PCI-e bandwidth/capacity limitations

[SteveWillis]

I'll just mention that my system/motherboard won't recognize all 4 1080 TIs unless I use powered risers. I'd love to find some good ones.

[bruce]

Unpowered risers draw power from the motherboard and the total for all PCIe slots is limited to 75w. I'm surprised that you can run several.

Even with power to the external power plug, I suspect that some power is drawn from the slot itself -- and I think you've just confirmed that. I have no idea which power would be drawn from the powered riser but it's apparently not zero.

[Aurum]

I'll just mention that my system/motherboard won't recognize all 4 1080 TIs unless I use powered risers. I'd love to find some good ones.

May I suggest a MB with a supplemental power plug?
https://www.newegg.com/Product/Product. ... 813128772R

[SteveWillis]

Mine has that but it's just a plain molex. Still if I don't use a powered riser on at least one GPU only three get recognized.
Also I was kind of limited in choices as I wanted AMD with at least 4 16X pcie slots

[Aurum]

Hmm, that's odd. Sometimes a molex pin pushes back out and does not make electrical connection. What's the MB model? Are all 4 folding now?

[Aurum]

Has anybody ever tried this EVGA Power Boost
https://www.evga.com/products/product.a ... MB-PB01-BR

[SteveWillis]

Has anybody ever tried this EVGA Power Boost
https://www.evga.com/products/product.a ... MB-PB01-BR

I tried one and the motherboard would never boot again. Had to replace it

[SteveWillis]

Hmm, that's odd. Sometimes a molex pin pushes back out and does not make electrical connection. What's the MB model? Are all 4 folding now?

I had the same experience on two different rigs using the same MB It's kind of out of date but I liked it. ASUS Crosshair V Formula-Z Republic of Gamers, AM3+, AMD Motherboard

The one I killed was a different board

[SteveWillis]

Has anybody ever tried this EVGA Power Boost
https://www.evga.com/products/product.a ... MB-PB01-BR

I tried one and the motherboard would never boot again. Had to replace it

However I Googled a lot and never found another report of that

[foldy]

I know if you run fast GPUs on pcie 2.0 x1 using WINDOWS then your PPD is reduced significantly.
If you run a GTX 1070 or GTX 1080 Ti on pcie 2.0 x1 risers using LINUX, how much PPD would you get?

Can I interpret these results from @rwh202 that on LINUX you can run a gtx 1080 on x1 risers with 90% performance?
viewtopic.php?f=38&t=28847&p=292130&hilit=riser#p292130

I'm asking because coin miners typically run many GPUs using x1 risers and want to evaluate if they could run folding@home efficiently.

[SteveWillis]

My 1080 on a 2.0 x1 typically gets better than 600K PPD. my other 1080s normally get 750-800K so I'm estimating I lose about 9% on that card. Linux Mint.

[foldy]

@SteveWillis: Is this with USB x1 risers? Anyone tried USB risers for x1? Does USB risers hurt performance more compared to flat cable risers?

[bruce]

IN THEORY: the ribbon cable version should be slightly faster. While the x1 bandwidth would be the same, there would be a bit more latency because the parallel bits would have to be serialized and then serialized.

The ribbon cable would be more susceptible to noise/interference but over that short distance, that's probably insignificant.

I've use both and have never measured a difference. The USB variety just "feels" better, but that's probably just my prejudice.

[SteveWillis]

@foldy, Yes with a usb riser

[rwh202]

AFAIK, 'USB' risers should be nigh on identical to the ribbon versions.
They are not USB nor serial - it's just a cheap, convenient cable and connector type with enough conductors and shielding to support PCIE 1x.