Posted January 26, 2019 Looking forward to catching up with @Ittaku @joz and @Kaynin tomorrow for the M Scalers second tour lol
Ittaku Posted January 26, 2019 Posted January 26, 2019 12 minutes ago, Sime V2 said: Looking forward to catching up with @Ittaku @joz and @Kaynin tomorrow for the M Scalers second tour lol Keeping the DAC warm for you. The valves will be cranked first thing in the morning to be lovely by the time you arrive.
joz Posted January 26, 2019 Posted January 26, 2019 39 minutes ago, Sime V2 said: Looking forward to catching up with @Ittaku @joz and @Kaynin tomorrow for the M Scalers second tour lol Looking forward to it. may even have a few free seats?
Ittaku Posted January 26, 2019 Posted January 26, 2019 Wish I could invite more people to mine, but my room is embarrassingly small.
Cruncher Posted January 26, 2019 Posted January 26, 2019 On 25/01/2019 at 3:13 PM, davewantsmoore said: .... As an example. The parallel processing pipe in my (2016 era) computer, can handle over 10 trillion ops/seconds.... @davewantsmoore What is the brand and model number of this processor ?
bhobba Posted January 26, 2019 Author Posted January 26, 2019 10 hours ago, davewantsmoore said: It's fine that you don't understand this stuff Bill Really. Or maybe it is you that does not. I could point to the degree I have in computer science and 30 years working as a computer programmer - but I will not. I will however point you to a site where professors in computer science hang out and we can take the discussion over there: https://www.physicsforums.com/threads/xilinx-xc7a200t-vs-intel-processor.964960/ We will see. Thanks Bill
bhobba Posted January 26, 2019 Author Posted January 26, 2019 8 hours ago, davewantsmoore said: Why do people keep posting back to argue if it's so irrelevant?! Because Dave it goes against what Rob Watt's has stated. He had to wait for the processor to come out to do it. Yet you seem to think it can be done on an Intel processor. Basically you think Rob is telling a fib. I have zero idea why you would want promulgate such a notion. It has been mentioned not all cores are equal - of course - but the engineer who wrote the code for it says he needs the power of the 740 cores to do it - the Intel processor simply does not have that kind of power. My PC uses a pretty high end Intel processor but costs much less that the M-Scaler. If it could be implemented on such hardware Chord would have done it and be able to sell the M-Scaler at a cheaper price with a higher profit margin. Unless you have a conspiracy type view of such things simple logic would indicate it cant be done with the processors you find in PC's Can anyone be 100% sure of this? Unless you have actually written the code on both processors of course you can't - all you can do is take the word of the person that did it. Thanks Bill 1
bhobba Posted January 26, 2019 Author Posted January 26, 2019 (edited) 10 hours ago, rocky500 said: HQ player already does 2 million taps or something. Now we may be getting to what the issue really is. There is a long standing argument between the maker of HQ player - Miska (you can find his posts etc over on Computer Audiophile) and Rob on what a Tap is. Miska claims exactly what Dave claims eg he wrote - 'Funny thing is that computers doing stated upsampling to 16x with 1M taps doesn't even break a sweat. It is so easy to pull it to completely different figures with computer, if one wants....'. That will not be resolved here - except to say what Rob considers a Tap requires more processing power than what Miska thinks of as a Tap. BTW HQ player is the real deal - it does sound good but the reports are the M-Scaler is better. I will need to compare the two when my health improves - that is if I can find the windows machine I had HQ Player on - I think I let someone else have it - I switched over to Audirvana which to me had exactly the same sound quality as HQ player - besides I got sick and tired of every time you got a Windows update it was Russian roulette if the DAC drivers would work. Apple has no such issues and my sanity is better for it. Thanks Bill Edited January 26, 2019 by bhobba
Volunteer sir sanders zingmore Posted January 26, 2019 Volunteer Posted January 26, 2019 3 hours ago, bhobba said: I could point to the degree I have in computer science and 30 years working as a computer programmer - but I will not but you just did ...
davewantsmoore Posted January 26, 2019 Posted January 26, 2019 9 hours ago, Ittaku said: For what it's worth, I used to upscale on my PC at extreme levels but once I upgraded my DAC, it sounded better at the native sample rate. Whatever my DAC does in terms of upscaling and filtering is better than what I could achieve on my (very powerful) PC. .... but not like Chord are doing, right!? (whereby you intervene in the "approximation of sinc", more like Phasure do) 5 hours ago, bhobba said: Really. Or maybe it is you that does not. I could point to the degree I have in computer science and 30 years working as a computer programmer - but I will not. I have the same qualification. It is not about whether one knows the answer to these sort of questions .... but how one approaches answers they do not know. Forrest Gump tells us straight "stupid is, as stupid does". Asking for peoples "opinion".... on performance numbers which are not "opinions" ..... is just a perfect example of that. Quote The designer claims it was only the release of processors of this power that allowed him to do it. There are almost certainly other caveats at play here ..... like that it was this level of performance available in the power and cost envelope they wanted/needed to use. Evidenced by the fact that the XC7A200T is not even close to the fastest part in the series..... for example, you can have a version of it with 5000 "cores" (and faster everything else also). The A200T just happens to be the fastest part in the "optimised for performance per watt" category of this processor. Why don't you just run a filter on your x86 processor ... and you will see the orders of magnitude at play here? 5 hours ago, bhobba said: We will see Hardware efficient filters like this are difficult to implement. I certainly understand why they chose the processor they did. You can however do it many many many times faster (with horrible efficiency) on a x86 CPU.
davewantsmoore Posted January 26, 2019 Posted January 26, 2019 7 hours ago, Cruncher said: @davewantsmoore What is the brand and model number of this processor ? Nvidia GTX1080TI
davewantsmoore Posted January 26, 2019 Posted January 26, 2019 6 hours ago, bhobba said: Because Dave it goes against what Rob Watt's has stated. He had to wait for the processor to come out to do it. Yet you seem to think it can be done on an Intel processor. Basically you think Rob is telling a fib. No, I don't think that ... I think you are misunderstanding exactly what he is saying.... which is embarrassing for you. He could have used an Intel x86 processor .... but you would never do that... an Intel x86 CPU is waaaay to big and powerful, and needs too much supporting infrastructure.... and has an instruction set which is totally unnecessary for this task. He needed to wait until the processing power was available in a power and cost envelope which made it appropriate for an M Scaler. RW would probably say that other processors would "not sound as good", due to power/efficiency, etc.... and he may be right about that (I don't know, of course). The point is that he could have had more taps right now .... and in the near future many more taps will be available (in an "appropriate" form). .... and point is that the article posted by Chord on their website which says "a computer can do it better" .... is not incorrect. The problem is that many people here have misconstrued exactly what "it" is. Seems many of you have done so to fit your "outrage", which I find quite pathetic. 6 hours ago, bhobba said: I have zero idea why you would want promulgate such a notion. I've gone into a fair amount of detail on where I'm coming from on this. 6 hours ago, bhobba said: It has been mentioned not all cores are equal - of course - but the engineer who wrote the code for it says he needs the power of the 740 cores to do it - the Intel processor simply does not have that kind of power. My PC uses a pretty high end Intel processor but costs much less that the M-Scaler. Bill this is stupid.... If you don't understand what you are talking about, then just stop man. The XC7A200T costs approximately the same as a mid-range Intel x86 CPU to buy .... although the x86 then needs much more support in infrastructure (like much of the stuff you see on a typical "motherboard"). This put the price for the x86 hardware higher. 6 hours ago, bhobba said: If it could be implemented on such hardware Chord would have done it and be able to sell the M-Scaler at a cheaper price with a higher profit margin. No they wouldn't have .... as there are many other reasons they would not use such a CPU. Too big, too hot, too much power hungry, too much unnecessary stuff needed >_< 6 hours ago, bhobba said: Unless you have a conspiracy type view of such things simple logic would indicate it cant be done with the processors you find in PC's Your simple logic is wrong ..... and the datasheets of these processors are correct. 6 hours ago, bhobba said: Can anyone be 100% sure of this? Unless you have actually written the code on both processors of course you can't Oh dear.
davewantsmoore Posted January 26, 2019 Posted January 26, 2019 7 hours ago, bhobba said: I will however point you to a site where professors in computer science hang out and we can take the discussion over there: https://www.physicsforums.com/threads/xilinx-xc7a200t-vs-intel-processor.964960/ A few who also like to quick fire reply about things which they don't really understand, from the look of few of the replies you already have. "My hearing cuts off at 12000Hz" Heh.
Guest Posted January 26, 2019 Posted January 26, 2019 This is my limited understanding of Why that particular chip was used, and I don’t have the technical knowledge of either type of chip. My understanding is that an FPGA chip can be given a task though code, which that code is stored on the chip itself (?) and once it’s given it’s instruction set, this is all it will ever do. Providing it’s on a mother board that suits its needs for the job, it doesn’t need external software to run it, unlike a traditional CPU (widows etc) giving it the advantages of reliability, cost and more importantly, instant boot times. The only processor that can handle the type of task within an framework to do the job of an M-Scaler with extreme efficiency. Of course this one was probably chosen because of cost and the right amount of horse power to achieve the 1m tap goal. And as far as I know, he’s only using 540 of the total cores. Running an M-Scaler with a X86 will be plagued with traditional PC issue, where a FPGA has zero (?) Sound about righ?
Recommended Posts