RETRO-MEASURE: 2002 Lynx L22 PCI audio card in a ~2008 Intel Q6600 Computer. (On "computers are noisy!" for audio)

As many of you probably know, Mitch "Mitchco" Barnett is a regular contributor to the Computer Audiophile website with his articles. He's also the author of the book Accurate Sound Reproduction Using DSP (now available in paperback). A great guy who has logged many hours back in the day in the audio production world!

The cool thing is that Mitch doesn't live far from where I am and last year, we met up at the Vancouver Audio Show 2016 (which by the way appears to be running again this year June 24-25, 2017). We chatted a bit before the show about various topics both in life and audio. From those discussions, one of the questions we wondered about was this whole notion about electrically "noisy" computers and just how "bad" the inside of a typical PC is for high-fidelity computer audio. After all, these days we see all kinds of products claiming the importance of "noise isolation" from USB cards to power supply upgrades.

At the Audio Show, Mitch kindly lent me an old computer audio card he had at home in storage which he was no longer using (in part because he's up to date with his computers and no longer runs a machine with PCI slots :-). Here it is:

Behold, the Lynx L22 24/192-capable PCI "digital audio interface" designed circa 2002. Back in the day, this package had an MSRP of US$749. It filters the power supply, is said to have very clean ground planes, and utilizes high quality components (here's a review from 2003). At a time when probably many of us were rockin' with Creative Soundblaster Live! type sound cards, this is something meant for home studios and maybe some DVD audio production... At 15 years now since the design of this device, we had to wonder - how does this sound card measure and sound sitting in a "noisy" old computer enclosure with components dating back to the era of conventional PCI (not PCIe) slots?

Here's a closer look at the board itself. Notice that it doesn't have any shielding over the components at all. At the heart of the analogue signal conversions are the Cirrus Logic CS4396 DAC (released 1999) and AKM AK5394 ADC (released 2002) among other Xilix Spartan-II FPGA chips, ISSI IS61LF6436A (static RAM), and a Cirrus Logic CS8420 sample rate converter/transceiver.

Notice that the card uses "break-out cables" for the connectors:

The main connector leads to some analogue stereo XLR in/outs. In total it's a good length - about 5.5' long, quite thick and well shielded from the look and feel of it. There are a couple of still unopened XLR-to-RCA adaptors, and another unopened cable which attaches to the smaller D-sub "Sync Port" connector providing AES/EBU S/PDIF in/out and BNC clock in/out (I guess Mitch didn't need to use these cables and features so never opened the plastic!).

This certainly looks like quite a complete "home studio" kind of package for back in the day.

Unlike Mitch, who is quite up-to-date with his computers at home, I still have a machine that my kids use for homework and gaming with PCI card slots to install this device in... Here she is opened up with the Lynx L22 installed:

Old computer build... Yuck! Wires all over and looks like I need to clean off the power supply fan filter! Nice big heat sink and fan for noise control - 105W TDP CPU.

Inside, we have an old Intel Q6600 "Core2 Quad" CPU (2.4GHz, 65nm, 105W TDP, LGA775 socket, released early 2007), 8GB (4 x 2GB) of OCZ DDR2 RAM, Gigabyte GeForce GTX 560Ti PCIe graphics card (1GB GDDR5 RAM, ~2011), and the motherboard is a Gigabyte GA-P35-DS3L (early 2008). The power supply is a Seasonic S12-430HB (switching 430W, ~2005). There's a 4-year old SanDisk 128GB SSD in there for OS and files. In the world of computer technology, clearly these are not "current" components. The "Core 2" CPUs have been discontinued since 2009 but I must say that this old Q6600 keeps chugging away and remains very responsive with most day-to-day software. I think this is a good reminder of how computing power although continuing to increase with each generation, no longer is something most users have to worry about unless one is a gamer or workstation power user.

Notice in the image above, I installed the Lynx audio card adjacent to the nVidia GTX 560Ti graphics card. Furthermore, it's one of these enclosures with the power supply at the bottom so the audio card is close to the switching 430W device as well. The Corsair Carbide SPEC01 ATX case was closed during testing for typical (higher) internal temperatures. Might as well "max out" the electrical and thermal noise in there and see what it looks/sounds like while testing! To make things even "worse", as you can see, I made zero effort to clean up the cable management with power and data cables all over just like in the picture (oohhhh the horror!).

Even though the hardware is old, this machine is running the last Windows 10 update just before the Creators Update and the latest Lynx L22 drivers (May 2016, version 2.0 build 23b WDM & ASIO 2). I also took the liberty to update the card's firmware to the latest "W15" (May 28, 2015). The firmware update and drivers installed without a hitch. Also no problems with audio playback and recording - good job Lynx for the software support after all these years with this product!

Lynx Mixer - very extensive. GUI looks basic by today's standards... Notice the "Analog Out 1+2" can be set as -10dBV or +4dBu output levels.

Set-up:
I'll use my current measurement setup for these tests unless otherwise noted...

Intel Q6600 computer + Lynx L22 card --> Interface XLR cables (~5.5') --> Focusrite Forte ADC --> 6' generic USB --> measurement Windows 10 laptop

The Lynx cable is approximately 5.5' which is a reasonable length to directly connect to the Focusrite's XLR inputs without another patch cable in the way.

Part I: Digital Oscilloscope, Digital Filter, Impulse Response

For the digital oscilloscope, I started with the usual 1kHz 0dBFS 16/44 square wave. However, interestingly, there must be some kind of filtering in place - there was no output with a square wave. Alas I did not have an easily accessible square wave generator on the computer to produce variants like a -3dB or -6dB signal to check. However, a 0dBFS 1kHz sine wave played back without issue or clipping:

That's the -10dBV "Analog Out 1+2" setting in the Mixer which corresponds to the voltage expected of most consumer electronics. With a peak 2.8V, that's ~2Vrms. The alternative professional signal level is the "+4dBu" setting which corresponds to about an 11dB or so gain:

Alas, ~10V is a little "hot" for my Focusrite Forte's XLR input so for the remainder of the tests, I'll stay with the consumer level "-10dBV" setting.

What we can say looking at the sine wave is that the stereo channel balance is looking excellent! Very well balanced, very clean looking with no over/under shoot or DC shift.

Here's the impulse response (16/44.1 signal):

A typical linear phase filter. Absolute polarity maintained, expectation therefore would be a reasonably sharp roll-off.

An orthodox looking "Digital Filter Composite" image. We see that intermodulation (19 & 20kHz sine waves) and imaging are well suppressed. The 0dBFS wideband noise does show intersample overloading (as with most DACs), no problem at -4dBFS - again, all quite normal.

Part II: RightMark Tests

Let's now jump into the usual frequency response, distortion and noise measurements.

16/44:

As usual, 16/44 is really straight forward for essentially any reputable DAC. It would truly take a remarkably incompetent design (or purposely "defective by design" like noisy tube output stage) to mess up 16/44 performance these days. Nonetheless, even among the devices here, you see that there are slight differences such as higher THD especially with the Pi3/HiFiBerry DAC+ Pro. I also have the Squeezebox Transporter RCA measurements. The Lynx L22 obviously performs well though you can see the THD is a little higher than the other two XLR devices offered for comparison - the Oppo BDP-105 and TEAC UD-501. Not a difference one would be able to hear though.

Composite graphs:

The only things that stick out are the slightly higher high frequency crosstalk and obviously higher IMD+N for the HiFiBerry DAC+ Pro which to be fair, at <$100 is way less expensive than the other devices it's being compared to!

24/96:
Now we get into the world of "high resolution" with the 24/96 test signal.

Notice the * for the TEAC UD-501 DAC XLR result. The reason is that I had to apply -3dB to the signal due to the XLR voltage being higher than what the Focusrite Forte ADC was able to handle. This isn't a problem with a 16-bit signal above, but will limit the result at 24-bits. In reality, the noise level is actually more like -113dB for the TEAC using my measurement chain.

The numbers look really good. Notice how good the noise level is for the Lynx L22! Remember, this is an unshielded sound card inside an almost-decade-old computer. Clearly we can see the 2013 Oppo BDP-105 achieving an excellent level of performance. Comparatively, the RCA outputs of the HiFiBerry DAC+ Pro and Transporter are still excellent although again the HiFiBerry has limitations with the crosstalk and IMD+N.

Composite graphs:

Note that the Focusrite Forte has a noise peak up at 37kHz; that's not from the DACs. Can safely ignore that.

The old Lynx L22 is clearly able to hold its own in the objective measurements department compared to excellent, more contemporary products.

24/192:
Finally, for completeness, here's what the measured results look like at 24/192, the maximum sample rate for the Focusrite Forte ADC and for the Lynx L22. The Logitech Transporter isn't capable of 192kHz so it's not on the list.

Again, note the * for the TEAC UD-501 DAC. The true noise level should be down at around -113.6dB or so due to the compensation I had to make to reduce balanced output level.

Composite graphs:

Impressive how flat the frequency response of the Lynx sound card is only dipping below -0.5dB way out at over 50kHz.

Part III: Jitter

Above is plotted the standard 16 and 24-bit J-Test FFT's off the Lynx L22's XLR output. Beautiful! In fact, as I have done on occasion over the last year, here's the 24-bit test accelerated to 96kHz which can enhance any jitter issues:

That FFT is excellent. The Focusrite Forte has a noise peak up at ~37kHz which I suspects intermodulates with the primary signal at 24kHz here to cause the signal around 13.5kHz.

The primary signal is otherwise very clean with no "skirting" around the base and a small symmetrical sideband at +/-8kHz below -130dB. For an example of what high jitter looks like, check out the measurements of the Chromecast Audio from early last year using this 24/96 jitter signal.

Part IV: Let there be LOAD!!!

So far, the measurements are being done with the computer just playing music and otherwise nothing going on in the background. As I mentioned in the introduction, the idea here was to also see if the "generally accepted" idea that computers are noisy devices full of hum and RF interference can be demonstrated and inform us as to how concerned we should be.

Since we're using a Windows 10 multitasking machine, why not hammer the machine and create as much electrical, thermal (and acoustic fan) noise as possible. Let's do some measurements under heavy load.

First, I loaded up the machine with 64-bit Prime95 for a CPU torture test. I used the "balanced" setting which not only runs the CPU at 100% but also creates a lot of memory access. Bottom line is that this did nothing to the RightMark or jitter tests at all.

Then what I did was loaded up FurMark 1.18, an absolutely insane torture test for GPUs to hammer the nVidia GTX 560Ti which not only increases load, but created a ton of heat and the fans were running full tilt as well by 5 minutes. With a maximum power requirement of 170W, the graphics card is even more demanding than the CPU. Check it out...

Here's what RightMark looked like comparing 24/96 and 24/192 results with normal low-load playback vs. 100% CPU & GPU load:

Notice that there was indeed a very small change in the noise level, dynamic range, and deterioration in the crosstalk. Here are the graphs:

As you can see in green, there are a number of new noise peaks while under load, particularly 120Hz, 350Hz, and 600Hz. It's important to remember that we're not looking at huge amounts of distortion though, the peaks are still below -120dB.

How about jitter then since audiophiles seem to always worry about this :-).

As you can see, although increasing CPU and GPU load will induce some small amount of noise and reduce the noise floor of the audio card along with some change in stereo crosstalk, it does nothing to change jitter itself. Clearly, the Lynx L22 has a stable clocking mechanism immune to the effects of the internal computer environment. Is this at all surprising?

Part V: Quick ADC Evaluation

To cap off the measurements, since I have the card installed, remember that this is an ADC device also. I wanted to just have a peek at the Lynx L22's XLR inputs compared to the Focusrite Forte. For some apples-to-apples comparison, I hooked up the PonoPlayer to each ADC using the same phono-to-RCA cable to the same RCA-to-XLR adaptor. Here's the summary data:

As expected, there are variations between the two ADCs. The Forte seems to have a slightly lower overall noise floor (20-20kHz), but crosstalk is a bit higher. Despite differences, notice the intermodulation and harmonic distortion results are very close, as it should be given the same device being tested!

Composite graphs:

Apart from the crosstalk difference which BTW is quite sensitive to interconnect cable used as shown years ago, it's the noise level graph that is the most interesting. Basically, what we see is that the Lynx ADC has picked up a 60Hz AC hum and another spike at 180Hz is evident. One of the great things about the Focusrite Forte is that it can operate with USB power only from the laptop which is why it can be free from the 60Hz hum, and this also keeps my measurement system very compact and mobile.

Notice the concordance of the measured "THD+N Sweep" graph between the Focusrite and Lynx. Over the years I've found this pattern to be quite unique and reproducible with various ADCs for each DAC tested almost like a unique "fingerprint" for the audio device.

As above in the "Let there be LOAD!!!" section, I wondered, would the noise floor of the Lynx ADC be affected by pushing the CPU and GPU to 100% and increasing the electrical noise of the computer?

Apologies that the overlay wasn't perfect with the screen capture a few pixels off; but I think the difference is obvious. As you can see, running Prime 95 and FurMark to strain the computer system does increase the noise floor particularly in the lower frequencies. The electrical noise (essentially from the GPU) seems to affect the frequency range between 120Hz to 1kHz mainly. Again, like with the DAC playback during heavy computing load, the extra noise is not of high amplitude and these frequency spikes individually remain significantly below -120dB. 

BTW there's also a spurious very high frequency noise spike up at 75kHz with the L22 (better than the Focusrite Forte which has a spike at 37kHz unfortunately; not that either would be audible in actual use of course).

One last comment is about the nice flat noise floor from the AK5394 ADC chip in the L22. Many ADCs including the Focusrite Forte have increasing noise floor going up 30dB or so towards 96kHz due to noise shaping as can be seen in the RightMark 24/192 "Noise Level" graph above. Not so with the AKM AK5394(A) which makes this device an excellent base for a measurement ADC. A limitation however of the L22 is that there are no built-in clean preamps to provide gain to an input signal for easy calibration/leveling.

Part VI: Subjective Listening

You might be wondering... What do you subjectively think about the sound of this audio card then!? Glad you asked :-).

One evening after installing the audio card and made sure it worked, but before I got the measurements done, I lugged the computer downstairs to my sound room for an audition of this computer/card hooked up via XLR to my Emotiva XSP-1 preamp --> Emotiva XPA-1L monoblocks --> Paradigm Signature S8 speakers.

It was easy to hook-up with an extended length of XLR patch cable and I had a small spare LCD monitor so I didn't have to hook it up to the big-screen TV. I placed the computer behind an acoustic panel to reduce the fan noise. A good quality recording like Vilde Frang's Sibelius, Prokofiev: Violin Concertos (2009, DR15) sounded beautifully articulate and dynamic. Excellent rendition of details. I love the Sibelius Violin Concerto in D minor on this disk. Not that I specifically pay attention to it, but for evaluation of detail, one can make out the fingering and changes in positioning of the violinist while playing as well as slight noises made by the orchestral members.

A more rock/pop oriented recording like Graham Parker's Squeezing Out Sparks (1979, 1996 remaster, DR10) was appropriately "aggressive". The standout track on this album for me has always been "Passion Is No Ordinary Word". It's not the best recording but the tonal qualities of the L22 were excellent, and there was certainly no issue with this sound card in terms of bass quality and quantity. The drums had a nice "slam" to them with great transient control through the system. Certainly sounds very comparable to my Logitech Transporter and TEAC UD-501 DAC, both hooked up to the same pre-amp.

Modern highly processed pop/rock/synth/distorted guitar work like Joe Satriani's latest Shockwave Supernova (2015, DR9) sounded appropriately "processed" with the use of soundstage-expanding DSP, and dynamic range compression :-(. Good hooks, nice rhythms; "easy listening" for the guy who doesn't mind slightly heavier and more distorted rock instrumentation. On that point about the distortion, I seriously wish this record was better engineered. I tell ya, rock instrumentals need to have dynamic headroom; as it stands, it sound OK but I just get this feeling of troubling disappointment thinking that if only they had given the instruments room to breathe, it'd sound a lot less muddy and a lot more punchy with the volume cracked up on a high-fidelity sound system especially with a powerful sub connected!

Bottom line is that this audio card sounds great 15 years after it was designed. Nothing "veiled", or "dull" or "colored" about the sound. I certainly would not be able to detect that it came from an internal sound card if I didn't know about it! What is a hassle of course is that it is plugged into a large computer motherboard, uses an old PCI form factor, and a computer of this vintage typically isn't as power efficient and demands cooling fans which of course creates acoustic background noise. Thankfully this computer is not too loud, but it's still noticeable even behind the acoustic panel and it's this acoustic noise that is clearly worse than any electrical noise measured...

Part VII: Conclusions

Okay then, let's summarize what we have learned in looking at the Lynx L22 "old-skool" PCI audio interface card designed back in 2002 running inside an elderly (by tech standards) Intel Q6600 Core2 Quad computer that's close to a decade old...

1. As a DAC, the Lynx L22 with its Cirrus Logic CS4396 performs very well. I remember back in the day reading about this CS4396 chip with complements made to the sound quality. Objectively, this sound card is capable of ~18.5-bits of resolution through the XLR analogue output at the consumer -10dBV level. I'm sure the sound would be considered excellent back in the early 2000's and certainly still excellent now even though a few more noise-floor dB's can be squeezed out these days with some of the better DACs on the market.

2. As an ADC, this card performs great. Nice to see the close cross-correlation of measurements using my Focusrite Forte compared with this Lynx card in measuring the PonoPlayer with RightMark. In normal operations, the noise floor is very low and remarkably flat all the way to 96kHz when recording at the 192kHz sampling rate except for that unfortunate noise spike around 75kHz (-120dB, not of any concern). Remember, this was an ~US$700 pro-level sound card back in the day made by a well regarded pro audio company so this level of quality is to be expected.

3. Yet again, don't worry about jitter. Aren't people getting tired of companies dragging out the jitter "boogeyman" after all these years yet? Even in this "retro", "legacy" PCI card, clearly jitter is not an issue. I even accelerated that 24-bit J-Test to 96kHz speed and this device took it like a champ. Impressive. As I've said before, I don't think jitter is audible unless very extreme. The only time these days where it's worth measuring jitter is when using S/PDIF coaxial and TosLink, or with HDMI simply because these are the interfaces more prone to timing irregularities. I don't believe there's any point worrying about jitter with reasonably designed asynchronous USB or networked communications (ethernet, WiFi) these days. I'd certainly be very cautious about listening to people who claim they can "hear jitter" unless they have some measurements to confirm the veracity of such testimony!

4. Yes, high CPU and especially GPU loads can increase noise in the machine. For a device like the L22 audio card, it does translate to higher noise floor when functioning as DAC and ADC. However, as you can see, even with a legacy computer system, the noise is handled very well considering it's inside the computer and doesn't even need a metal shield around the components. I suspect to a large extent it is because of the use of balanced circuitry and cabling. I'll see about measuring a more pedestrian sound card at some point (I have an old Soundblaster Live! somewhere around here). Clearly, this sound card blows away the on-board audio from last year's Gigabyte GA-Z170X-Gaming 7 motherboard.

By the way, I see that some people like to use a powerful GPU to perform upsampling with HQPlayer. Not that I think it'll be a problem, but considering that the electrical noise I'm finding is generally when I start putting load on the GPU card, perhaps it's worth putting some thought into this realization for those who seem to worry about computer noise in general. Even though modern nVidia CUDA graphics cards are more efficient for the speed, a GTX 1080 can suck up to 180W, GTX 1070 up to 150W and GTX 1060 up to 120W.

In any event, this is demonstration that even 15 years ago, pro audio engineers were able to handle the electrical noise inside a multi-purpose computer without needing to resort to any special audiophile brouhaha. With modern machines, obviously the job can be done easier and cheaper.

5. Nope, CPU and GPU loads do not contribute to higher jitter. Computing load certainly did not create any more jitter for a PCI internal sound card designed in 2002! I remain perplexed as to why some people think that having more CPU threads or background services somehow results in "bad" sound. Seriously folks, these days if you've got bit-perfect output (ASIO, WASAPI) to an asynchronous USB DAC, there is no jitter to worry about unless there's something wrong with your DAC! OS tweaks and "optimization" apps like Fidelizer or AudiophileOptimizer will achieve nothing. Likewise, player software makes no difference as long as it's set up properly. Yes, I know people have provided testimony about how great certain optimizing software works or feel that a certain bitperfect player is better than another. So what? People testify about all kinds of things all the time which we safely ignore in day to day life!

Remember that way back when (2013), I did a video about "the hunt for load induced jitter" :-). Didn't find an issue then, and still finding nothing now.

Well, looking back now to the early 2000's, a 33.33MHz synchronous "conventional" 32-bit PCI bus might seem "slow" by today's standard (the slowest single-lane PCIe card these days provides twice the data transfer). But remember that for audio, an old PCI architecture still provided up to 133MB/s speed; way more than fast enough for stereo audio up to 24/192 as with this device. Furthermore, this was a ~US$700 card with great components for the day, and pro-level functions, so of course we should expect excellent sound quality.

What is more important from these tests is a reminder that sound quality in the form of technical advancement in fidelity has obviously not kept pace with development in so many other ways. Machines are smaller, faster, more efficient, lighter, have better battery life, pack more features as the years go on. But ultimately, even though I can show that the Oppo BDP-105 might be extending the noise floor lower compared to this Lynx audio card, is it that important any more? The obvious answer is no, because no matter how great a DAC is, ultimately the sound will need to be transduced (speakers, headphones), interact with your sound room, and finally confront the limitations of one's own hearing/perceiving ability. The Lynx L22 is a great reminder that very high quality products have been available for a long time now; even in devices like this which might not be recognized by "hardware audiophiles" who may have preconceptions and worries about bad "computer noise" and "jitter". Sadly, I think over the years, these beliefs have been perpetuated by the typical audiophile press who have never bothered to check these beliefs and provide proper education to audiophiles at large.

A big thank you again to Mitch for letting me borrow this audio card to play with! Looking forward to meeting & catching up. Will get the Lynx card back to you then :-).

----------------------------

I believe it is AXPONA this weekend so it'll be interesting to hear about what's new.

In the world of Windows 10, we now have the Creators Update (build 1703) rolling out. I've updated my main workstation and HTPC already without issues. I have also updated my nVidia GTX 1080 driver to go along which now allows HDR10 on my Vizio P75 TV as Windows desktop... Can view HDR YouTube now although the colors are still a little washed out - back to SDR until a few things get sorted I think!

Along with the HDR video update, finally Windows has native USB Audio Class 2 drivers support (like OS X and Linux have had for years). Will have a look/listen to this in the coming week.

Have a great week ahead everyone... Hope you're all enjoying the music!