  |
CMP /
05ComponentsChapter 5 - Selecting components for a cMP² project This chapter makes general suggestions about component choice and setup. Careful selection can offer good results quickly but, while a PC can overcome engineering issues associated with reading CDs in real time, it is still an electrically noisy environment. This limits the quality of audio reproduction. Later chapters describe how to address the issue further by minimising power consumption, mechanical vibration and motherboard traffic and offer step-by-step design examples. Full assembly is explained in Appendix A.
CasePerhaps unexpectedly, case design is important for good performance. Most cases rely on fans, offer poor vibrational damping, allow electrical radiation from the power supply unit (PSU) to permeate the case and so on. The chosen motherboard runs cooler than most once redundant features are disabled but attention must still be paid to cooling. A key design aim is to avoid fans as they vibrate and consume power. Cases for Home Theatre PCs that address these issues are available from manufacturers like Thermaltake and Zalman, whose TNN-300 case can be recommended. It is heavy (reducing vibration) with a solid aluminium chassis, uses heat-pipe cooling (no fans) and features rubber mounts for disk drives and a good, externally mounted Power Supply Unit (PSU). It certainly looks the part of a high-end CD player. Power Supply UnitEven if a conventional case is chosen on cost grounds, it is imperative to use a quality PSU. Note that peak output will not even be approached: the aim is to operate the PC with the lowest possible power consumption, typically under 25 watts. Good results have been obtained from the Antec EarthWatts (430 watts) which provides good filtering, low ripple voltages and active PFC. Many other good choices are available. The PSU fan can be disabled only when all optimisations are implemented. Motherboard & CPUAlthough cMP² can perform ultra-high-quality upsampling of music data, it needs a powerful processor to do so. Intel recently released a Core 2 Duo CPU using its innovative 45nm technology. This provides adequate processing capability without unduly impacting the design goal of low power consumption. A small motherboard was chosen as larger boards have longer runs of embedded copper (which suffers greater antennae effects) and have more components which have no role in this application but consume power and associated noise. The Intel G31 northbridge, which supports 45nm processors and has itself low power consumption, is used on the recommended motherboard, the Gigabyte GA-G31M-S2L (or GA-G31M-S2C or GA-G45M-UD3). A micro-ATX board with passively-cooled chipsets, it supports DDR-2 RAM (which uses less power than DDR) and allows flexible configuration. Even with fanless cooling throughout, processor temperatures while running cMP² are no more than about 45°C during playback. This illustrates the superiority of 45nm technology. When upsampling to 96k, CPU load is 60% (at individual CPU core level) even when underclocked to 1.2GHz and with the FSB at 800Mhz.
CPU RF noise increases as clock speeds rise and there are audible benefits to be had by running at the lowest viable clock rate. However, particularly when upsampling, a trade-off has to be made between processing speed and power consumption. The Intel E7xxx uses 45 nm processor technology. Despite power consumption as low as anything in the range, it offers ample processing speed even when, as here, under-clocked. Its performance is due in part to a 3 MB cache. DisplayUse only motherboards with integrated graphics – do not use a graphics card. Chapter 7 explains how to configure displays to minimise power consumption and data traffic. RAMRAM affects sound quality. Extensive trials with a range of memory types suggests that high-end memory is worth the premium, enabling stable operation at low clock speeds. Best results are obtained with DDR2 533/667 RAM which supports low latencies such as Kingston’s HyperX range. Use at least one GB of high quality RAM. The consensus among system builders is that a 1 GB module sounds noticeably better than two 512 MB modules. (cMP² setup leaves about 920MB of this for music data.) Smaller RAM sizes if available are best (a fully configured cMP² will operate on just 256MB of RAM). Disk drivesNoise and vibration are critical – drives with low power consumption and low rotational speeds are best and 2.5” laptop drives close to ideal. They typically consume two watts while reading compared to 12 for a 3.5” drive. Fujitsu’s 300 GB, 4,500 rpm drive uses even less. Capacities up to 500GB is available and is good for ~800 CDs (ripped as uncompressed wav files). SSD is becoming cheaper and offers further advantages by way of no vibrations and very low power consumption. Another option is using eSATA through an adapter that connects to the mobo's SATA ports (2 at a time) and provides eSATA ports on a back plate. This way, an external storage unit can be used and is a better choice when storage requirements exceed 2TB. SoundcardA soundcard usually connects to the CPU via the PCI bus regardless of whether it is a PCI, USB or Ethernet card and, for best results, it needs the lion’s share of access to it. Disk drives connected via these buses compete with it for access and tend to degrade sound. However, many southbridge chips provide separate bandwidth for SATA drives. To use cMP², the soundcard must support ASIO. Beyond that, this manual makes no firm recommendations for a soundcard or DAC: its remit ends once audio data are presented by the music-playing software to the chosen interface. Prototype systems were designed to use SRC upsampling and the soundcard chosen accordingly. Other builders report good results with other techniques (including NOS DACs) but they are not examined here. It is strongly recommended that a good 24/192-capable soundcard is selected with quality drivers and an optical output. This last permits galvanic isolation, which is not possible with other interfaces. As demonstrated in Chapter 3 and despite a bad press, Toslink excels when providing 24/96 output over good cables such as the Audioquest Optilink 5 or the Van Den Hul Optocoupler.
Asynchronous USBWords like reclockers and asynchronous may suggest something new is happening. For enthusiasts, USB Device Class Specification for Audio Devices is an informative read. Although developed in 1998, designers were fully aware of jitter: An essential issue in audio is synchronization of the data streams. Indeed, the smallest artifacts are easily detected by the human ear. Therefore, a robust synchronization scheme on isochronous transfers has been developed and incorporated in the USB Specification. The Audio Device Class definition adheres to this synchronization scheme to transport audio data reliably over the bus.
Isochronous (audio) streaming refers to constant data transfers to/from a USB device. For example an approach is to use a timing interval of 1ms so for 44.1k output, 44 samples are sent per transfer (45 on every 10th). For the isochronous transfer method, the USB specification allows for 3 types of sync: synchronous (like the example), asynchronous and adaptive. Of interest is the asynchronous method (hence the name Asynchronous USB) which allows for an external 'master' XO to be used. USB receiver chips need to cater for this type of sync method. Is this new? Comparing to PCI soundcards, this is not new. For PCI, we have a 'master' XO (internal or external via Word Clock input or S/PDIF input for Juli@) that drives buffer loads through interrupts (PCI read/write transactions). With PCI, 192k streaming is possible (USB is limited to 96k). |
