Thx, one last question. How do you extend the (I believe) 3 available PCIe slots to accommodate the 9 additional drives?
Taiko Audio SGM Extreme : the Crème de la Crème
- Thread starter CKKeung
- Start date
There is a pcie card card that uses 4 nvme m2 cards on the pcie card.
Hi Emile, there are 2 LAN ports in Extreme. Any preference in choosing the port? And what is the purpose of having 2 LAN ports?
Thanks
Thanks
There is no preference, you can use either. It does allow you to connect an ethernet based endpoint directly to the Extreme.
It would not be used for firmware updates but it could be used to connect future technology.
Hi Emile,
do exist with Extreme any SQ benefits to use an endpoint vs direct USB connection?
Thanks
Matt
USB, but it’s all pretty close. USB has an obvious advantage in supported sampling rates of course.
An "appropriate" CPU is not any more important than the other components in a music server, it's just most servers today are using what I would consider less than appropriate CPUs.
Of course, it's not just about high-power CPUs. Having tried my Hewlett Packard Z820 workstation which has dual Xeons and my son's high-powered gaming PC as music servers, it's hard to listen for very long because they sound quite harsh but I would consider a high power CPU a necessary starting point when looking for a music server. It's what initially drew my interest to the Extreme.
I'm sure there will be many who will be skeptical of this claim because how can >95% of the music server manufacturers be wrong? Considering how a non-upsampling music server that uses dual high-power CPUs, no fancy clocks, and outputs through the stock USB port of an off-the-shelf motherboard is being acclaimed by many as the best music server in the world right now and has had a waiting list to buy one that has been as long as 3 months despite its $25k price tag is pretty strong support of this claim.
Once again, I don't want people to think that the Extreme is just about its dual Xeons because it is not but it is fundamental to its design. Had Emile gone with the same low power Pentium used by the Statement, as an example, there would have been no need for such a large expensive chassis with such an elaborate passive cooling system along with a power supply capable of powering a small amplifier. The engineering and cost to properly implement 2 high power Xeons are obviously not trivial hence the Extreme's large price tag but unlike many high-end audiophile components that are often over-engineered to diminishing returns, based on my own experience, the gains from going high power are significant and cannot be overstated.
This brings up the question of obsolescence. I had barely owned my InnuOS Zenith SE Mk 2 for a year when it was already obsolete. Not that the Statement made it obsolete because while I felt the Statement was a significant step better, it was not revolutionary better. It was my discovery of the benefits of high power CPUs that did the Zenith SE in. Once I heard what a music server with a high power CPU could sound like, there was no going back and so this raises the question of whether the Extreme's dual 10-core/20-thread Xeons will be outperformed by future CPUs. When I pressed Emile with this question, he indicated that with the Extreme's tally of 20 physical cores, "a couple of cores are being wasted" meaning they aren't being used suggesting that going with more cores is unnecessary, In fact, Emile shared with me that he purchased a whole range of AMD Epyc Rome CPUs having as many as 64 cores per CPU and did not find them to improve upon what he already had. Barring any major software change that would have vastly different CPU requirements than Windows and Roon, the processing capabilities of the Extreme should safely see it through the coming years.
Emil
Eloquently written. Emile should put this in his brochures!
The concept behind an endpoint was to mask the noise generated by a noisy upstream server or NAS and so the general belief has been that the endpoint should be a low noise device suggesting once again a low power CPU. When I first discovered the sound quality benefits of a server with a high power CPU, I continued to use my endpoints that housed a low power CPU including such things as a NUC, uRendu, and sMS-200ultra Neo. As I improved the quality of my high power server, I found that these endpoints were no longer necessary and in fact, became detrimental as they significantly hindered dynamics and shrunk the sound stage. It became clear to me that the endpoint needed to be equivalent to the server in order to avoid this and Pink Faun obviously agrees since they are marketing a dual 2.16X solution. With the Extreme having dual CPUs and with each CPU having its own independent bank of memory, you have the capacity for the same distribution of tasks but housed in a single chassis which is really quite elegant. In Emile's testing, he indicated this distribution between 2 CPUs sounded better than a single CPU having at least 20 cores.
Hi Romaz,
The Extreme is a wonderful piece of equipment and a masterpiece of product development. I am more interested in discussing high vs. low power CPUs (HPCPU & LPCPU) The question for me is; what is the HPCPU actually doing to improve the sound quality? If you look at the reason for going to LPCPUs is the first place it was about managing noise generation. The higher the CPU’s power, the more noise it generated....that was the generally accepted rule of thumb.....but now we discover that with certain high power CPUs servers combined with LPCPU end points , the low power end points negatively influence SQ. Well that’s understandable, logical even from a noise generation standpoint. The server’s HPCPU is cleaning up the incoming stream, but adding its own noise in the process that the LPCPU endpoint has insufficient capacity to deal with, without adding latency, so dual HPCPUs is the solution. Lets follow the logic:
Incoming noise—HPCPU server clean-up + HPCPU noise—LPCPU Endpoint clean-up + latency= Lower SQ
Incoming noise—HPCPU clean-up—HPCPU endpoint clean-up + HPCPU noise = Higher SQ
The conclusion I’d reach from that is that 2 HPCPUs are capable of correcting substantially more noise than they generate, while still leaving plenty of capacity to avoid latency affects, so in the above scenario a HPCPU Endpoint is key to delivering lower noise, low latency signal to the DAC.
My approach has been entirely different. Instead of using HPCPUs to correct incoming noise while adding their own noise to the outgoing signal I get rid of the incoming noise before it reaches the server. This has several benefits. 1. It means the CPU load is lower. 2. It means that a LPCPU can be used so 3. I get less incoming and CPU-generated noise going to the DAC and less latency throughout.
So IMO, LPCPUs may still be the way to go, but only when you do something to ameliorate the upstream network noise and jitter In order to offload the CPU in the server. This gives you the following ‘POTENTIAL’ scenarios:
High incoming noise and jitter....2 HPCPUs sound best
Low incoming noise and jitter.....LPCPUs sound best*
* This is so far entirely unproven, also by me, as I’ve never compared my system to a system based on 2x HPCPUs. But the fact that 2 HPCPU’S makes the system more immune to incoming noise does indicate I may be on the right track. All I do know is that the sound of a Statement with a low noise incoming stream is quite remarkable and the sound actually has a physical impact on the listener, generating shivers down the spine, whoops of joy or tears at the sheer beauty and gloriousness of the music, sound quality that so soundly thrashed my vinyl set-up that it was 100% clear that in my particular set-up vinyl no longer had a viable role i.e why go to all the bother of playing records when the sound quality thus produced is overall poorer in every aspect.
But we both know how things are in audio, that everything is relative...i.e improvements very much depend on what went before. So it could be that what the Extreme is capable of producing is still superior to what I’m achieving...all I know is that there are some huge benefits to be had from optimizing and removing noise and jitter from the entire network stream, where every step of the way is rewarded by substantial gains.
I’m now at the point where the addition of a single DC cable on a network device has the capacity to substantially diminish listener enjoyment until its run-in. That tells me a lot about the importance of the incoming stream’s ‘condition’ and ability to influence the system’s final SQ.
Best regards
Last edited:
I guess the proof is in the pudding. I did compare the statement and extreme side by side in the same system but not to the extent that I would like and I was not familiar with the system. I do intend to repeat the comparison before I make my decision.
Frankly am unconvinced that high or low power CPUs are the sine qua non of the desired state. Rather more likely is how the ingredients are blended, and that I suspect is the essential differentiator. Or put another way the Extreme - basis the eulogising in this thread - works because it is well-engineered as a system. Recognising from the ground up the import of physical and electrical isolation, power/current supply resilience, processing capabilities, memory latency, and software selection/pathing...realising the desired performance through considered integration not just as the result of one element.
Could I see a similar outcome using low powered CPUs.....if approached with similar competence/intent.....I would not be suprised. After all the world of vinyl seems to accomodate more than a few paths to superlatives.
Suffice I look forward to auditioning an Extreme in early January....and I fully anticipate it will be plug and stay...
Could I see a similar outcome using low powered CPUs.....if approached with similar competence/intent.....I would not be suprised. After all the world of vinyl seems to accomodate more than a few paths to superlatives.
Suffice I look forward to auditioning an Extreme in early January....and I fully anticipate it will be plug and stay...
That is a very interesting question because Romaz, Taiko and Eurostar are always putting his HPCPU as and advantge.
I see very clearly that SGM hasn´t a clear goal from its beginning and it explains the actual solution for the Extreme.
They believe that HQPlayer was the solution for cheap Dacs. Classical T+A Dac with SGM2015.
That was the argument they said to me:
" With HQ Player doing PCM to DSD conversion, and the 1-bit data stream being fed to a DSD converter, the actual / functional Digital to Analog conversion is not going on in in the DAC, but in software, in the PC, in PC clock time. This realization started the experimentation with replacing the computer motherboard clock with more short term frequency stable types including TCXO, and OCXO's with different frequency stability specs. From our extensive testing of motherboard clocks, the sound improvements are coming from three sources, the high frequency power supply we power the clock with, the lower low frequency phase noise of the crystal, and the greater short term frequency stability. A very interesting effect is we achieved a 6% reduction in total system power consumption and a 2 degree Celsius overall lower operating temperature depending on the clock model and its surrounding circuitry. The sound quality improvement is not as huge as we get from pushing the RF noise floor down, but it brings a pristine quality to the sound and a coherence, that our other customizations don't bring. For us on the team, it's a must have."
As you can read the only improvement effect was less power consumption and temperature. Even they said the improvement wasn´t as huge as reducing noise floor....
That was a bad solution that i´m glad to know they don´t implement on the Extreme because replacing the oscillator that the Asus board itself carried by means of a coaxial cable, welded to the motherboard ruins all accuracy of the device. The difference between the masses to be mounted in this way, the welds of the cable to the plate, etc... and the consideration of that the server runs as slave of the dac. So it is the clock´s dac who manage the jitter.
This is only one example that experimenting isn´t the way to reach your goal.
On the price side, it wasn´t very logic to spend 16000€ on a Server for a 3000€ Dac.
But now they have discovered that bit perfect sounds better than HQPlayer because now top Dacs doesn´t require it.
It is very funny that affirmation.
But why they continue using HPCPU?
Because their technology is still based on HQPlayer.
Bit Perfect is an unaltered signal and both CPU and Ram are a noise focus.
In digital audio application, the critical RAM/CPU parameters are bandwidth and latency, but there are other details that have final influence. The design of the memory chips within the module itself has an influence on the final result. It is the contamination of the masses and the currents of return to the source that makes the final result vary. The only solution is to develop a sophisticated power supply that dampens that 'pollution' obtained from rapid memory and CPU switching. We are not talking about 0 and 1 as Romaz said, each number represents a voltage. The best final processor is the result to balance power and 'silent' electric.
What power needs Bit Perfect? Nothing from bit perfect side. Only for metadata, etc....
Even now less than ever with the launch of Roon Valence...
Roon Signal Path shows the speed processor up to 100x. So as you can see, playing both streaming as internal storage the Speed Processor isn’t showed because the processor only works less than 1% of its potential.
So SGM is a victim of itself technology.
IMHO the Extreme is a multifuction Server.
It is designed to work in any Dac putting different solutions that are contradictory each one with the other.
Resuming. A top Dac doesn’t need a HPCPU server because all the processing needed is inside the Dac.
Cheers.
Last edited:
It would be very interesting to validate Blackmorec's assumptions, compare an EtherREGEN + Statement vs. Commercial Switch + Extreme.
Your comments assume that Emile did not try low powered CPUs in a configuration and build similar to the Extreme and preferred the high powered CPUs. My understanding is that he did.
Clearly the entire build defines the quality of the server but I do think that Emile isolated the effect of the CPU and decided that high powered was best.
Maybe he can chime in.
I do indeed test most CPUs and embedded systems when they are released, including non x86 architecture.
