Further Isolation of the Silverstone PCIe - USB Card

[acg]

Following on from a couple of other threads to do with the PCIe-USB card in the pc, this thread will look at isolating the PCIe-USB card as much as possible from the PCIe socket and the impact, if any, on sound quality.

To implement this, I purchased a $6 PCIe riser cable from fleaBay , read up on the PCIe protocols and formed a plan of attack.  The first thing is to divorce the power to the card from the PCIe socket.  The PCIe socket provides 3.3V, 3.3V Aux. and 12V whilst the USB Host Controller on the Silverstone card needs 3.3V and 1.05V.  Additionally the USB ports on the card need a 5V supply and this is supplied from an external source.

The pinout for the x1 PCIe connectors is shown in the image below.  The forward 11 pins handle power, connection testing (JTAG) and System Management Bus (SMB):  the JTAG and SMB functions are not required by a USB card.  Likewise, my aim here is to cut the power supplies from the PCIe socket in the computer, so theoretically the 12V, 3.3V, 3.3V Aux and ground pins are not required either.

With these things in mind I set about butchering my brand new PCIe riser cable to experiment with pins 1-11 on both the B (component side) and A (solder side) sides.  I cut wires incrementally and have determined the following things for the Silverstone card:

• The Silverstone card does not use the 12V or 3.3V Aux supply pins at all.  It only uses the 3.3V supply pins.
• The card does use the "Wake" pin, although this is an optional pin used for waking the card from low power states (ie. from sleep).  The "Wake" pin is the only pine used on the B side of the card (the component side).
• The only pins used on the A side (the solder side) are the "PWR Good" and "PRSNT1" pin (use to determine what size card is attached to the PCIe socket - in this case a x1 card)
• If I supply 3.3V from my lab LPS to the PCIe pins, and return the ground via pin 7 the card runs perfectly.  Note I also have 5V feeding the molex plug.

So, it can be done...I am now powering the Silverstone card from two rails of my lab LPS and it functions perfectly.  The orange pins from 1-11 on both the A and B sides in my PCIe pinout below are the only ones connected to the PCIe socket (every pin 12-18 is still connected).  Pin 7B goes to ground in the 3.3V rail of my lab LPS, and pins 8B/9A/10A all go to the positive.

How does it sound?  I will leave that for another day because I have only just started playing music through it, but the signs are promising.

The next step is to fiddle with pins 12-18.  All of these pins are required (CLKREQ would not be required if the card only supported a high power state i.e. always on) but there are 6 ground pins which depending on the design of the Silverstone card may or may not be related to the data operations of the card and not the power operations.  I will look into it.

Cheers,

Anthony

[acg]

I found a little more time this afternoon and managed to have a hack at pins 12-18 on the riser cable.  Have a look at the attached diagram (orange are the pins connected between the PCIe socket on the mobo and the Silverstone Card)...ground is now _completely_ severed and power is now completely supplied from my lab LPS (3.3V to the PCIe pins and 5V to the molex plug).

It's probably placebo...but 5 minutes in the sound in freakin' awesome.

Can somebody else have a go at this?  You will need a two rail supply and a cheap PCIe riser cable.

Anthony

EDIT: There is a better image attached in the PDF file.

[PeterSt]

Hey Anthony,

I didn't read your text yet, but is it possible that you downsize that large picture (width 800 or so) and put that as *additional* attachment ? Put it in as the first attachment (so detach the other two (in the first post I mean), start with the downsized and then add the other two again.

Navigation in such a large picture is almost undoable (per this forum means), but works quite well if you can see where you want to navigate to *first* (that's why the smaller picture).

Will dive into the content later ...
Peter

[acg]

Done Peter.  I have to remember that not everyone has a couple of giant hi-res screens in front of them...that image fits on the page on one of my screens.

Anthony

[PeterSt]

Ah super Anthony. But notice that you can (or should) just attach that large picture as the third attachment now (might people want to see details).

Btw, I already noticed that you have a nice camera there ...

Peter

[PeterSt]

Btw, here is something I had under my sleeve since two days, and I should have put it up right away; It is from Silverstone Taiwan and the most useful part you already found yourself (12V not used at all) :

(do notice : this is a response to what has been read in our forum about the card and is not much about my explicit questions (I asked for the schematic which they won't provide))


However you request about the power current for the USB port, since EC04-P does not limit the power current in each of the USB port, so in theory it can be based on the maximum power current which is provided by the power supply to use, but the user must  consider the USB connector has to withstand power current the restrictions to the standard USB 3.0 interface as standard power current term, usually its typical power current is 1.8A for each USB connector.

and

Notice for this user., our EC01-P/EC03-P/EC04-P used PCIe 3.3V to convert voltage to host ic(NEC) and molex power connector provide the 5V power to each USB port and we didn't use 12V power for these models.

[acg]

Btw, here is something I had under my sleeve since two days, and I should have put it up right away; It is from Silverstone Taiwan and the most useful part you already found yourself (12V not used at all) :

(do notice : this is a response to what has been read in our forum about the card and is not much about my explicit questions (I asked for the schematic which they won't provide))


However you request about the power current for the USB port, since EC04-P does not limit the power current in each of the USB port, so in theory it can be based on the maximum power current which is provided by the power supply to use, but the user must  consider the USB connector has to withstand power current the restrictions to the standard USB 3.0 interface as standard power current term, usually its typical power current is 1.8A for each USB connector.

and

Notice for this user., our EC01-P/EC03-P/EC04-P used PCIe 3.3V to convert voltage to host ic(NEC) and molex power connector provide the 5V power to each USB port and we didn't use 12V power for these models.

Ha, well I guess that is now confirmed!!

[CoenP]

Hi Anthony,

Its a miracle this works. Except for the tx and treceive lines that are balanced all other pins are referenced to groud. Without a ground the voltage cannot be determined. I expect that a pin that is 'low' provides the grnd connection necessary for operation.

I would be really concerned about ESD between card and mobo as there is no other path to travel besides the 'hot' pins.

Did you reference the powersupply to any pc ground?

Regards, Coen

[acg]

Hi Coen,

I am surprised this works too.  Not only that but it works straight away.

The lab LPS is plugged into the same power socket as the rest of the audio gear including the pc, but other than the four wires running to the Silverstone card there is no other connection.

All I can say is that I ran this for a few hours last night and did not have any issues and that it sounded great.

Anthony

EDIT: The two PRSNT pins are directly bridged on the PCIe card so that the PCIe Host Controller on the mobo knows what sized card is in the slot, so these two pins are not referenced to ground on the PCIe card at all.

[acg]

So I have done a little bit of listening this afternoon and I must say that the change for me here is as big as minimising the o/s.  My friend Dave was here as well.  Dave is a music lover but not an audiophile (lucky bloke!) and has a good ear for music but does not spend any time reading a audio or visiting forums and such.

Anyway, Dave was convinced of the changes in my system with this tweak and heard them pretty much as I did.  His most telling comment was that it was just "less noisy - more clear".  My own comments are significantly increased clarity with a speed and separation that is easy to hear particularly in the voices and with normally harsh instruments such as the cymbals and bells.  Remember this is with an isolated Silverstone card, which I did not like initially, but with further isolation from the PCIe bus the sound is sweet and delicate and dynamic and beautiful.  The "sssss's" sizzle and decay but don't bite. 

But the most important improvement for me is the "presence"...I am getting more nuance, micro-detail, clarity and believability.  It is a beautiful sound and a major step forward in my system.

So, what is causing this?  Is it that the ground for the USB card is now separated from the XXHE pc?  Is it that the cheap lab LPS is injecting cleaner power into the Silverstone card which in turn is helping the USB Host Controller IC and clock to perform more optimally?  Is it a combination of the above?  I don't know.  All that I know is that it is a major step forward in sound quality and it is something that I can make easily and it is something that Peter could also make easily.

My next thoughts are to the dexa clock guys.  Can you give this a try?  I will post a more detailed procedure to get this to work.  What I want to know is if this further isolation of the UCB card improves your sound with the clock mods.  If there is no change then we may be able to say that the clock mods account for the just about all of your improved sq and my isolation scheme is not warranted in that setup.  What I would like to hear (but of course I want the truth) is that this isolation scheme helps further with your setups and that the clock quality is not the be-all-and-end-all, although that would make things easier!!

Cheers,

Anthony

[CoenP]

Hi Anthony,

Did you disconnect the chassis ground wires in the NOS1 like Peter suggested in the " best sound ever"  disconnected ground thread?

If you didn't its about the USB grounding wire from the USB receptacle to the PSU and the one from the PE to the chassis.

regards, Coen

[AlainGr]

Hi Anthony,

Did you disconnect the chassis ground wires in the NOS1 like Peter suggested in the " best sound ever"  disconnected ground thread?

If you didn't its about the USB grounding wire from the USB receptacle to the PSU and the one from the PE to the chassis.

regards, Coen

Hi Coen,

There are two of them to disconnect ? There is one that I disconnected from the printed circuit (it is attached to the upper right screw in the right branch when facing the NOS1), but the other ?

Regards,

Alain

[CoenP]

The other runs at the power connector end of the right leg (fron facing) to a screw at the top of the right side of that leg....

regards, Coen

[juanpmar]

Hi Alain,

Before opening my NOS1 I wonder if you noticed something after disconnecting the cable.

Incidentally, I think you just have to disconnect the black wire to avoid doing so in the USB cable that is more difficult. At least this is what I understood from Peter posts.

Best regards,
Juan

Edit: these are Peter words: "under the top right bolt of that PSU is a black wire; unscrew the bolt, detach the wire (it is not soldered), ...do NOT detach it from the USB interface as of now because it is too fragile there."

[PeterSt]

If you didn't its about the USB grounding wire from the USB receptacle to the PSU and the one from the PE to the chassis.

H*ll no Coen. I never said anything about the wire from PE to the chassis ... Leave that be please.

Peter