Linus Tech Tips - How many PCIe Extensions is TOO MANY??

[Nookyava]

[video=youtube;q5xvwPa3r7M]https://www.youtube.com/watch?v=q5xvwPa3r7M[/video]

This is to the point that it's absurd, why would you ever need that many.

Have we really strayed that far from the light that we need to create such a monstrosity?

[Michael haxz]

That's actually pretty impressive. Imagine having all the GPUs to all your systems in air conditioned centralized location. You might as well put the entire system there but still it's pretty cool.

[Jelman]

He notes this in the video but this is with Shielded PCIE extenders.

With unshielded he'd have issues by the second or third

[meppers]

those cyptominers will find this handy after they buy that new asus mining board

[t]http://cryptomining-blog.com/wp-content/uploads/2017/08/asus-b250-mining-expert.jpg[/t]

[Johnny on Flame]

Is the furry torus hellspawn a big enough bandwidth test? I imagine that with better benchmarks he'd start getting more fps loss.

[Big D Duchovny]

Obviously no one should do this anyway, but he's running a benchmark where I'd imagine there's very little to do on the PCI-E bus after stuff has been loaded into VRAM, so saying there's no performance degradation is, well, technically true in that benchmark, but still not very rigorous. Not like it matters, though.

[Adam1172]

[QUOTE=Michael haxz;52634444]That's actually pretty impressive. Imagine having all the GPUs to all your systems in air conditioned centralized location. You might as well put the entire system there but still it's pretty cool.[/QUOTE]

Or you can do the opposite, get extensions for everything so that each component is in a different room.

So where's your PC?
Oh didn't you know? My whole house IS the PC.

[paindoc]

[QUOTE=GoDong-DK;52634654]Obviously no one should do this anyway, but he's running a benchmark where I'd imagine there's very little to do on the PCI-E bus after stuff has been loaded into VRAM, so saying there's no performance degradation is, well, technically true in that benchmark, but still not very rigorous. Not like it matters, though.[/QUOTE]

Have to agree, with how important memory bandwidth and latency is nowadays he should've tried something with much more texture streaming going on: Titanfall 2, for example, managed to bring my 1070 to it's knees at the highest texture settings due to how much streaming it does. I imagine any extra latency might be measured in percentage points at most, since transferring through PCIe and/or using DMA is still considered fuckin slow

[Mitsuma]

[QUOTE=meppers;52634609]those cyptominers will find this handy after they buy that new asus mining board

[t]http://cryptomining-blog.com/wp-content/uploads/2017/08/asus-b250-mining-expert.jpg[/t][/QUOTE]

I would assume that PCIe 2x or 1x might have lower range.
It might also have other implications that Linus did not test.
All he did was a Furmark stress test. There was no test that could display latency issues or other potential problems.

[pebkac]

[QUOTE=paindoc;52634688]Have to agree, with how important memory bandwidth and latency is nowadays he should've tried something with much more texture streaming going on: Titanfall 2, for example, managed to bring my 1070 to it's knees at the highest texture settings due to how much streaming it does. I imagine any extra latency might be measured in percentage points at most, since transferring through PCIe and/or using DMA is still considered fuckin slow[/QUOTE]

Or better yet, get one of those PCIe SSDs and you can then directly measure how the length of the extension affects the transfer speeds.

[paindoc]

[QUOTE=pebkac;52636429]Or better yet, get one of those PCIe SSDs and you can then directly measure how the length of the extension affects the transfer speeds.[/QUOTE]

that's actually a really clever thought, and now im honestly fairly curious to know if it does affect transfer speeds in a measurable way

[Joey Skylynx]

[QUOTE=Thomo;52634434]This is a good idea if you need to heat another room of your home.[/QUOTE]
Run a mining rig with all the GPUs put into your closet with a fan, and some beef jerky racks.
It's like knocking two birds with one stone!

[pebkac]

[QUOTE=paindoc;52637825]that's actually a really clever thought, and now im honestly fairly curious to know if it does affect transfer speeds in a measurable way[/QUOTE]

When there's enough signal degradation to cause a significant number of corrupted packets, I bet there'd be a noticeable difference. I mean you can see it even in the video, once they go past 3m they're getting huge framerate drops, seems like at that point there's barely any data getting through uncorrupted.