How many PCIe Extensions is TOO MANY??

[nicklmg]

Buy PCIe extensions
Amazon: http://geni.us/Ej85Tkv
Newegg: http://geni.us/Xu5qQN

 

Can PCIe extensions cause a loss in GPU performance? We investigate!!

 

 

[NvidiaIntelAMDLoveTriangle]

Pointless video is pointless.

[nicklmg]

Just now, NvidiaIntelAMDLoveTriangle said:

Pointless video is pointless.

Pointless comment posted 1 minute after a ~10 minute video is set live is pointless.

[Glenwing]

3 minutes ago, NvidiaIntelAMDLoveTriangle said:

Pointless video is pointless.

Actually I'm curious to know about this

[theminnesotanman]

6 minutes ago, NvidiaIntelAMDLoveTriangle said:

Pointless video is pointless.

gets cucked by admin because point made is pointless

[Guest]

You've given her all she's got captain!

[Imbellis]

Nice! I've been recommending this for a while!

[AshleyAshes]

6 minutes ago, Glenwing said:

Actually I'm curious to know about this

Yeah, while PCIE may be a bit extreme, testing 'practical maximum cable length' is an interesting metric since most people quote on paper numbers which are usually VERY conservative to offer a 100% chance of everything working right.

 

I'd actually enjoy seeing the max cable length tested on a range of cables.

[Mira Yurizaki]

13 minutes ago, NvidiaIntelAMDLoveTriangle said:

Pointless video is pointless.

Sometimes the greatest discoveries were about answering the most dumb sounding questions.

 

Not saying we're on the verge of a revolution here, but it's always nice to answer said dumb sounding questions.

[Daniel644]

1 hour ago, NvidiaIntelAMDLoveTriangle said:

Pointless video is pointless.

not really, lots of people ask about this on here.

[fourwordsalluppercase]

Next - Linus makes a CPU riser, and then a RAM riser and then his entire computer is split between 18 rooms (+RGB)

[Flavio hc 16]

actually, it was a very interesting video

[AshleyAshes]

1 minute ago, michaeldpgrace said:

then a RAM riser

[minibois]

Using Thermaltake risers.. I too like to live very dangerously.

But I like these type of videos. Just trying stuff to see how it turns out.

 

I like the banter between Linus and Ivan too.

[Progressor]

45 minutes ago, AshleyAshes said:

Jesus fucking Christ! What kind of sorcery is this. Had no idea this was possible, thanks for sharing, TIL.

 

Anyway, awesome video. I was always wondering how much do these extensions affect performance. Looks like as long you buy a fancy one and not the cheap gray ones off eBay, you're all good.

 

[AshleyAshes]

1 minute ago, Progressor said:

Jesus fucking Christ! What kind of sorcery is this. Had no idea this was possible, thanks for sharing, TIL.

 

Anyway, awesome video. I was always wondering how much do these extensions affect performance. Looks like as long you buy a fancy one and not the cheap gray ones off eBay, you're all good.

 

Nope, these were cheap ones off eBay. FYI, they've run in my server for 18 months without issue.

 

 

[Streetguru]

2 hours ago, nicklmg said:

 

Are you guys going to do the same with external GPU docks? Just need a bunch of powered USB-C extensions.

[Progressor]

1 minute ago, AshleyAshes said:

Nope, these were cheap ones off eBay. FYI, they've run in my server for 18 months without issue.

 

 

In regards to the performance I was talking about Linus's video. Anyway, it's good to know that the ram extensions work flawlessly too. Thanks for the info, I wish I was aware of this a month ago when I was shopping for second hand desktop ddr3, I would have went with laptop ram which I found was much cheaper.

[maksakal]

From what I understand FurMark doesn't stress the databus as much. Becuase it's designed to overheat the GPU it will load into GPU memory and go from there or CPU would be a factor too. On some systems it would stress out the CPU before GPU. This could explain why it's stable as there isn't as much data being transfered.

 

Would've liked to see a game.

[HWTag2]

3 hours ago, nicklmg said:

Pointless comment posted 1 minute after a ~10 minute video is set live is pointless.

REKT ROASTED LIKE A PEANUT

 

(nice one Nick)

[NvidiaIntelAMDLoveTriangle]

3 hours ago, nicklmg said:

Pointless comment posted 1 minute after a ~10 minute video is set live is pointless.

Have you considered that I've watched it somewhere else? Or that I've skipped through the video and got the conclusion quickly?

[Snowmirage]

Can someone PLEASE try with an RX 480 or 580?

I'm currently trying to use a perfectly reasonable length (~7") with an RX 480 and getting crashes as soon as I try to install the drivers (works fine directly in the slot).
The extension I'm using is similar to this


If I use one of these 

It works but I suspect an RX 480 might get bottle necked on PCI-E 2.0 1x (I'll try to do some benchmarks to test that soon)

Building a 2 gamer 1 CPU + other stuff machine with an EVGA SR-2 board 

From my research it seems the RX 480 cards draw a LOT of power through the motherboard (more than the PCI-E calls for)

I ordered a couple different replacement extentions



And this thermal take (different than what LTT used in the video I think)

But they are taking forever to ship

My theory is I'm getting a power drop across the extension but I don't know that a higher quality or shielded cable will make a difference.  Maybe the auxiliary molex connector will help it seemed to work on that 1x extension I tried.

[maksakal]

3 hours ago, NvidiaIntelAMDLoveTriangle said:

Pointless video is pointless.

I've built 3 custom cases until now, last 2 could've been better if I used a PCI-e extension. Held of because extension cables have a bad rep, Linus made a video on that way back. Good to see reliability has improved on these cables.

[sydh]

Well that was kinda of interesting. Totally random but yet interesting.

[mariushm]

As usual, loads of problems with this video as well.

 

You guys managed that long distance because several reasons:

1 extensions were well made and SHIELDED

2. extensions were kept relatively flat and in one direction only (no bends, no twists)

3. the pci-e 3.0 controller is probably way more powerful than the pci-e 2.0 controller on the other card.

 

By keeping them flat, you basically kept pretty much the same distance between data signals (there's data signal pairs and then ground wire, and repeat for up to 16x) , so if the signal quality went down, it kinda went down the same amount on all lanes.

Had you twisted the extensions or shaped them in any way without good shielding that extension would have become an antenna.

There were also few devices between the computer and the video card that would affect the signal. I suspect even something as simple as the monitor or the power supply between the video card and the computer would have caused enough noise (leeching into the extensions through the weak points, the unshielded slot / connector mating points)

 

There's one more thing... there's no numbers about how reliable the connection was.

Yeah, you got the pci-e 3 controller working but there's no data about how many failed transactions have occurred on the pci-e channels. For all we know, by the 2 meters point, maybe 20-30% of all data packets were corrupted and had to be re-transmitted but they were re-transmitted correctly so fast the benchmark wasn't slowing down at all.

 

Also, no mention if the video card / driver maybe switched down to pci-e 2.0 or even 1.1 due to too many transmission errors, but kept the power of transmission and receiving to 3.0 levels. The pci-e 3 controller should have more power compared to pci-e 2 controllers and i wouldn't be surprised if they also have better error correction and recovery compared to controllers from pci-e 2.0 times.

I suspect that Furmark benchmark doesn't really use that much pci-e bandwidth.

 

Maybe a better benchmark would have been transferring a huge amount of data using a 10gbps or 40 gbps ethernet card (i think you guys have two 40 gbps cards around there.. well 32 gbps effective ... that's good enough of at least pci-e v2.0 x8 worth of throughput)