Project Hammerfall is an unusual name for a soundcard — but then RME's new card is somewhat out of the ordinary. Martin Walker enters a world of sample‑accurate sync and minimal latency.
The ADAT digital interface standard has is now in widespread use on a far greater variety of systems than Alesis's original tape‑based digital multitrackers. Most digital mixers, for instance, provide digital I/O in this format, and it's now possible to set up a very powerful recording system which uses ADAT interfacing to combine the mixing facilities and analogue I/O of a digital desk with the recording, sequencing, editing and processing capabilities of a modern computer.
Indeed, many of those musicians who actually own ADAT recorders must be looking enviously at the facilities provided by the latest computer‑based audio sequencers. The ability to see the waveforms and edit them right down to sample accuracy, as is possible with such graphical random‑access systems, makes the process of editing a lot easier. However, ADAT machines still have their own virtues — in particular, they are very easy to use, especially to those familiar with analogue multitrackers. So, rather than sell them and move to a wholly computer‑based system, many existing ADAT owners are likely to consider running them in parallel to gain the best of both worlds — a straightforward and reliable tape‑based recording and playback system synchronised to a computer supporting MIDI and audio tracks.
To link either an ADAT or an ADAT‑format digital mixer to a computer, you need a computer soundcard with an ADAT‑compatible interface. SOS has looked at several such cards before: the Alesis ADAT Edit reviewed in last month's issue supports a single eight‑channel ADAT interface, while the Sonorus StudI/O reviewed in the September '98 issue supports two. Like the MOTU 2408 reviewed in February '99, however, the RME DIGI 9652 (aka Project Hammerfall) under review here can support up to three ADAT interfaces simultaneously. Another of its main selling points is its comparatively low price — for £499, you effectively get an interface ideal for connecting your computer to up to 24 ins and 24 outs of multitrack digital recorder or mixing desk, along with a single stereo digital in and out for mixdown use. The MOTU system, by comparison, costs £999, though its I/O is mounted in a separate breakout box and includes eight analogue inputs and outputs with 20‑bit converters (Project Hammerfall has no analogue I/O at all).
The 9652 is also the first soundcard (to my knowledge) to be released with ASIO drivers that conform to the new ASIO 2.0 specification, as implemented in Steinberg's new Cubase VST version 3.7 for Windows (see review on page 160). This provides 'zero'‑latency monitoring inside Cubase VST by automatically routing the incoming audio signal to the equivalent hardware output during recording, completely bypassing any delays caused by software processing. However, even without this feature, the Hammerfall drivers still provide extremely low latency, down to a very impressive 3mS given a powerful enough host computer. In addition, it can transfer digital data to and from ADAT machines with sample‑accurate sync, and is even capable of 24‑bit 96kHz operation with other digital devices such as DVD.
RME may be a new name to some SOS readers, but the Germany‑based company has been quietly making a name for itself with the high quality DIGI range of soundcards, now distributed in the UK by Digital Media. In fact, readers with good memories may remember me mentioning the excellent selection of audio test files available from the download page of the RME web site in my March '98 PC Notes column. I still use some of these (particularly the 24‑bit ones) as part of my test procedure for each and every soundcard I review for SOS.
The RME '96' range includes the S/PDIF‑only DIGI 96 soundcard (£299), the DIGI 96/8 with added ADAT I/O (£349), the DIGI 96/8 PRO with a 20‑bit D‑A converter (£399), and the DIGI 9/8 PAD with both A‑D and D‑A converters (£449). The DIGI 9652 reviewed here is the latest and greatest card in the range, and provides three 8‑channel ADAT optical inputs and three 8‑channel ADAT optical outputs, along with a stereo S/PDIF input and output, word clock input and output, and a 9‑pin ADAT sync input.
All of the 9652's digital audio inputs and outputs are capable of 24‑bit 96kHz operation, albeit by a circuitous route. Since the ADAT optical spec only supports sampling frequencies of up to 48kHz, RME use what they term 'sample split', and distribute the doubled bandwidth for 88.2kHz and 96kHz sample rates between two channels, giving a halved channel count of 12 inputs and 12 outputs along with the same stereo S/PDIF in and out. The name of the card is derived from this 96kHz capability, and the fact that the total number of digital audio channels comes to 52 when running at a sample rate of 48kHz or less.
Project Hammerfall is one of the smallest PCI cards I've seen, and at just five inches long should fit in any PC or Mac without a shoehorn. The backplate contains four Toslink optical connectors (for ADAT 1 Input and Output and ADAT 2 Input and Output), along with a 9‑way D‑type connector. To this you attach a short adapter cable which terminates in a 9‑pin flying socket for ADAT Sync In and a pair of in‑line phono sockets for S/PDIF In and Out.
However, as with many recent soundcards, a single backplate is just not enough to house all the required socketry, so RME have placed the remainder on the 'Expansion Board' — a stand‑alone backplate (complete with a small circuit board) that you can fit instead of a blanking plate in front of any empty PCI or ISA card slot. This houses the third and final ADAT 3 Input and Output, along with a pair of BNC sockets for word clock In and Out. Next to these is a useful green LED which lights up when the word clock input has locked to a valid signal. You don't have to install this Expansion board: if you are short of space and happy with the twin ADAT support of the main board, it will happily run by itself.
The main card and expansion board are connected using a supplied three‑inch‑long ribbon cable. RME obviously expect you to place the two next to each other, but I just managed to piggyback one intermediate card to fit the expansion board next to a vacant slot. In a well populated machine you might need to shuffle your existing cards unless RME decide to splash out on a few more inches of cable.
With latencies of 6mS or lower now possible with many modern PCs, the whole issue of latency begins to take on less importance, and if you have a Pentium III machine capable of 3mS latency, you may not even be aware of it at all!
Installation proved to be quite simple in Windows 98, with the usual plug‑and‑play routine of inserting the floppy driver disk when asked. As with the drivers for most professional cards, the actual file sizes are tiny — five files with a total size of 100Kb. It only took me a few minutes to arrive at the desktop along with the extra Hammer icon on my Taskbar to launch the Settings utility. However, when I tried installation in my Windows 95 partition I had various problems getting the card to be recognised, as well as several system‑wide crashes afterwards. RME subsequently confirmed that Windows 95 can't handle the memory requirements of their card, and that the drivers are specifically written for use in Windows 98. I suspect that this may be the start of a new trend.
On reading the User's Guide, you could almost believe that RME have been religiously reading my SOS writings over the last few months. Not only do they provide specifications for jitter (see Brief Specifications box), but they also have a signal‑flow chart showing how the card works, which makes understanding the options much easier. The S/PDIF input and output can both be switched between co‑axial and ADAT (optical) sockets, and the input can also be switched to 'Internal', which allows you to connect the digital audio output from an internal CD‑ROM drive to the 2‑pin connector provided on the soundcard.
The S/PDIF switching is carried out using the 9652 Settings utility shown opposite, and the S/PDIF Output is even more versatile than at first appears. Its co‑axial socket is always connected, but if you tick the 'ADAT 1' box you switch the ADAT 1 optical output from its normal 8‑channel function to another S/PDIF output in parallel with the co‑axial one. The Emphasis box needs to be ticked if you are outputting any recordings that were originally made using 50/15 microseconds pre‑emphasis. This is because the S/PDIF output header information is created from scratch, losing any previously set bits on input signals.
The 'Non‑Audio' tick box should be used when playing back Dolby AC3 signals, since many external decoders such as those in surround sound receivers or digital TVs won't otherwise recognise the signal as AC3. The fourth tick box is labelled 'Professional', and doubles the output voltage levels (from 0.8 to 1.6 Volts), making them suitable for AES‑EBU operation using the co‑axial output with a suitable adapter lead ending in a male XLR plug. The co‑axial input will directly accept the higher levels from an AES‑EBU adapter lead with a female XLR plug. Clever stuff!
When working with any digital audio card, getting the correct clock settings is vital, and any visual feedback here is extremely useful. As I mentioned earlier, a green LED next to the word‑clock input lights up as soon as a valid input signal is detected, which is very helpful. In addition, the Settings utility shows the Sync Check status of all three optical inputs (ADAT 1, ADAT 2, ADAT 3), and the S/PDIF In: 'Lock' and 'No Lock' indicate the presence or absence of a valid input signal, and 'Sync' shows that a valid synchronous input signal is present. The Settings utility also displays under Time Code the information received using the ADAT Sync In socket, so that you can check this against the display on the ADAT machine itself.
Clock Mode determines how everything syncs up. If you have a valid word clock input (such that the green LED is lit) then you can switch to this as the system reference by selecting 'Word Clock', or you can use the 'Master' setting to force the 9652 to run from its internal master clock, in which case all other devices will have to slave to it (either using the embedded clock in the digital outputs, or the dedicated word clock output). However, the easiest option is to select 'AutoSync'. In this mode the card runs by default from its internal clock, but also scans the digital inputs for a valid clock signal. If one is found at any time it switches to this automatically and becomes the slave, allowing 'on the fly' recording. If this happens the Sync Reference is shown in the utility, along with its current sample rate. This can theoretically be anywhere between 25kHz and 105kHz, although I didn't manage to confirm this. If you have more than one input providing a valid sync reference, you can tick a different box in the 'Pref Sync Ref' section to force the issue.
The Audio Buffer size can be adjusted over a wide range to determine the latency value (see the Size Isn't Everything box), but the beauty of the Hammerfall Settings utility is that you can make any adjustment in real time. You don't need to click on any OK button for a change to take effect, and you don't even need to stop recording or playback, even when changing the latency setting.
Since the 9652 has no analogue I/O, no comments can be made concerning its analogue audio quality, or any measurements of background noise. However, I did have the opportunity to try it out with RME's ADI‑8 Pro Interface, and this gave impressive results, as one might expect from converters with a dynamic rane exceeding 110dBA (see the Partners In Sync box). The 9652 has been initially released with ASIO drivers only, and as such is only currently compatible with those MIDI + Audio sequencers that support this standard, such as Steinberg's Cubase VST and Emagic's Logic Audio.
The ASIO inputs are labelled DIGI9652 1 to 26, and the outputs are arranged in stereo pairs labelled DIGI9652 1, 3, 5 and so on. The first eight channels correspond to ADAT 1, while 9 to 16 are ADAT 2, 16 to 24 are ADAT 3, and finally 26 and 26 belong to the S/PDIF I/O. The S/PDIF connectors are colour‑coded, but it turned out that the preliminary user guide is lying when it claims that the red socket is the output — in fact it's the input. Once I discovered this, everything burst into life.
The Enhanced zero‑latency monitoring provided by ASIO 2.0 and the RME drivers was a real treat: those who select 'Global Disable' in the Monitoring choices of Cubase and arrange your monitoring externally using a hardware mixer will be able to safely return to the Cubase options. 'Record Enable Type' lets you hear the input as soon as the channel is record‑enabled, while 'Tape Type' only lets you hear the input when in Stop or Record mode, but not while playing back. You can't control the monitor level or pan position inside Cubase with the Hammerfall card, but being able to switch the hardware from Cubase transparently is still a major leap forward, and certainly beats having to set up separate soundcard utilities!
However, a little of its thunder is stolen by the low‑latency option. With a suitably powerful PC, a latency of 6mS or even 3mS is possible (see the Size Isn't Everything box), and this made for a very immediate response both when monitoring an audio performance and for playing VST Instruments. I expect latencies of this order to become a prime requirement for many of next year's soundcards. With my Pentium II 300MHz PC I managed to monitor my input signals with 6mS latency, while listening to the input signal with added effects from Waves' Trueverb as a channel effect, plus a Waves REQ 6 'analogue' EQ and RCL compressor running as inserts.
Although some people might grumble about the initial lack of MME drivers, you can see why RME have concentrated on ASIO performance. With latencies of 6mS or lower now possible with many modern PCs, the whole issue of latency begins to take on less importance, and if you have a Pentium III machine capable of 3mS latency, you may not even be aware of it at all!
To ensure sample‑accurate positioning when transferring audio digitally between devices you need two levels of sync — sample rate (word clock) and sample position (time code). I connected a pair of optical cables between Hammerfall and an ADAT machine, and then set ADAT 1 as the sync reference using the Hammerfall Settings utility (using AutoSync mode to ensure that Hammerfall became slaved to the embedded word clock signal coming down the ADAT optical cable). Timecode sync was then established by connecting the 9‑pin ADAT Sync Out to the Hammerfall Sync In.
Cubase VST 3.7 needs its Timecode Base set to ASIO 2.0, and the Sync button selected on the Cubase Transport Bar. Another setting to check is that your song's sample rate is also set to the same value as that of the ADAT; otherwise Cubase may intermittently drop in and out of sync. With these settings in place, transferring tracks to and from ADAT proved very easy and gave no problems, with the Time Code being correctly displayed in both Cubase and the Hammerfall utility. Being able to monitor with zero latency also helped during these transfers, and I was impressed by the rapid lock‑up time of no more than a second — it can often take three or four seconds with some gear combinations.
The new ASIO 2.0 positioning protocol worked well, and essentially the ADAT and Cubase tracks became a seamless system with sample‑accurate sync between the two sets of tracks. The system must be controlled from the ADAT transport controls, but if you have a suitable conversion box, it's also possible to select 'ASIO 2.0 with MMC' as Timecode Base, and then you can use the Cubase transport bar controls instead. However, I doubt that most musicians will consider having to use the ADAT controls a disadvantage, since a dedicated set of tactile buttons is nearly always preferable to clicking with a mouse.
The ADAT 8‑channel optical interface has grown into a versatile general‑purpose interface used not only by ADATs but by A‑D and D‑A converters, digital mixers and other devices. This ensures that the Project Hammerfall card can remain a versatile nerve centre for your studio even if you stop using ADAT recorders in the future, or don't intend to use them at all.
The only other soundcard to my knowledge to currently have ASIO 2.0 drivers is the StudI/O from Sonorus; with its Backplate option you also get sample‑accurate sync, but this only provides support for two ADAT machines rather than three. However, RME are promising two‑card support in future driver updates, and judging by the driver improvements made even during the review period, I fully expect this to happen. The ADAT Sync Input on the second card would of course be redundant, but with a powerful enough computer the combination of two Hammerfall cards would then support six ADAT interfaces for 48‑in/48‑out operation, along with two stereo channels of S/PDIF.
As with any other soundcard, you may need to experiment a little to find the best value for the Hammerfall driver latency to suit your computer, but this is less important with the ASIO 2.0 enhanced 'zero'‑latency option. However, having feasible latency values down to 3mS also makes this an excellent choice for those interested in a multitrack soundcard that can run some of the latest breed of 'real‑time' ReWire and VST Instruments. This is a versatile and impressive product, and at £499 should attract many users.
For those without ADAT machines who want a suitable set of A‑D and D‑A converters to use with Project Hammerfall, one ideal partner would be RME's own ADI‑8 Pro Analog/Digital interface. This has eight analogue ins and outs on quarter‑inch TRS balanced/unbalanced jack sockets, along with a comprehensive set of digital I/O. This includes two pairs of ADAT optical sockets and two pairs of TDIF sockets, along with word clock I/O. Its converters are all 24‑bit, with a quoted 114dBA dynamic range on the A‑D side, and 110dBA on the D‑A. It supports sample rates of 44.1 and 48kHz, and when synced to word clock can lock to sample rates between 27kHz to 57kHz.
However, for those with 16‑bit ADAT machines who aspire to 24‑bit recording, the ADI‑8 Pro's Bit Split function allows a single 24‑bit signal to be separated into two 16‑bit ones. This allows two ADAT machines to record eight channels at 24‑bit resolution. The corresponding Bit Combine function lets you reassemble the 16‑bit digital input signals back into a single 24‑bit output. Apparently the method is compatible with the Yamaha 02R mixing desk, so that the ADI‑8 Pro can be used with it to provide 24‑bit operation. Yet another useful feature is Copy Mode, which allows digital input signals to be simultaneously sent to the TDIF and ADAT outputs to provide real‑time ADAT‑TDIF and TDIF‑ADAT conversion, along with digital patchbay and signal distribution facilities. As always, such comprehensive facilities and high audio quality don't come cheap, but at £999 this should still find many enthusiastic purchasers.
Where general circuit design is concerned, RME have seemingly left no stones unturned in their quest for glitch‑free audio. Their PCI buss‑master interface is claimed to reach transfer rates of up to 130Mb/second for recording and playback, and in case other PCI devices grab too much of the buss bandwidth, a 96Kb 'Burst FIFO' buffer prevents dropouts. When transferring the maximum of 52 tracks, PCI bus load is claimed to be an extremely low 9 percent. In addition, Hammerfall uses a straight 'double‑buffer' ASIO design, which handles data transfers without involving the main computer CPU at all, instead using its own PCI DMA controller to move audio between the card and Cubase. Even when transferring the maximum 52 tracks there is 0 percent CPU overhead. Apparently other buss‑master cards still need to massage the data format slightly to interface with the Cubase ASIO requirements, which takes a small amount of CPU overhead.
The ASIO drivers operate with a 32‑bit (4‑byte) transfer mode for each mono channel (like the Yamaha cards), since this is more efficient than the packed 3‑byte mode when dealing with 24‑bit data. The forthcoming MME drivers will also work in this mode for 20‑bit and 24‑bit use, and the standard 16‑bit (2‑byte) mode for 16‑bit use.
When lots of audio channels are used in a single application, some previous soundcards have had problems with them starting out of sync, or even stereo pairs getting occasionally swapped. The SyncAlign feature of the 9652 makes sure that however many channels are being simultaneously recorded and played back, they all remain sample‑aligned. The Enhanced Sample Buffer acts similarly to a sample‑rate converter, and can be used to overcome otherwise insurmountable problems such as recording the digital audio output of a CD‑ROM player when the card is sync locked to an ADAT machine. Although RME don't recommend such combinations for general use, it's good to know that you could probably get away with it in an emergency.
One of the main claims to fame of the 9652 card is that the supplied ASIO drivers for Windows 98 can have their latency adjusted over a wide range, from 186mS down to a jaw‑dropping 1.5mS! They accomplish this in just the same way as any other ASIO driver, by changing the size of the buffers, but since some people may not yet have realised the direct link, let me explain further. With a fairly typical buffer size of 1024 samples (or 1K), the delay when running at 44.1kHz works out to be 1/44100 times 1024, or 23.2mS (or 21.3mS at 48kHz). When the buffer size is made smaller, it needs to be filled and emptied more often, but the latency figure drops accordingly. So the latency figures for any ASIO driver depend solely on the available choices for buffer size.
Most drivers provide a small selection of options for buffer size, and the more powerful your computer, the smaller you can use, with a corresponding drop in latency. Once you go below a certain point on any machine, you will start to get occasional glitches and clicks, and you should then increase the buffer size until they go away — this is the optimum setting for your particular machine. So, the fact that RME provide options down to a buffer size of 64 samples doesn't mean that you will automatically get 1.5mS latency on your machine. In practice this is just not possible with today's computers — but my Pentium II 300MHz PC managed 6mS comfortably, and I know that Pentium III machines have managed 3mS.
However, although the Hammerfall design takes no CPU overhead (see Tech Talk box) to move the data about, the smaller the buffer size the more interrupts (IRQs) occur. This will still show up as a rise in the Cubase CPU Performance meter when using very small buffer sizes. Sadly the current version 4.0 of Emagic's Logic Audio can't use settings below 1024 bytes (23mS) without crackling, due to its internal buffer design.
- Digital Audio Inputs: 3x Toslink ADAT‑compatible, 1x stereo S/PDIF phono.
- Digital Audio Outputs: 3x Toslink ADAT‑compatible, 1x stereo S/PDIF phono.
- Sync Inputs: word clock, 9‑pin ADAT Sync.
- Sync Outputs: word clock.
- Supported sample rates: 32kHz, 44,1kHz, 48kHz, 64kHz, 88.2kHz, 96kHz and variable (word clock).
- Sample split mode: 12 channels of recording/playback at 24‑bit/96kHz.
- ASIO design: 0 percent CPU load when running all 52 channels.
- Zero Latency Monitoring: Hardware bypass per track, controlled by Punch‑In/Out.
- Settings: all adjustable in real time.
- Jitter: <1nS in PLL mode (44.1kHz, optical in), <2nS in PLL mode (44.1kHz, ADAT optical in).
- Good value for the quantity of I/Os provided.
- Very low‑latency Windows 98 ASIO drivers.
- Sample‑accurate sync
- Enhanced 'zero'‑latency monitoring.
- Not supported in Windows 95.
- No Mac, NT or MME drivers with launch package.
- No 9‑pin sync output.
A well designed and versatile soundcard with a powerful set of features that could form the nerve centre of a 24‑track computer‑based system.