The second thing that I consider essential is the hardware, and, in this point, I'll allow myself to extend a little bit more, once it divide in a lot of subjects that need analysis. To facilitate, therefore, I'll separate by subitems:
a) absence of moving parts: the moving parts of a computer consist, basically, of coolers and hard drives. These two things are powered by motors that, apart from being noisy mechanically, are high frequency noise generators, even on a small scale, which reflect indefinitely by the computer. Some people say that, to avoid this problem, external power and a few filters are enough, and the noise won't be audible anymore. I disagree. I've already tried to power my HDD and a cooler with an external power supply, and even with a 5v battery (20000mA power bank). It didn't work. The motors not only generate noise from the cable that powers them, but also emit electromagnetic waves that spread without direction. Since motherboards have very sensitive and microscopic rails, and the processor, memories and internal controllers already work at high frequency, this noise, in the house of MHz and even GHz, generated by the motors, affects the boards and the cables and cause distortions, harmonics, and this reflects in the loss of quality. It's worth remembering here that the computer is the birth of the audio, and that the digital information can be lost and/or distorted until reach the speakers. So, the more care you take with intrusions, the better it will be for the end result, and all of this is perfectly audible. It's also worth remembering that the electric power also makes noise and, especially on the motherboard, this is audible and deserves special care..
Some people maintain that this way of thinking (loss or distortion of computer information) became obsolete with the advent of asynchronous DACs. Well, if it is true, and if the data is buffered previously in the DAC, then, doesn't make sense that devices like ISO Regen, UltraRendu, JCAT USB Isolator and other things bring a positive result. Even a few USB/SPDIF interfaces, such as Mytek, Audiophilleo, SOtM dx-USB HD, M2Tech HiFace would work just as well, instead of the USB connection. What I think is that the purity of the circuit as a whole is, yes, one of the key points of the overall yield. As a solution, I defend the use of solid state storage and passive cooling.
b ) the fewer onboard controllers used, the better. We know that a computer has a lot of onboard controllers, like SATA, RAM memory, VGA, LAN, USB, PCI-Express, processor clock and many others. In my experiences, I noticed that, the smaller the number of controllers in use, the better, clearer and more faithful the audio, and the closer it gets to the analogue, and it gains attributes of microdynamics, decay, spatiality and fidelity of tones and harmonic body. The controllers work at high frequency, and this generates electromagnetic noise that affects everything inside the PC, from the cables to the boards and storage devices.
To allow the use of the controllers in a small number, we must connect to the computer as few elements as possible, and it mean using only USB, LAN and power connections, externally to the computer. The USB connection is only for connecting the PC to the DAC. The network connection is used to connect one PC to another, as well as to control them remotely. There's no need of video connection, mouse/keyboard connections, external hard drives connection. Everything else is completely unnecessary to the audio playback (this is, in my opinion, one of the reasons for why a notebook or a Mac Mini aren't so good as a desktop computer: they have unsurpassed rigidity in hardware set-up). This means that the VGA controller, the integrated USB controller, the fan speed controller (point "a" above), the mouse and keyboard controller and some others, are not in use, at least actively, and only the essential ones remain active for correct operation of the computer and for musical reproduction. From what I've heard, the worst controller for the audio quality is the VGA one. The onboard USB brings a bit of noise, but on a smaller scale, it's something less obvious, but with which I do not want to hang out. This can all be achieved in the absence of a monitor, using a mobile phone or tablet for remote control, and when longer maintenance is required, with remote access via another computer or notebook. Video connection, never.
c) the fact of having hardware is not enough for good performance. You also need to adjust the hardware: in my research, I came across instruction manuals from Computer Audio Design, Computer Audiophile, Fidelizer, and a few others. Each one has its own ideas, and through reading and listening, I have tried to bring together the best of all in a material I have assembled. Several manuals argued that certain features inherent in the motherboard and processor would be harmful to the audio and should therefore be disabled. As in this hobby we cannot believe everything, but also can not doubt anything, I tested what was recommended, one by one. I absolutely disabled everything related to Intel SpeedStep, Turbo Mode, C-States, Virtualization Technology and Hyper-Threading (Audiophile Optimizer recommendatoin). I found improvements in audio quality, but the most vital was Hyper-Threading. This detonated the sound quality. I noticed that this all caused the processor cores to work very unsteadily with energy, now consuming more, and heating more, and now consuming less, for cooling and energy saving. This v-core oscillation, however, brought dirt to the sound, and the deactivation left the frequencies of the cores locked at a fixed value (2GHz) and the sound more fluid and clean. I also deactivated the onboard audio, the SERIAL driver and the USB 2.0 drivers (I only left 3.0 port active) and decreased the video memory to 64MB (in particular, I also deactivated the Audio-PC SATA controllers). I left HPET (High Precision Event Timer) enabled.
Having said that, I ask the license to digress a little to talk about something important: for those who don't know, Hyper-Threading is a feature that processors use to execute multiple threads (threads) in a single core. The practical result of this is that, with it active, Windows "sees" the processor with twice as many cores and processing performance goes up. Therefore, a Core i3 has two physical cores and four virtual cores, as well as some Core i5. Some Core i7 have four physical cores and eight virtual cores, and so on. There are some Intel Xeon with ten physical cores and twenty virtual cores. As a consequence, with Hyper-Threading disabled, a Core i3 processor works only with two cores, as well as some Core i5. This in essence. In addition, disabling the Turbo Mode prevents the processor's frequency from rising beyond the base. It turns out that, because of this sum, performance drops substantially, and then that very good processor with a very high benchmark, isn't so good anymore. The latency of this whole process is very low and more than enough, and the processing is far from itself limit.
I say all this to come to the conclusion that a processor of very high performance isn't essential for high fidelity audio, except for three exceptions that I want to mention in the sequence. To elucidate, let's take as example two processors, a Core i3-6300T and a Core i5-6400T. The first, with Hyper-Threading disabled, will only have two cores working at 3.30GHz. The second, with Hyper-Threading and Turbo Boost disabled, will work with four cores at 2.20GHz. For audio purposes, therefore, what would be the difference of an Intel J5005, which has four physical cores working at 1.50GHz and only 10w of TDP? What would be so significant compared to an Intel N3700 with four physical cores working at 1.60Hz and only 6w of TDP? Only in the 8th generation of processors is the Core i3 now have four physical cores. For more RAM, USB cards, M.2 connections are good, I have no doubt that processors like the J5005 are still the fastest part of the entire computer, and that with everything off, expensive processors with expensive motherboards are underutilized. Despite this, the exceptions are very important.
The first exception I referred to is the Xeon line. These processors have a completely different operation, but more importantly, server motherboards have a much higher, more robust construction with very high precision components, and this greatly benefits the performance of the computer in terms of music reproduction. So, for certain applications, I just find interesting to leave the motherboards with embedded processors to go to the Xeon line. The second exception I mentioned above is the processor cache. Tests carried out by several users are unison in the sense that a large cache brings very significant benefits to the quality of the audio, and that is one of the characteristics of greater responsibility in the approximation of the analog audio. But what is this cache? 6MB? 8MB? Something to conclude, then, that a Core i3-8300, with its 8MB cache, is superior to a J5005, with its 4MB cache? Not at all! I mean something on the order of 25MB, 30MB, up to 60MB cache. Using a 6 or 8MB processor will not even tickle an audio file, which allows me to come to the conclusion that, again, if it's not to mount a PC with an Intel Xeon, it's even better to use a motherboard with an Intel J1900, N3700, N3710, N3160, J4105 or J5005, because they will allow to build a computer with consumption in the order of 20w, as it's what I have here, that consumes something between 20 and 25w, when in reproduction, and this with a SOtM USB card and a linear power supply that I believe is responsible for at least 5w of that total.
The third and more important exception, in my opinion, is related to a Single-PC or Dual-PC setup. When using the Single-PC or Dual-PC configuration, there may be a greater demand for processing, bitstream, latency, and so it may be interesting to have a more powerful computer being used as Control-PC or as Single-PC. Here, using (for now) a computer with an Asrock J1900-ITX motherboard, 16GB of DDR3 RAM, a 1TB SSD, and a dedicated USB card, I could play 192/24 with JPLAY v. 6.2 at 700Hz and 0.01s without interruptions in Single-PC mode, and I can now play 192/24 at 1000Hz also without interruptions with v. 7.0. As it's now the Audio-PC, it's still more than enough to play 192/24, and I think it wouldn't have problems even with DSD256. However, in this case, the Control-PC will import more, so I think that for DSD playback, in both Single-PC mode and Dual-PC mode, at least one computer needs to be powerful (I repeat, not for 192/24) and preferably an Intel Xeon (because I really don't see an advantage in using a Core i5 or i7 if it's to disable various features in the BIOS).
What we can observe, in the end, is that building a dedicated computer, after all these observations, may seem extremely complex, but it isn't. But it's not cheap either. The fact is that for certain applications, it's much smarter to direct the investment to particular components, such as a dedicated USB card and/or a dedicated LAN card, and the return on audio quality will be much more significant than using a Core i3- 8300 compared to a J5005. The issue is not performance, benchmark, but process exclusivity, low heat generation, low latency, low jitter, minimum use of onboard controllers and low power consumption.
To summarize: if you're going to use a Single-PC configuration and will not use an Intel Xeon, buy a motherboard with an embedded processor and direct your investment in the amount of RAM, storage capacity (1TB SSD, at least), linear power supply and dedicated USB board. No need for Core i3, i5 or i7. If you're going to use a Dual-PC setup, the most perfect would be to use both Intel Xeon-based computers, but if that's not possible, I also don't see a need for anything other than an embedded processor, keeping the investment focus on the USB card, on te dedicated network cards and linear power supplies. And, for those who would like to know my opinion, I say: yes, I think the best USB and network cards in existence today are the JCAT FEMTO. Better than SOtM and much better than PPA. Currently, I have a SOtM tx-USBexp, but I intend to buy a JCAT as soon as possible, because I don't consider SOtM a reliable brand in terms of effective troubleshooting, and have literally abandoned the evolution of USB cards. This differentiates JCAT from other brands.
In the next post I'll write about the software.