Other The HDPLEX Silicon flexible PCIe riser

[iFreilicht]

Received the HDPLEX Silicon riser (click "Package contents" for more pictures) today and boy what a great riser that is.
This might be the one I've been looking for all this time.


Short facts:

• Costs about 1/3rd of the 3M riser, about the same as the LiHeat
• Is thinner than any other riser in existence
• Absolutely solid, can be folded and kinked with no performance impact
• No retention clips but higher friction force

EDIT: This review was also posted to the HDPLEX website!

 

[confusis]

Awesome This would make for an interesting front-page article with some extra writing

 

[jeshikat]

Looks very promising.

 

[Tek Everything]

I wonder why there is a performance increase when folded... That seems to run counter to what one would expect.

 

[Ceros_X]

Did you run the straight tests first, or the kinked tests?

I wonder if there were some residual heat built up from the first runs? Maybe try again straight from a cold start.

 

[Terse]

That doesn't look shielded.

 

[iFreilicht]

I wonder why there is a performance increase when folded... That seems to run counter to what one would expect.

I edited the album and added benchmark runs without a riser. It could be that the non-bent test wasn't done after a cold restart, maybe something was running in the background that impeded performance.

Did you run the straight tests first, or the kinked tests?

I wonder if there were some residual heat built up from the first runs? Maybe try again straight from a cold start.

Straight first. Well the cable is kinked now, so I'll try it again but it won't be exactly the same.

That doesn't look shielded.

It doesn't look that way, but from what I've established, this riser is indeed shielded. It is a multi-layer FPC, so the top and bottom layer would be ground (and thus shield) and the signal lanes would run in between. As @EdZ already pointed out somewhere else, the signal lines also form twisted pairs with different twisting frequencies, which reduces crosstalk between pairs and is something no other riser on the market currently has, at least to my knowledge.

Also, I've measured the thickness of the riser with a dial caliper. It's a 0.14mm thin.

 

[Phuncz]

Finally a riser that seems like a clear win in most aspects. I'm trying to find an excuse to buy one.

 

[K888D]

Is this of any use to anybody, availabel for £17.99:
https://www.scan.co.uk/products/thermaltake-pci-express-extender-cable-for-core-p5-chassis-black

 

[Ceros_X]

Is this of any use to anybody, availabel for £17.99:
https://www.scan.co.uk/products/thermaltake-pci-express-extender-cable-for-core-p5-chassis-black

Looks like it is 220mm and shielded, $30 on amazon but out of stock.

http://www.thermaltake.com/Chassis/Accessories_/_/C_00002784/PCI_e_X16_Riser_Cable/design.htm

 

[iFreilicht]

Finally a riser that seems like a clear win in most aspects. I'm trying to find an excuse to buy one.

"Research"

Is this of any use to anybody, availabel for £17.99:
https://www.scan.co.uk/products/thermaltake-pci-express-extender-cable-for-core-p5-chassis-black

This looks similar the one ModDIY is selling. Someone recently mentioned the name of the manufacturer, but I can't remember. The problem with those is that they are pretty much regular ribbon cables wrapped in aluminium foil and fabric tape, so they work fine, but are quite stiff. They can also exhibit crosstalk on the connection from the wire to the PCB.

 

[Supercluster]

I'm quite inexperienced with PCI-E extensions so please forgive my ignorance.
I have a question: is there an extender (of any kind) that retains the performance in the 99th percentile (in the order of margin of error) ?

 

[CC Ricers]

I got my "excuse" to buy one already, now I just need a job and money for all the parts I need . I also have the ModDIY one in my sights because it looks well made and presentable. But if cross talk is a problem (and I will need to wind the cable a 180 degree bend at least once) then the HDPLEX riser it is.

 

[iFreilicht]

I'm quite inexperienced with PCI-E extensions so please forgive my ignorance.
I have a question: is there an extender (of any kind) that retains the performance in the 99th percentile (in the order of margin of error) ?

Good question, but it is extremely hard to test. As you saw from the benchmarks I made, the margin of error already is quite high. Personally, I haven't tested the 3M riser but the current consensus is that it is the best (and most expensive) you can get.

The other problem with "performance" is that there's perceived vs. measured performance loss. It might be that you lose 100 points in Heaven and 2 FPS in your games benchmarking tool, but if your GPU is capable of running the game at 70 FPS, you're not going to notice the difference. If you really want to get every last bit of performance from your PC, though, long flexible risers are certainly not the way to go.

 

[EdZ]

I'm quite inexperienced with PCI-E extensions so please forgive my ignorance.
I have a question: is there an extender (of any kind) that retains the performance in the 99th percentile (in the order of margin of error) ?

In theory, the performance loss should be zero. Either the link works, or it doesn't and you get ungraceful failures like crashing or graphical corruption. GPU testing has such high variance doe to changes in run-to-run environments (for non-canned benchmarks), system load (background processes), local heating, external temperature and airflow, etc. This is born out in testing of risers.

 

[iFreilicht]

BTW, after taking a closer look, I've found out that the riser is in fact NOT shielded, and it is not a multi-layer FPC. It's just double sided. Between the two sides, the differential pairs are staggered, so crosstalk isn't going to be an issue, but EMI in general could be, as each pair is only shielded from three directions (back and both sides, not the front).

How dangerous is it to run a microwave while it's open? Could that be a good test for EMI problems?

 

[QinX]

running a microwave upon isn't a good idea, but also wouldn't be a good test.
A microwave emits around 2.4Ghz which is why with some wireless headphones you get a bad connection when standing close to one.

But EMI isn't just a single radiofrequency. It's multiple and at varying strengths. You'd need to run a sweep to see what frequency the PCIe riser is sensitive at and also at what angles.

A simple example could even be your regular AC wiring. Although only 50/60hz a multiple of that frequency can end up being the cause of problems.
The frequency of the motor in a fan or a particular inductor can be the cause of issues.

 

[iFreilicht]

So I've been having a bit of correspondence with one of the guys at HDPLEX (great service, BTW, they answered my query in less than 10 minutes on a Saturday) and asked them a few of my questions. The less technical ones I already got answers for.

Firstly, GPUs without PEG connectors:
Multiple GTX 750Ti models without PEG connectors were tested by them, no problems were encountered, 75W power delivery shouldn't pose a problem. The new GTX 950s with slot-only power were not tested as the guys couldn't get their hands on them yet. I also measured the 12V delivery trace on the FPC, it's a whopping 7mm wide on the top and 6.5mm on the bottom, much wider than the pinout would require. I guess that can only help.

Secondly, the strange connector solution:
Because the riser is FPC based, you can't get the signals to both sides of the PCB on a card edge connector easily. That's why they have female connectors on both ends that are soldered directly to the FPC, the PCB is only there for physical stability. All other risers don't have this problem because they use two ribbons for each side of the PCB.

Thirdly, custom length:
As expected. Possible, but only with a very high MOQ.

Answers to EMI and some design decisions should be expected from one of their engineers within the next few days.
Pretty nice to work with them so far.

They also read this post and apologised for the damage on the riser, promising to improve the packaging on future deliveries of this. To be fair, they probably only shipped it with the H5 so far.

EDIT: Thanks QinX for explaining that a bit!

 

[jeshikat]

That's great you can contact them and get technical answers back.

 

[CC Ricers]

GTX 750Ti's have a lot of headroom since their power is capped by its BIOS at 38w. It will not pull more than that from the bus. It remains to be seen how much actual power draw the slot-powered 950's will do, but I feel confident that those risers would be able to handle it.

On a different note, anyone know where I could find a good and affordable shielded flex riser (like the HDPLEX one) that is very short, like 5cm? I would need one to fit in my slim case if I cannot flip the GPU completely around.