Loading...
> >

Cloud rendering

2»

Comments

  • RayDAntRayDAnt Posts: 1,147
    Jack Tomalin said:
    RayDAnt said:
    Jack Tomalin said:
    RayDAnt said:
    Jack Tomalin said:
    kenshaw011267 said:
    Jack Tomalin said:
    kyoto kid said:
    Jack Tomalin said:
    kyoto kid said:

    ...for my scenes, If I can get access to a couple NV LInked RTX Quadro 8000's I'd be very happy. 

    Honest question - what would you willingly pay for that?

    ..do you have such a system?  Looking to rebuild /revise railway station and girls at the bus stop scenes (both were still WIPs which were lost in the drive crash). Then there is an idea I had that will be very "complex".  Remember a person long ago who wen't by the name of Alpha Channel? Yeah like his stuff.

    Don't even think my Titan X could handle that.

    No Quadro's at the moment, the current render server has 4 x 2080ti's, with another 3 in my work/backup server.  I was thinking of putting all 7 in one rig, then potentially adding something spicier like some Qaudro's in the backup server, and have that as a bit more an 'extreme case' server.  I heard that the memory pooling slows down the GPU's, so it might be worth trying to get the best of both worlds that way.. speed and then RAM capacity.

    The usage isn't too bad at the moment, but certainly doesn't have the userbase to really warrant that level of investment yet.

    Does iRay server support NVLink? Have you tried it on the 2080ti's?

    I haven't no, tbh I'm not sure it does support it yet actually..

    I'm not able to get to a proper computer to dig it up right now, but if you take a look at the complete Iray changelogs (not just the piecemeal excerpts of them included with each new release of Daz Studio) nvlink vram pooling has been fully implemented in Iray software-wise since the summer of 2018 when it was updated for Volta. Hardware-wise, Linux users seem to have an assured case of it working. Windows users, on the other hand, are said to be subject to limitations because of how Windows enumerates graphics memory. However there's been a lot of major code/feature changes to Iray since then, and no one with capable hardware has yet gotten around to experimenting with things.

    Even the server side of it?

    On Linux, apparently yes (remember: Iray Server has parallel Windows/Linux releases.) On Windows - that's where the big question mark lies.

    I tried it a while back, but then realised you couldn't cluster between different OS'es.. so went back to Windows on it.

    Yeah, I remember you mentioning that.

  • Hurdy3DHurdy3D Posts: 1,058
    Jack Tomalin said:
    RayDAnt said:
    Jack Tomalin said:
    RayDAnt said:
    Jack Tomalin said:
    kenshaw011267 said:
    Jack Tomalin said:
    kyoto kid said:
    Jack Tomalin said:
    kyoto kid said:

    ...for my scenes, If I can get access to a couple NV LInked RTX Quadro 8000's I'd be very happy. 

    Honest question - what would you willingly pay for that?

    ..do you have such a system?  Looking to rebuild /revise railway station and girls at the bus stop scenes (both were still WIPs which were lost in the drive crash). Then there is an idea I had that will be very "complex".  Remember a person long ago who wen't by the name of Alpha Channel? Yeah like his stuff.

    Don't even think my Titan X could handle that.

    No Quadro's at the moment, the current render server has 4 x 2080ti's, with another 3 in my work/backup server.  I was thinking of putting all 7 in one rig, then potentially adding something spicier like some Qaudro's in the backup server, and have that as a bit more an 'extreme case' server.  I heard that the memory pooling slows down the GPU's, so it might be worth trying to get the best of both worlds that way.. speed and then RAM capacity.

    The usage isn't too bad at the moment, but certainly doesn't have the userbase to really warrant that level of investment yet.

    Does iRay server support NVLink? Have you tried it on the 2080ti's?

    I haven't no, tbh I'm not sure it does support it yet actually..

    I'm not able to get to a proper computer to dig it up right now, but if you take a look at the complete Iray changelogs (not just the piecemeal excerpts of them included with each new release of Daz Studio) nvlink vram pooling has been fully implemented in Iray software-wise since the summer of 2018 when it was updated for Volta. Hardware-wise, Linux users seem to have an assured case of it working. Windows users, on the other hand, are said to be subject to limitations because of how Windows enumerates graphics memory. However there's been a lot of major code/feature changes to Iray since then, and no one with capable hardware has yet gotten around to experimenting with things.

    Even the server side of it?

    On Linux, apparently yes (remember: Iray Server has parallel Windows/Linux releases.) On Windows - that's where the big question mark lies.

    I tried it a while back, but then realised you couldn't cluster between different OS'es.. so went back to Windows on it.

    what does this mean? cluster between different os'es?

  • Daz Jack TomalinDaz Jack Tomalin Posts: 13,499
    gerster said:
    Jack Tomalin said:
    RayDAnt said:
    Jack Tomalin said:
    RayDAnt said:
    Jack Tomalin said:
    kenshaw011267 said:
    Jack Tomalin said:
    kyoto kid said:
    Jack Tomalin said:
    kyoto kid said:

    ...for my scenes, If I can get access to a couple NV LInked RTX Quadro 8000's I'd be very happy. 

    Honest question - what would you willingly pay for that?

    ..do you have such a system?  Looking to rebuild /revise railway station and girls at the bus stop scenes (both were still WIPs which were lost in the drive crash). Then there is an idea I had that will be very "complex".  Remember a person long ago who wen't by the name of Alpha Channel? Yeah like his stuff.

    Don't even think my Titan X could handle that.

    No Quadro's at the moment, the current render server has 4 x 2080ti's, with another 3 in my work/backup server.  I was thinking of putting all 7 in one rig, then potentially adding something spicier like some Qaudro's in the backup server, and have that as a bit more an 'extreme case' server.  I heard that the memory pooling slows down the GPU's, so it might be worth trying to get the best of both worlds that way.. speed and then RAM capacity.

    The usage isn't too bad at the moment, but certainly doesn't have the userbase to really warrant that level of investment yet.

    Does iRay server support NVLink? Have you tried it on the 2080ti's?

    I haven't no, tbh I'm not sure it does support it yet actually..

    I'm not able to get to a proper computer to dig it up right now, but if you take a look at the complete Iray changelogs (not just the piecemeal excerpts of them included with each new release of Daz Studio) nvlink vram pooling has been fully implemented in Iray software-wise since the summer of 2018 when it was updated for Volta. Hardware-wise, Linux users seem to have an assured case of it working. Windows users, on the other hand, are said to be subject to limitations because of how Windows enumerates graphics memory. However there's been a lot of major code/feature changes to Iray since then, and no one with capable hardware has yet gotten around to experimenting with things.

    Even the server side of it?

    On Linux, apparently yes (remember: Iray Server has parallel Windows/Linux releases.) On Windows - that's where the big question mark lies.

    I tried it a while back, but then realised you couldn't cluster between different OS'es.. so went back to Windows on it.

    what does this mean? cluster between different os'es?

    You cluster two render servers together - so I can leverage all 7 GPU's between the 2 machines.  That has it's own issues, but it's quite handy at times.

  • Hurdy3DHurdy3D Posts: 1,058

    ah... but you don't cluster anyway right? so you could switch to linux and enable memory pooling for us?

  • Daz Jack TomalinDaz Jack Tomalin Posts: 13,499
    gerster said:

    ah... but you don't cluster anyway right? so you could switch to linux and enable memory pooling for us?

    I do on occassion.

    Possibly, if it's actually deemed to work. The last time I tried, I couldnt even get Linux to report all the 11gb of RAM free, it was showing the same amount as it does on Windows... then on top of that, theres no way to connect all 4 GPU's.. so would it work with two bridges and 22gb split across the 4 cards? I don't know.

  • Hurdy3DHurdy3D Posts: 1,058

    Well, theoretically RTX support NVLink for two 2xGPUs groups:

    https://mc.ai/4x-rtx-2080-ti-with-quadro-nvlink-performance-test/

    Of course, that doesn't say if it will work with iRay pratically :)

  • DemiurgentDemiurgent Posts: 97

    For the record, Jack Tomalin's service works well and Jack's incredibly responsive. If you're still looking for a Daz-friendly iray server, he should be your first stop.

  • kyoto kidkyoto kid Posts: 41,252
    edited December 2020

    ...wow are they going to fix the quoting so that all the previous quotes are collapsed like they used to be?  This is plain ugly and as bad as trying to read the forums on a smartphone

    That said.

    Well the 3090 is out with the same memory (and more twice the cores) as the RTX Quadro 6000 and at a much lower price.  2 of these and NVlink would give you 48 GB VRAM.

    NVidia has also released the A6000 (follow on to the Quadtro line) with 48 GB of ECC GDDR6 and similar core count to the RTX 3090. for 4,650$  (96 GB VRAM with 2 thought NVLink)..

    Post edited by kyoto kid on
Sign In or Register to comment.