How to enable NVLink in OR 2018.1 XB1 ?

[glimpse]

as far as I remember to enable SLI You need at least x8 lanes. On certain motherboards You might get like x8 x8 x4 meaning if one is NVLINKed cards is placed in the last slot, SLI option might not possible.

[acc24ex]

did I just screw myself by getting the most expensive asus strix oc thinking I'm gonna be great :/ (the gainward I got first crashed my pc, so I figured lets not go cheap on this expensive thing)

- it's 3x the pci size - if I put another same card next to it, it's gonna overheat, so now I'm looking for mainbords with 4 sizes between them and a nvlink that is 80mm wide (4 slot size) .. and now I'm not sure if that 5th slot is gonna support 8x - damn

.. BTW does SLI has to be activated in order to NVLINK to work - my main concern is for it to run octane obviously, if it's only for gaming boost, it's fine I guess

..I'm planning in gettin a i9 9900k and a suitable motherboard to put up 2x 2080ti 3 slot size monsters - with a single slot space between for space .. any recomendations I found so far ASUS Z270-WS, but for some reason it is unpopular, I don't see it on the tests, and can't buy in my area ?

[J.C]

it seems required that the display is connected …

I think that is correct. You also often have to use specific slots on your motherboard.

Can you use HDMI or Displayport dummy plug display emulator? Something like this one:
https://www.amazon.com/Displayport-Disp ... B074FT1P8M

[nejck]

This might be considered necroing a thread but just to let everyone know... Users appear to be having success in getting NVLink / SLI working with non-display cards if you plug the monitor into them, enable SLI and then disconnet the monitor and plug it into the actual display card. Afaik you do need at least x8 PCI-E slots for SLI to work. You possibly also need a dummy plug for the non-monitor pair of GPUs.

[jdinh]

I read in the FAQs that SLI is advised to be disabled for performance hits. Is that not the case anymore?

[nejck]

I read in the FAQs that SLI is advised to be disabled for performance hits. Is that not the case anymore?

If your card supports NVLink and you have an NVLink bridge installed then its probably wise to have SLI turned on for memory pooling. Good question as to whether it has a performance hit without actually pooling the memory though (small scene for example)

[Timmaigh!]

I read in the FAQs that SLI is advised to be disabled for performance hits. Is that not the case anymore?

If your card supports NVLink and you have an NVLink bridge installed then its probably wise to have SLI turned on for memory pooling. Good question as to whether it has a performance hit without actually pooling the memory though (small scene for example)

I have been told on some other hardware related message boards, its not "memory pooling", what is going on with Octane and RTX Geforce cards. Apparently the memory pooling is it only in case of Titan/Quadro cards, where the operating system itself sees their memory buffers as single large one. In case of Octane its peer-to-peer, where the data from one card is copied into memory of the second card.

It may seem like a technicality, and ultimately of no relevance for us as end users, since we get larger memory at our disposal either way, but i have to wonder about those little pesky details, since they can be fairly significant... so does this mean, that in case of Octane, if you have large scene and that P2P kicks in, is then data from the second card copied into the frame buffer of the first card? Thus only one card is doing rendering at that point and the other one is just used as a memory storage accessed via NV-Link (with faster access time than if the data were stored in system RAM, as it the case of OOC?) Or do both card still do the rendering? And what about Quadro/Titan RTX, since these have that "real" memory pooling ability, does NV-Link feature work different for them in Octane?

Would love to have some light shed on this by the devs, since it is truly interesting and nobody in the community seems to have any clue, what is really going on.

[nejck]

Yes indeed, I would probably phrase that differently if I were to write that post today.

That said, I think NVLink is still in its infancy. From what I've heard Quadro NVLinks are pretty much the same, its just the software that treats them differently which means the devs have perhaps less work to do on that end. Do take it with a grain of salt, I might be right but I might also be talking out of my ass. Would be cool to hear what the (awesome) developers here think about it

[voon]

Hm, with the new gaming cards announced today, only the expensive 3090 has NVLink. If I want Octane to make use of two GPUS, can you buy two 3080's? I assume it then can use the compute power of these two cards, but not concatenate the VRAM into one large pool, while the 3090 with NVLink can both use compute power of two cards and give you more virtual VRAM?

[sreinger]

Same question here. Gives news ?
My two 2080ti support nvlink but i have never can use it
When i understand it, we need Octane 2018 for nvlink (peer?)
I have Octane 4 only. No peer function here.
SLI is activated.

Today problem with a scene, not enough memory
I hate this situation every time. I wish i can double my Vram.

When i buy two 3090 next months, then i hope it works sometime