Gramophone Dreams #74: Elekit TU-8900 kit amplifier Page 2

Feedback issues
I am glad I listened for five straight months without disconnecting the TU-8900's feedback. During that time, I never forgot the feedback was connected, but everything sounded so alive and energized that I never felt motivated to remove it—never, that is, until I began to write this report. When I finally took the feedback out, I was surprised how much rawer the sound became. Without feedback, a deep-and-silent, wet-look transparency appeared. The amp actually got quieter into my 97dB/2.83V/m– sensitive Heretic AD614s.

I found these new, raw, uncorrected sonics quite appealing, especially in the upper octaves and especially on flute, harp, and violin. But at first, despite some effort, I could not grasp what effect the absence of feedback was having on the lower octaves.

I thought my favorite piano recordings would make it obvious, but they didn't. Imagine a hammer hitting a low-octave piano string. A wavefront is produced by that string, followed by a delayed reaction from the sound board; these sound energies combine to produce a complex wavefront, the time alignment of which is critical for full-pleasure listening. Only the best amplifier-speaker pairings get this wavefront alignment right. Push-pull amps need feedback to even get close to preserving this complex phenomenon. Single-ended amps without feedback can capture it, but too often their puny damping factors round the leading edge of bass transients and mute reverb crispness. This is why many single-ended triode amp users like a little "touch-up" feedback to tighten up the bass and make it more tuneful. My Line Magnetic LM-518IA and Elekit's TU-8600S both use such "touch-up" feedback to keep sounds taut and moving fast.

That's why I felt good about the feedback in both the TU-8600S and TU-8900 amps. If you're gonna have solid state rectification, which, generally speaking, tightens and hardens the sound of tube amps, and fixed bias, which has a similar effect, why not wire in a few dB of global feedback to assist the amp in driving lower impedances? I've always felt that these three design choices steered the Elekit 300B amps toward a fresher, crisper, brighter, more contemporary sound than their softer-sounding no-feedback cathode-bias counterparts.

When I first removed the feedback from the TU-8900, I was driving the Falcon Gold Badges. Suddenly they sounded—I can only describe it as weirdly different: inexplicably grainless with a possibly too-dark deep-space transparency that made images look cinematic: very Blade Runner 1982.

I kept listening for chamfered transients or limp bass, but the low octaves seemed taut as ever. Then, a few records later, that darkness I first perceived began changing to brightness and light—sometimes even a slight glare, depending on the recording. The dominant sonic characteristic of the no-feedback TU-8900 was this raw-file, no-glass transparency, which felt unmitigated in some kind of extreme way. A week passed, and the bass was still tight on my Falcons, but I still wasn't sure if I wanted feedback in or out.

And then I realized: What I was experiencing—possibly for the first time ever—was feedback-less 300Bs in concert with the dramatic clarifying effects of amorphous core output transformers combined with the enhanced small-signal conductivity of those silver-foil caps and tantalum resistors.

Sequence is everything (again)
When I replaced the tiny 83dB/2.83V/m–sensitive, 15 ohm Falcons with the barrel-chested, 8 ohm, 97dB/2.83V/m–sensitive, $7290/pair Heretic AD614 coaxials, the no-feedback 300B Elekit experience became more physical and less dynamically restrained. The projected soundspace went from deep and tall to shallow and wide. (I picture speakers as camera lenses, with different speakers generating soundspace effects analogous to changes in aperture and focal length.) The Heretics are sharp-focused but with a shallow depth of field. The bulk of their stout cabinets appears to flatten and broaden the soundstage and reduce the experience of three-dimensionality. Nevertheless, the larger-scale images and the intensity of the Heretics' energy delivery is something I've needed, something I've craved emotionally. For my listening pleasure, I am willing to sacrifice some soundstage depth for image density and what feels like electrified intensity, with singers and instruments presented closer to full size.

The river was whisky and I was a divin' duck ...
I am attracted to the Heretic speakers because nothing else I've been using plays artists like R. L. Burnside in a manner that feels strong enough, textured enough, and tangible enough for this kind of raw musical art. Burnside's 1998 anthology The King of Hill Country Blues: Rollin' & Tumblin' (16/44.1 FLAC, Wolf Records/Tidal) needs big paper cones, fat boxes (and possibly multicell horns) to properly launch his voice and rough-textured guitar into the room.

Without feedback in the TU-8900, the Heretic's 12" paper cones fed me more corporeal energy. As a result, everything felt more impactful, more emotionally fleshed out. Every Elekit 300B virtue was enhanced, especially texture and transparency.

Western Electric's 300B was used primarily in audio-frequency amplifiers in commercial applications and as a pass tube in regulated power supplies. The 2A3 was the premier console-radio tube of the 1930s and '40s, famous for its clarity and the naturalness of voice reproduction. The Cossor LinLai WE300Bs were my preferred 300B for the TU-8900, but every time I tried the LinLai 2A3s, I'd end up switching to the richer, earthier-sounding 1940s RCA 2A3s (which I sorta collect). In the TU-8900, the RCAs generated more realistic tone and denser body.

In my system, driving the Heretic AD614s, the no-feedback Elekit TU-8900 with the vintage RCA 2A3s excelled at playing Gregorian chant recordings. Playing my beloved "Mare Nostrum" Orient – Occident: Dialogues, a 2011 recording by Jordi Savall, Montserrat Figueras, and Hespèrion XXI (16/44.1 FLAC, Alia Vox/Qubuz), in which Sephardic, Turkish, and Christian chants overlap and complement each other, the sound via 2A3s showcased a coercive intimacy coupled with startling jump factor. I enjoyed how the 2A3 Elekit–Heretic combo played this record in a manner that exposed every drop of Jordi Savall's and Montserrat Figueras's beauty and talent as well as the recording's myriad sonic wonders.

The headphone output
While I'm turning you on to a supervalue SET amp, I may as well give you a quick introduction to a supervalue headphone that I am in the process of reviewing. It's a new model from HiFiMan that, before I knew its price, I assumed was in the $5000 range because it looks expensive and dances sonically with the best: the $1999 Audivina. The 20 ohm, 97dB/mW Audivina is a wood-shelled closed-back with a "NEO supernano" planar-magnetic diaphragm that played crazy clear when powered by the TU-8900's front-panel headphone jack.

I chose the Audivina for this report because I wanted to see how the TU-8900 would drive low impedances and because it was the headphone I'd been auditioning when I started this report.

My ears scanning for problems, I used the Audivina closed-backs to listen to Danza Española played by Andrés Segovia (16/44.1 FLAC, Masar/Tidal). More than with XIAudio's Broadway or HeadAmp's GS-X mini, the sound coming through was vivid and lustrous in a lush, microfocused manner reminiscent of the TAD CE1TX loudspeakers. The beauty of headphones like this is how the amp is connected directly to the headset's raw, responsive full-range drivers with no crossovers or filtering. This allows listeners to experience an unmolested view of the recording. You know how much the purist in me loves that.

Curious, I decided to pause here and put the feedback back in, fully expecting to prefer the sharper transients and better sortedness I had enjoyed for many months. But now the feedback felt annoying, like a politician squeezing my hand too hard when they shake it. Tight is only right when it does not feel like the sound is being restrained or manipulated, and that's the subtle feeling I was getting with every recording with feedback engaged. This was strange in an unsettling way because,

I swear, feedback in the Elekit never felt too tight before I removed it. Once again, sequence is everything.

I left the feedback in for a couple of days to see if my brain would adapt, and it did—a lot. But, ultimately, my artist's spirit preferred the darker, looser Blade Runner aesthetic of no feedback.

Because the TU-8900 sounded authoritative powering the 20 ohm Audivina, I switched the front panel jumpers to a position the manual said would favor my 250 ohm, 96dB/mW Beyerdynamic DT880. I put on "Time Is Tight" and "Green Onions" by Booker T. and the MGs (16/44.1 FLAC, Stax/Tidal) and was disappointed by the disembodied weakness of the sound. Then I remembered that I hadn't used the 880 in months, so I gave it a few hours to loosen up. After that, it felt a lot more solid and 3D playing "Funky AECO" on The Third Decade, by Art Ensemble of Chicago (16/44.1 FLAC, ECM/Qobuz). After some loosening up, the Beyerdynamic sounded enough looser and better to say the Elekit's headphone amp was rocking the high-impedance DT 880s. The modestly priced DT 880's popularity rides on the extreme naturalness of its detail and the correctness of its tone; the Elekit's headphone output honored those traits while adding super PRaT and corporealized textures. I spent the entire night playing every Mississippi Fred McDowell song I could find and went to bed feeling satisfied.

Because I couldn't stop myself, I had to try JPS Labs' stylish, planar-magnetic open-back, the Abyss Diana TC, which, at $4495, cost more than the Elekit TU-8900. The Diana TC is specified as 69 ohms and only 90dB/mW. I believe it is only a click or two less resolving than its stablemate, the Abyss AB-1266 Phi TC. South African Abel Selaocoe playing his own composition "Ibuyile iAfrica," off his debut album Where Is Home/Hae ke Kae (24/96 FLAC, Warner Classics/Qobuz), came through as whole, clean, and exquisitely gold-toned.

The TU-8900's headphone output sounded good, but it got a bit tense and dulled trying to drive the 60 ohm, 83dB/mW HiFiMan Susvara. I'd say 90dB/mW might be a wise bottom limit for headphone sensitivity with this amplifier.

I never thought I'd say this about the headphone output in a moderately priced integrated amplifier, but this one's got the stuff to be all the headphone amplifier most people will ever need.

Over the last three decades, I've enjoyed an enviable cadre of famously expensive 300B amplifiers. I wouldn't tease you with these Elekit stories if I did not believe Victor Kung and Yoshitsugu Fujita had created a uniquely interesting product, one that delivers large amounts of engagement factor, serious tube-rolling fun, and extraordinary sonics at a very reasonable price.


georgehifi's picture

If you keep your head, and keep it simple.

I traded in a new unassembled TU-8500 kit preamp of Elekit's, the owner couldn't face it when he saw the bags of parts.
I can say that even though very plain looking the parts content were 1st Class. So instead of selling it when things were quiet and the wife was away, I decided to assemble it.
They are very well thought out kits, with clear and concise assembly instruction, if you can solder and don't rush and have a patient demeaner, you will be able to build these kits.
I did bench measurements on it afterwards, and it was as good as any $$$$$ tube preamp I've measured from the big esoteric names.

BTW: just go with the basic $1945 kit, it will sound very good, instead of the glitz and glamour upgraded $3280!! (without tubes)!!!

Cheers George

Vinylleroy's picture

If I understand it correctly. All the upgrades you did on this KIT are not available via the Elekit website, but came to mind by the Elekit distributer?

Kind regards,

Herb Reichert's picture

that is correct


Vinylleroy's picture

for your reply, and excuse me for posting my reply 5 times. Is this a Kit that can be build by someone with no experience building kits at all?

And thank you for picking out these gems and telling us about it.

Kind regards,


MatthewT's picture

The manual is very well done, tons of support on DIYAudio, and Mr. Kung is very easy to reach for help. Give it a try!

Herb Reichert's picture

Successfully building your first kit requires both hand and mind skills.

If you can build a plastic airplane model and you can complete a jigsaw puzzle

it will be the same kind of fun


vkmusic's picture

Is this a Kit that can be build by someone with no experience building kits at all? long as you read and follow the manual step by step.

Ortofan's picture

... with other tube-type power amps available for around $4K, or less, such as the Rogue Atlas Magnum III, the PrimaLuna Evo 200 and/or Evo 300 and the Cary CAD-120S MKII.

Again, particularly when there are reviews of such low power output amps, could the reviewer use an oscilloscope to determine whether or not the amp is ever being driven into clipping?

vkmusic's picture

if you want to compare which amp has good value. Here is the simple math.
You can add the cost of a pair of LL2785c amorphous core output transformer (5.4KG) + a pair of Audio Note silver 0.1 (600V). It is more or less 50% of the total cost.. There is no amplifier in the market at this price range will offer you Audio Note silver cap or Amorphous Core output transformer.

The amorphous core is a 5.4kg opt...
DIY kit does not have high markup...

TU-8900 uses an over spec output transformer. Clipping is very common in low cost amp. When an amplifier signal makes the amplifier try to produce output voltages that are beyond its capability, which is what causes the clipped signal.

Ortofan's picture

... output transformer with a core made from HI-B steel?

johnnythunder1's picture

use transformers made from "hi-b" steel ? I would think a transformer's metal is only as good as the thought behind its windings in the same way that an alnico magnet doesn't guarantee a nice sounding speaker.

Ortofan's picture

... specifies the use of that type of core material in the construction of their output transformers.

vkmusic's picture

LL2785C AM is amorphous core (for TU-8900). If it is a steel core LL2785c (no AM) it is $250 cheaper.
TU-8600S is using steel core not amorphous core.
LL2785C AM is a custom made for TU-8900.
I only want the best and afforable parts.

georgehifi's picture

"Is this a kit anyone can build"

As I said in my first post:
"if you can solder and don't rush and have a patient demeaner, you will be able to build these kits."

If you can't solder, stay away you'll make a mess of it.

Cheers George

taipan254's picture

Herb - great write-up. Why did you use a preamp here? Isn't this considered an integrated amp?

Thanks from a fellow Brooklyn-ite!

georgehifi's picture

Sort of, it's "really a power amp with a passive volume control" at the front of it. Similar to the 8600 and his others

Looks like the headphones are driven from the speaker outputs

Cheers George

vkmusic's picture

The headphone is tapped to the OUTPUT Transformer with 3 resistors
You have an option to run TU-8900 as an integrated amp or power amp with a HEX socket..

Herb Reichert's picture

the TU-8600 and the TU-8900 have enough gain to possibly (with the Heretics) skip the pre but it has only one input and I use three sources.

I have never used either of them as an integrated, but I am certain many do. You should speak to Victor Kung about this.


JRT's picture

I have not seen the schematic for the TU-8900. Looking at the schematic for the TU-8600, the secondary windings on the output transformers have taps for 8 Ohm and 5 Ohm nomonal loads, and are single ended outputs with both channels referenced to a signal ground that is also referenced by both channels' single ended inputs. Though routed through a switch to select between taps (higher/lower Z) and through switched contacts in the headphone jack (1/4 inch TRS receptacle), the outputs remain ground referenced single ended with unbalanced impedance at the terminations into the rear panel output connectors.

Each of the two (stereo) ground refenced single ended outputs could be attenuated through a simple two resistor L-pad attenuator (voltage dividing network) using low noise noninductive resistors of sufficiently high power dissipation to feed into the ground referenced single ended input of a downstream preamplifier and amplifier, which if sufficiently low in noise and distortion, the seemingly desirable euphonic nonlinearities added by this tube amplifier would dominate well above the vanishingly small linearities added downstream.

The resistors in the L-pad can be selected to provide suitable load on the secondaries while also providing suitable attenuation of the signal, while signal level into the tube amplifier is optimized for low noise and low nonlinear distortion. The downstream power amplifiers can be selected to suitably drive the loudspeakers at outputs well above what the flea power 2A3 triodes can deliver.

With 2A3 triodes installed the amplifier is rated at 3.5W per channel at 10% THD, and presumably very much lower distortion at half (-3dB) of that power output.

5.29 Vrms across 8 Ohm load for 3.5 Wrms
presumably using the 8 Ohm taps

3.74 Vrms at half of that power

0.374 Vrms is -20dB relative to 3.74 Vrms

+3dB is useful headroom for intersample overs in the top couple of octaves, and 10% THD would not be horrible on those brief maximum peaks, especialky considering that much of the nonlinearities would be in ultrasonic frequency spectrum.

With the DAC fed a pink noise signal at -20 dBFS signal level, and an AC voltmeter measuring output across loaded 8 Ohm tap on the outputs, the volume control could be adjusted to output 0.374 Vrms. That would place 0dB at 3.74 Vrms, and there would be +3dB headroom above that where THD would rise to 10%.

Those would be target signal voltages entering the L-pads. Target output voltages need to be determined, likewise the design load impedance. Using that aforementioned -20 dBFS pink noise signal, the volume knob would be adjusted to the target signal voltage at the output of the actual L-pad as-built.

So then what load impedance? The headphone circuit including headphones, exhibits a load impedance that can be varied between the limits of 13.5 Ohm minimum and 15 Ohm maximum, applied to the output transformer secondaries at either the high-Z 8 Ohm taps or the low-Z 5 Ohm taps. That load impedance will reflect back to the primaries approximately as the square of the turns ratio, so load impedance on the secondaries affects the loading of the output triodes. The output power is low enough that the designer was not constrained much by considerations of excessive power dissipation, rather presumably designed that load impedance in the interest of optimizing sound quality
via the headphones. So perhaps similar load impedance target might also be selected for the L-pad for sinilar reasons.

The inverse of the RMS of the inverse values of the two limits would be:
So maybe a good approximate target for the L-pad sum might be R1+R2= 14.2 Ohm.

The idea of hybrid amplification to boost the output of flea power tube amplifiers is not new.

georgehifi's picture

If you have enough volume with a passive pre you'll get a better sound, because there's no need for active preamp components that have even more gain/distortion/noise etc only then to knock it all back down again to the passive pre's level

Cheers George

JRT's picture

In his review of the Schiit Freya Plus, Kalman Rubinson did not prefer the non-buffered passive preamplifier mode, rather Kal clearly expressed a preference for unity gain buffered output. He supposed that might be attributable to driving relatively long line level interconnection cables between preamplifier and amplifiers in his system. In support of that, consider that buffered output mode presents a low source impedance into the cable capacitance and the non-buffered output of the passive attenuator presents a high source impedance into same.

Schiit Freya S and Freya Plus are similar preamplifiers, each having remote controllable stereo pair of 128 step stepped attenuators, relay switched R-2R resistor arrays, and with three operating modes, non-buffered passive, buffered with unity gain, and buffered with gain. The Freya S uses solid state for the latter two modes, while the Freya Plus uses solid state for the buffered unity gain mode, and octal (eg 6SN7) tube based gain stage for that third mode.

There was also a third variant of the preamplifier which is now an obsolete product, the Freya N which utilized less expensive noval tube types instead of the octal tube types. There are a few of those available at a discounted sale price from the Schiit outlet.

Both current versions of the preamplifier include balanced and unbalanced inputs and outputs, and include conversion of single ended unbalanced impedance input to symmetric differential balanced impedance output in the latter two modes. The passive mode is a little different in that because there are no active amplifiers in the passive mode for that conversion, or transformers, so rather in the passive mode the preamplifier converts the single ended unbalanced impedance input to an asymmetric differential output with the audio signal on one leg of the balanced impedance output, and with zeroed signsl voltage on the other leg (rather than the mirrored signal voltage that would be exhibited in symmetric differential output). In all cases single ended unbalanced impedance inputs pass to single ended unbalanced impedance outputs, and differential balanced impedance inputs pass to differential balanced impedance outputs.

windansea's picture

This might be one of my top 5 Stereophile reviews ever, one which I will re-read several times for the sections comparing zero feedback against light feedback. Love the observations and the prose, and the addendum on headphones listening. Art Dudley is reading this one from heaven.

There is one extra element that I would have liked: if Herb had recruited a friend to switch the feedback jumper a few times, just to verify that he was hearing something differently. Better yet a clueless friend who has no idea what the jumper switch does. I'm thinking once per day, hey neighbor can you flip this switch to up or down every day, write it down but don't tell me, for one week. That extra effort would have taken this wonderful review from a beautiful and probably accurate piece to a beautiful and verifiably truthful piece.

Lastly, I want to buy an Elekit now. Thanks Herb for a classic review!

LouisB's picture


I recently bought the kit without having any experience of building an amp before, but after reading all the feedback figured that this was my only way to a really high quality 300B/2A3 tube amp, and I chose to trust the experience of all the builders on DIYaudio and Herb's reviews with the 8600 and 8900. I was torn between permalloy and amorphous core transformers as I hadn't heard AC before but in the end wanted some of that esoteric magic so went for the 8900 with amorphous core transformers. I had to buy a soldering iron and practice soldering (a lot) before beginning but the manual is very clear and all the resistors and caps are labelled on the board (once you figure out how to read the colour codes of resistors... ). I built it in two days with only one or two fiddly bits. The box pins and volume pins, for example, are very tight and I was trying to avoid any solder bridges etc. so real took my time to double check all the joints. In the end I powered up accidentally as the power button was pressed on... It lit up and nothing exploded!

I didn't get the upgraded resistors and AN caps, I went with all stock so I can gradually replace parts and hear the differences.

It sounded great from the start. I only have a few hours on it and the immediacy, intimacy and clarity are present and will improve. No fancy tubes yet. The emotional connection that I associate with tube gear is developing and my system is not yet set up to get the best out of it. My speakers are studio monitors and need current to bloom. There's not a lot of gain on the amp (only one gain stage) so the volume needs to be cranked to get music to flow, but there are so many options and pathways to improve and know that the amp will deliver as I upgrade and discover silver or tinned copper caps and high sensitivity speakers and passive or active preamps...

So yeah I'm happy so far... Feels like the beginning of a relationship...

S_Procee's picture

This, really, great sounding amplifier can be further improved by applying pieces of adhesive bituminous foil, the stuff used for cars e.g., on the bottom and sides of the main chassis and on the transformers cover. This results in less upper bass bloom hence deeper bass, and a cleaner overall sound. Just tick your fingernail on the casing and when that sounds bright and metal'ish apply more damping until it sounds dull. Although some sources on the internet warn for overdamping, which might be true for equipment relying on vibration i.e. speaker and turntable, I can't think of a risk in applying too much damping in an amplifier. This with the exception that you'd block e.g. the cover by a too thick layer. The cost was less than 15 euro (for one 50x50cm sheet of 3M foil).
Enjoy! this amp is a delight, I use it with EML 300B and 12BH7 tubes non-NFB; PAP trio 15H1 open baffles and Music First Classic 2 TVC. The latter controlling input and volume with the TU8900's VC at 100% so effectively shortcutting the amp's potentiometer.