Loading...
> >

Nvidia GPU shopping tips for large scene Iray rendering.

2»

Comments

  • kenshaw011267kenshaw011267 Posts: 3,805
    kyoto kid said:

    ...as I understand RTX 2070s are not NVLInk compatible, only RTX 2080 and on up.

    RTX 2070 super is NVlink. The RTX 2070 isn't.

  • JamesJABJamesJAB Posts: 1,766
    kenshaw011267 said:

    Those look like 80 or 92mm. That's on the small size. But a workstation running those crappy E5's and not packed to the gills with PCIE cards should still have enough airflow.

    I got curious and looked into that PSU, never seen a workstation with a server PSU and it sounded like you were describing one. But of course its Dell. That thing looks proprietary, It looks based on a a pair of 1U's side by side. If it goes bad good luck replacing it. 

    The fans in question are as follows: 1x120mm and 2x80mm.
    Also the power supply is the standard Dell server/workstation PSU.  The 1300w can be purchased used on ebay for around $75 and is super easy to swap out without taking the computer apart.

     

    TheKD said:
    JamesJAB said:
    kenshaw011267 said:
    kyoto kid said:

    ...however, AMD is useless for Iray or Octane so no point in wishing.

    As to using a Tesla card with say a Titan-X, if the paired with a Tesla M40, will Iray defer to that as the primary rendering card?

    I wouldn't count on it working at all. If it did you should be able to set which one is the primary render card.

    It is important to know that Tesla's are not intended for consumer grade motherboards. I strongly doubt you could get one up and running on a board if it wasn't in a x16 slot. Based on the experience of someone trying to get 2 working that won't work at all on consumer motherboards. So be cautious about picking up cheap Tesla's on Ebay for DS.

    Yes that may be true, but...

    If you look at the used market on ebay, you can get a well equipped dual CPU (No GPU or HDD) Dell Precision T7600 that have 4 full PCI-e 16x slots and a 1300W PSU starting under $350 after shipping.
    These are dual processor workstations that use the Xeon E5 CPUs.  The E5 line of Xeons come with 40 PCI-e lanes.  Dell set this motherboard up to use the PCI-e lanes from both CPUs
    CPU1 : PCI-e 3.0 16x, PCI-e 3.0 16x, PCI-e 3.0 4x  (full length)
    CPU2 : PCI-e 3.0 16x, PCI-e 3.0 16x
    Chipset : PCI-e 2.0 4x (full length)

    And BTW:  The cooling setup on the T7600 / T7610 is amazing.  It's quiet even when rendering with both of my GPUs running full blast.

    You got one of those for 350 on ebay? Cheapest I see on canadian ebay is CDN$1,210.51 :(

    You should look on the USA version of ebay because most of the inexpensive Precision t7600 sellers will ship to Canada. 

  • kenshaw011267kenshaw011267 Posts: 3,805
    JamesJAB said:
    kenshaw011267 said:

    Those look like 80 or 92mm. That's on the small size. But a workstation running those crappy E5's and not packed to the gills with PCIE cards should still have enough airflow.

    I got curious and looked into that PSU, never seen a workstation with a server PSU and it sounded like you were describing one. But of course its Dell. That thing looks proprietary, It looks based on a a pair of 1U's side by side. If it goes bad good luck replacing it. 

    The fans in question are as follows: 1x120mm and 2x80mm.
    Also the power supply is the standard Dell server/workstation PSU.  The 1300w can be purchased used on ebay for around $75 and is super easy to swap out without taking the computer apart.

    No, that isn't a 1U or 2U PSU. I deal with those all day long. It likely has a similiar connector but that is not a rack mount PSU.

  • JamesJABJamesJAB Posts: 1,766
    kenshaw011267 said:
    JamesJAB said:
    kenshaw011267 said:

    Those look like 80 or 92mm. That's on the small size. But a workstation running those crappy E5's and not packed to the gills with PCIE cards should still have enough airflow.

    I got curious and looked into that PSU, never seen a workstation with a server PSU and it sounded like you were describing one. But of course its Dell. That thing looks proprietary, It looks based on a a pair of 1U's side by side. If it goes bad good luck replacing it. 

    The fans in question are as follows: 1x120mm and 2x80mm.
    Also the power supply is the standard Dell server/workstation PSU.  The 1300w can be purchased used on ebay for around $75 and is super easy to swap out without taking the computer apart.

    No, that isn't a 1U or 2U PSU. I deal with those all day long. It likely has a similiar connector but that is not a rack mount PSU.

    Here, let me correct that sentence:
    "Also the power supply is the standard Dell workstation PSU".  This is the form factor that is on all of the larger modern Dell workstations (anything with an E5 V1 or newer CPU that's not a micro tower).  But they are designed like a server PSU in that there are no cables coming directly from the PSU.  Instead all cables come from plugs on the interface board that the PSU plugs into.

  • JamesJABJamesJAB Posts: 1,766

    Updated ebay prices.

    Take note of a couple of GPUs that dropped under $1000
    Quadro P5000 16GB = $970
    Quadro M6000 24GB = $997

    Prices as of 19July2019    
    List of Nvidia Large VRAM GPUs sorted by generation and VRAM size with current 
    ebay pricing range for new and used. (Used price are ebay buy it now prices)
           
    Double slot cards with 16GB and up    
    RAM   Name   New   Used
      Turing    
    48GB   Quadro RTX 8000 48GB (NVLink)  $5,500 $5,350
    24GB   Quadro RTX 6000 24GB (NVLink)  $4,000 $3,800
    24GB   Titan RTX 24GB (NVLink)  $2,499 $2,450
    16GB   Quadro RTX 5000 16GB (NVLink)  $2,300 $2,000
      Volta    
    32GB   Quadro GV100 32GB (NVLink)  $8,999 $7,000
    32GB   Titan V CEO Edition 32GB (never sold at retail)   $XXXX  $12,000
      Pascal    
    24GB   Quadro P6000 24GB  $4,099 $2,995
    16GB   Quadro GP100 16GB (NVLink)  $5,990 $3,395
    16GB   Quadro P5000 16GB  $1,899 $970
      Maxwell    
    24GB   Quadro M6000 24GB  $2,490 $997
           
    Single slot cards with 8GB and up    
    RAM   Name  New Used
      Turing    
    8GB   Quadro RTX 4000 8GB  $889 $849
      Pascal    
    8GB   Quadro P4000 8GB  $889 $590
      Maxwell    
    8GB   Quadro M4000 8GB  $760 $289
           
    Tesla Cards    
    RAM   Name  New Used
      Turing    
    16GB Tesla T4 16GB (Low Profile Single slot 70W) $2,332 $2,332
      Volta    
    32GB Tesla V100 32GB $7,397 $7,397
    16GB Tesla V100 16GB $6,718 $6,799
      Pascal    
    24GB Tesla P40 24GB $4,999 $4,195
    16GB Tesla P100  $3,563 $3,385
      Maxwell    
    24GB Tesla M40 24GB $2,500 $580
  • ebergerlyebergerly Posts: 3,255
    Keep in mind that there are multiple forms of NVLINK. I believe the supers are the SLI version which I think is unidirectional, and not suitable for VRAM pooling.
  • Takeo.KenseiTakeo.Kensei Posts: 1,303
    ebergerly said:
    Keep in mind that there are multiple forms of NVLINK. I believe the supers are the SLI version which I think is unidirectional, and not suitable for VRAM pooling.

    NVlink is always bidir

  • drzapdrzap Posts: 795
    Takeo.Kensei said:
    ebergerly said:
    Keep in mind that there are multiple forms of NVLINK. I believe the supers are the SLI version which I think is unidirectional, and not suitable for VRAM pooling.

    NVlink is always bidir

    Correct.  The difference is in the bandwidth, with the 2070 Super and 2080 being half as wide as the 2080ti.  Or something like that.

  • Takeo.KenseiTakeo.Kensei Posts: 1,303
    drzap said:
    Takeo.Kensei said:
    ebergerly said:
    Keep in mind that there are multiple forms of NVLINK. I believe the supers are the SLI version which I think is unidirectional, and not suitable for VRAM pooling.

    NVlink is always bidir

    Correct.  The difference is in the bandwidth, with the 2070 Super and 2080 being half as wide as the 2080ti.  Or something like that.

    yep, one link vs two or 50 GB/s vs 100 GB/s. Still vastly superior to PCIe 3.0 or 4.0

  • JamesJABJamesJAB Posts: 1,766

    Updated used prices and added a column for "Or Best Offer"

    Prices as of 22July2019      
    List of Nvidia Large VRAM GPUs sorted by generation and VRAM size with current   
    ebay pricing range for new and used. (Used price are ebay buy it now prices)  
             
    Double slot cards with 16GB and up      
    RAM   Name   New   Used OBO
      Turing      
    48GB   Quadro RTX 8000 48GB (NVLink)  $5,500 $5,841  
    24GB   Quadro RTX 6000 24GB (NVLink)  $4,000 $3,250  
    24GB   Titan RTX 24GB (NVLink)  $2,499 $2,450  
    16GB   Quadro RTX 5000 16GB (NVLink)  $2,300 $1,975  
      Volta      
    32GB   Quadro GV100 32GB (NVLink)  $8,999 $7,000 obo
    32GB   Titan V CEO Edition 32GB (never sold at retail)   $XXXX  $12,000 obo
      Pascal      
    24GB   Quadro P6000 24GB  $4,099 $2,850 obo
    16GB   Quadro GP100 16GB (NVLink)  $5,990 $3,395 obo
    16GB   Quadro P5000 16GB  $1,899 $1,099 obo
      Maxwell      
    24GB   Quadro M6000 24GB  $2,490 $997  
             
    Single slot cards with 8GB and up      
    RAM   Name  New Used OBO
      Turing      
    8GB   Quadro RTX 4000 8GB  $889 $825  
      Pascal      
    8GB   Quadro P4000 8GB  $889 $580 obo
      Maxwell      
    8GB   Quadro M4000 8GB  $760 $235 obo
             
    Tesla Cards      
    RAM   Name  New Used OBO
      Turing      
    16GB Tesla T4 16GB (Low Profile Single slot 70W) $2,332 $1,995  
      Volta      
    32GB Tesla V100 32GB $7,397 $7,397  
    16GB Tesla V100 16GB $6,718 $6,799 obo
      Pascal      
    24GB Tesla P40 24GB $4,999 $4,195 obo
    16GB Tesla P100  $3,563 $3,385  
      Maxwell      
    24GB Tesla M40 24GB $2,500 $580 obo
  • JamesJABJamesJAB Posts: 1,766

    Updated prices.

    Prices as of 03August2019      
    List of Nvidia Large VRAM GPUs sorted by generation and VRAM size with current   
    ebay pricing range for new and used. (Used price are ebay buy it now prices)  
             
    Double slot cards with 16GB and up      
    RAM   Name   New   Used OBO
      Turing      
    48GB   Quadro RTX 8000 48GB (NVLink)  $5,500 $5,200 obo
    24GB   Quadro RTX 6000 24GB (NVLink)  $4,000 $3,850  
    24GB   Titan RTX 24GB (NVLink)  $2,499 $2,375  
    16GB   Quadro RTX 5000 16GB (NVLink)  $2,300 $1,849  
      Volta      
    32GB   Quadro GV100 32GB (NVLink)  $8,999 $7,000 obo
    32GB   Titan V CEO Edition 32GB (never sold at retail)   $XXXX  $12,000 obo
      Pascal      
    24GB   Quadro P6000 24GB  $4,099 $2,700  
    16GB   Quadro GP100 16GB (NVLink)  $5,990 $3,195 obo
    16GB   Quadro P5000 16GB  $1,899 $1,000 obo
      Maxwell      
    24GB   Quadro M6000 24GB  $2,490 $998  
             
    Single slot cards with 8GB and up      
    RAM   Name  New Used OBO
      Turing      
    8GB   Quadro RTX 4000 8GB  $890 $890  
      Pascal      
    8GB   Quadro P4000 8GB  $889 $480 obo
      Maxwell      
    8GB   Quadro M4000 8GB  $760 $249  
             
    Tesla Cards      
    RAM   Name  New Used OBO
      Turing      
    16GB Tesla T4 16GB (Low Profile Single slot 70W) $2,332 $1,849  
      Volta      
    32GB Tesla V100 32GB $7,397 $6,995  
    16GB Tesla V100 16GB $6,718 $6,799 obo
      Pascal      
    24GB Tesla P40 24GB $4,999 $3,400 obo
    16GB Tesla P100 16GB $3,563 $3,365  
      Maxwell      
    24GB Tesla M40 24GB $2,500 $550  
  • p0rtp0rt Posts: 217
    Richard Haseltine said:
    kyoto kid said:

    ...however, AMD is useless for Iray or Octane so no point in wishing.

    If you are referring to the impact of AMD on nVidia prices that is irrelevant, it isn't sales to iray users that determine the price of the nVidia cards. If the AMD can compete in games then it may well lead to lower prices on the consumer-oriented nVidia cards, which would benefit us as a side-effect.

    kyoto kid said:

    As to using a Tesla card with say a Titan-X, if the paired with a Tesla M40, will Iray defer to that as the primary rendering card?

     

    AMD still run the business market with the EPYC super computers and vega+ cards to go alongside the openCL ProRender, which works on all GPU's, which I guess uses the openCL to Cuda bridge AMD have on the developement site 

    https://www.amd.com/en/technologies/radeon-prorender

  • ed_ulloaed_ulloa Posts: 0
    Hi All - can anyone comment on current laptops (Quadro RTX versus GeForce RTX) and any noticeable differences in render times in Daz studio? I just saw that DS version 4.12 beta has RTX capability to decrease render times. As a current GTX 1050 (2GB) user, I'm seriously looking at a few laptops and can find Quadro RTX in the $3K to $4K range right now but hesitant to spend if GeForce RTX is just as quick. Thanks and sorry if I'm posting at the wrong forum.
Sign In or Register to comment.