!nverse: A Highly Versatile Steam-Box Design

[Hahutzy]

Update 10/17/2015:

Hey everyone, a quick update from me. I have done the PCIe riser test with 5 different risers:

[...]

From this test, it is clear that DIY solutions to shielding the PCIe riser cable does not work. The LiHeat cable performs slightly worse than the 3M cable, but at less than half the price. Therefore, I think the LiHeat cables are what we'll be bundling with the !nverse case. Now we need to see what kind of bulk discount we can get.

I must say, I am quite perplexed by this result. Here's a few notes on the matter:

i) In my build (EDIT: with a 980 Ti), I am using an unshielded PCIe rise, the ones that are as cheap as they come. And to date, I've ran upwards of 100 runs of Unigine Heaven at 1080p with max anti-aliasing. I have not crashed once, let alone system not booting. The case and its components are currently my daily driver machine. It has been through gaming, watching videos, etc. and I have not once seen a crash/display failure.

ii) I doubt I got lucky with my one PCIe rise cable, because I've used 2 others in my earlier prototypes, and they all ran 3Dmark11 fine.

iii) Even more surprising to me is the difference between your test #2 and test #3. A faraday cage is a faraday cage. If you've set up the aluminum foil the same way as the EMI shielding tape, the results should be the same...

With that said, I'm sure you've moved on and went with the solution you felt the most confident about, and that is great.
Just found the results a bit interesting versus the data I collect from my project.

 

[iFreilicht]

How long were the risers you used, Hahutzy? The ones esplin uses are probably 20cm long, maybe more, so that can have a huge impact on the results.
EDIT: They should be around the same length, just checked the case design again.

A farady cage might be a faraday cage, but things aren't that easy in HF applications. Depending on the wavelength of your interfering EM waves, the strength of them, the connection of your shield to ground, results can strongly vary. I wouldn't want to make any statement on this matter that I can't back up with experimental data.

 

[Hahutzy]

250mm. You're right, it is possible there is more to it than what I said. Maybe the thickness of the foil vs the tape differed enough, or other possibilities.

To be honest, from all that I've read regarding this issue and sheer number of people that have had negative results, I'm questioning moreso why my card is running properly.

 

[iFreilicht]

It might be shielded by other means. For example, it could be that your GPU has a backplate, so when the ribbon is running between GPU and motherboard tray, and the latter doesn't have a cutout for the CPU, the cable is pretty much shielded by two grounded, highly conductive sheets of metal on both sides, and only small parts of it are exposed. That way, EMI doesn't couple through those metal parts but is redirected to common ground, where it should go. In the end, nobody can really know.

 

[EdZ]

Remember that shielding is not only against external EMF, you also have crosstalk between the PCIe pairs themselves (Alien Crosstalk in data cabling parlance). The 3M riser (and possibly the LiHeat, I don't have one to sacrifice) uses Bonded Pairs, effectively tiny shielded coax leads in a ribbon configuration. The cheaper ribbon-cable risers are effectively bare untwisted pairs. An add-on foil shield, even with correctly terminated grounding, will do little-to-nothing to prevent crosstalk.

 

[esplin2966]

It might be shielded by other means. For example, it could be that your GPU has a backplate, so when the ribbon is running between GPU and motherboard tray, and the latter doesn't have a cutout for the CPU, the cable is pretty much shielded by two grounded, highly conductive sheets of metal on both sides, and only small parts of it are exposed. That way, EMI doesn't couple through those metal parts but is redirected to common ground, where it should go. In the end, nobody can really know.

My GPU had a backplate as well, so I don't that that's what made the difference.

Hahutzy, where is your PCIe riser cable placed relative to the rest of the components in the case? My PCIe riser cable has to be routed below the GPU, so it's really close to a source of EMI interference. During my testing, I have tried taking out the GPU so that the unshielded PCIe riser cables are far away from the motherboard and the GPU, and the PC was able to boot, but it still crashes during Unigine Valley.

Remember that shielding is not only against external EMF, you also have crosstalk between the PCIe pairs themselves (Alien Crosstalk in data cabling parlance). The 3M riser (and possibly the LiHeat, I don't have one to sacrifice) uses Bonded Pairs, effectively tiny shielded coax leads in a ribbon configuration. The cheaper ribbon-cable risers are effectively bare untwisted pairs. An add-on foil shield, even with correctly terminated grounding, will do little-to-nothing to prevent crosstalk.

I've never heard of crosstalk. Can you elaborate on it a bit more?

 

[iFreilicht]

Remember that shielding is not only against external EMF, you also have crosstalk between the PCIe pairs themselves (Alien Crosstalk in data cabling parlance). The 3M riser (and possibly the LiHeat, I don't have one to sacrifice) uses Bonded Pairs, effectively tiny shielded coax leads in a ribbon configuration. The cheaper ribbon-cable risers are effectively bare untwisted pairs. An add-on foil shield, even with correctly terminated grounding, will do little-to-nothing to prevent crosstalk.

Good points, but that doesn't explain how the unshielded cables work so well in one case and so bad in the other one. I don't really want to sacrifice one of my cables, but I do have two of them so maybe I'll take a look inside.

My GPU had a backplate as well, so I don't that that's what made the difference.

Hahutzy, where is your PCIe riser cable placed relative to the rest of the components in the case? My PCIe riser cable has to be routed below the GPU, so it's really close to a source of EMI interference. During my testing, I have tried taking out the GPU so that the unshielded PCIe riser cables are far away from the motherboard and the GPU, and the PC was able to boot, but it still crashes during Unigine Valley.

Unless you've got the exact same model, we can't claim otherwise.

This is Hahutzy's case, the layout is very similar to the DAN A4-SFX: https://smallformfactor.net/forum/t...nce-densely-packed-gtx-980-ti-sub-5l-case.397
Not sure whether the cable is behind the GPU or the motherboard, though.

 

[esplin2966]

Good points, but that doesn't explain how the unshielded cables work so well in one case and so bad in the other one. I don't really want to sacrifice one of my cables, but I do have two of them so maybe I'll take a look inside.



Unless you've got the exact same model, we can't claim otherwise.

This is Hahutzy's case, the layout is very similar to the DAN A4-SFX: https://smallformfactor.net/forum/t...nce-densely-packed-gtx-980-ti-sub-5l-case.397
Not sure whether the cable is behind the GPU or the motherboard, though.

Different backplates make a difference? I guess that's a possibility. But it is indeed very odd that we're getting such wildly different results with PCIe riser cables.

 

[iFreilicht]

Not necessarily the shape of the backplates themselves, but the material and whether they're connected to ground or isolated could make a difference.

 

[Hahutzy]

My build layout step:

gpu -> backplate -> case mounting board -> PCIe cable -> mobo

Even back when I tested prototypes with my 750 Ti (no back plate, all data/power from PCIe), it ran 3dMark11 fine and did everything else fine. And that was another cheap ebay PCIe ribbon.

Spacing between components has always been ~5mm from GPU to case mounting board. Case board is directly in contact with PCIe ribbon, and then another ~5mm from PCIe to the mobo

I've only ever tested combos of Haswell CPU + Maxwell GPU on a H81 mobo

Maybe somehow, some component I chose sends/cleans up signals better than components in other people's builds?

 

[QinX]

I've never heard of crosstalk. Can you elaborate on it a bit more?

I'm not an Electronics Engineer by any stretch, but crosstalk is basically as you read it. The data that travels through a wire crosses over to another wire, disturbing/corrupting the data in that second wire. The same thing can happen with you have 20 ethernet cables bundled together, that's why they need to be shielded and sometimes individually shielded.

All wires emit some form of radiation, and depending on the frequency and power these can affect other wires. Ever tried using a 2.4 GHz wireless headset and turned on your microwave? The same can happen with regular wires.

Good shielding and proper termination are the best way to combat this. and although I don't know the theory, twisted pairs also have an effect, my best guess would be because they are differential pairs, when one is high the other is low, they cancel each other out?

Again not an Electrical Engineer but this is what I've learned over the years.

Good points, but that doesn't explain how the unshielded cables work so well in one case and so bad in the other one. I don't really want to sacrifice one of my cables, but I do have two of them so maybe I'll take a look inside.

If you want to be safe you might be able to cut it to be a 8x cable and still get some use out of it.

 

[jeshikat]

Flexible PCIe extenders can be tricky to test. An early Steambox-style case, the M3A2 tested perfectly fine for the designer, but some customers ran into issues with the GTX Titan in the field.

 

[jØrd]

could the wildly different results be related to QA / QC at the riser manufacturing stage?

 

[iFreilicht]

I've made a new thread for discussing this riser cable, you can find all the pictures there: http://smallformfactor.net/forum/threads/dissection-of-the-liheat-flexible-pcie-riser.434/

Flexible PCIe extenders can be tricky to test. An early Steambox-style case, the M3A2 tested perfectly fine for the designer, but some customers ran into issues with the GTX Titan in the field.

For anyone who's searching for the thread about the M3A2: https://web.archive.org/web/20130606192956/http://hardforum.com/showthread.php?t=1750863

 

[mitxlove]

This looks really promising! Looking forward to it! I like that it looks like receiver of some sort, will really blend into a home theater system! Great job!

 

[esplin2966]

This looks really promising! Looking forward to it! I like that it looks like receiver of some sort, will really blend into a home theater system! Great job!

Thanks!

 

[dondan]

Using a GTX 750 to test a riser is a bad choice because the GTX 750 Ti is one of the cards that don't realy use the possible speed of PCIe 3.0. But if you test it with a GTX 980, Titan XYZ s.o. you will see that your unshielded riser will fail.

The reason why the 3M riser is so good is the fact, that EVERY WIRE has a surrounded shield and not only a shield on the top and bottomside. Furthermore the cable resistent is 85ohm normal ribbon cables have 100ohm and more and will fail in some situations.

 

[iFreilicht]

Well a 750 Ti is a good choice if you want to test whether the riser can deliver the full 75W of power, but for speed I absolutely agree with you.

As the 3M riser is using micro-coax cables, the core of those wires is indeed shielded from all sides. It's just so darn expensive.

 

[Hahutzy]

The 750 Ti was long ago. I've since only tested with my 980 Ti in Hassium. And everything I said pertaining to my daily machine not having experienced any crashes/boot up failures is Hassium, not the early 750 Ti prototype.

 

[esplin2966]

The 750 Ti was long ago. I've since only tested with my 980 Ti in Hassium. And everything I said pertaining to my daily machine not having experienced any crashes/boot up failures is Hassium, not the early 750 Ti prototype.

Hmm, I'm wondering if the brand of your unshielded PCIe riser cable is super amazing? Could you reveal which PCIe cable you have and then some of us can test it? I think I might test it on my second prototype.