Question Cpu with 20+ Pci Lanes Pcie 4.0 & multi Gpu bottleneck For Gpu Rendering, Will it be Worth it.

[apjith]

Cpu with 20+ Pci Lanes Pcie 4.0 & multi Gpu bottleneck For Gpu Rendering, Will it be Worth it.
question 1 ; Its said For games 2 Gpus like rtx3090 It has stutters Even if accommodated with correctly in 40 Pci lanes on pcie 3.0. so will it be different in more cores though same clock speed & 20 Pci lanes of pci 4.0.
question 2 ; lets leave games which need full realtime horse power for real time performance. so now If Its for 3d Rendering using Multi gpu in same 20 lanes of pci 4.0 and same rtx3090 will it scale to support my render time of complex shaders & ray tracing.?

[jaslion]

Well sli and multi gpu for gaming is dead dead now so pointless anyways. For multi gpu rendering honestly bandwith doesn't matter too much. It really just depends on how quickly stuff is being processed by the card.

 

Also a 3090 is not that good of a card for rendering really. Nvidia really gimped it with their drivers and it performs worse than the last gen titan rtx in most tasks and is flat out worse than the last gen quadro in all render tasks basically. For the price of 2 might as well get a quadro and get almost identical to better performance than 2.

[apjith]

I then have another question ; 3090 piles of Cuda Cores? extra memory Pci 4.0 all adds up, is that doesnt help to compute better in comparison to similiar priced quadro p5000 16 gb pci 3? for now i only got 2k$ where in india thats price of 3090 and p5000.

Specifically for arnold gpu rendering i guess 3090 supports cause i see 2080ti does good job. and not sure for Houdini simulations like open cl or open gl im not sure how it differs but you get idea, does it helps supporting calculation of simulations. or still p5000 wins in this scenario apart from Cad & some scientific calculations. 

[jaslion]

4 hours ago, apjith said:

I then have another question ; 3090 piles of Cuda Cores? extra memory Pci 4.0 all adds up, is that doesnt help to compute better in comparison to similiar priced quadro p5000 16 gb pci 3? for now i only got 2k$ where in india thats price of 3090 and p5000.

Specifically for arnold gpu rendering i guess 3090 supports cause i see 2080ti does good job. and not sure for Houdini simulations like open cl or open gl im not sure how it differs but you get idea, does it helps supporting calculation of simulations. or still p5000 wins in this scenario apart from Cad & some scientific calculations. 

You have to look at it on a case by case basis. Yes the 3090 has A LOT more raw power compared to similarly priced quadros (why the p5000 that is from years ago I'm talking about rtx quadros) but nvidia just slows them down in the driver to prevent their gaming cards being better than their proffesional big money making cards.

 

So basically you have to look at every usecase on it's own to see if it's doing as good as a quadro of similar price or not.

[apjith]

question 1.1 ; does rendering efficiency is improvised than 2080 so thats similiar to quadro in this regard, and in houdini open cl or cuda supports calculation of simulation, is enough for artists like me. if that handles similar to quadros thats great. so if the card supports this case in particular is really good for me. But as said, for some one use it for scientific calculations and so on, then they should go for quadro am i right?

 

question 2.1 for all ; please answer my post question. does pci lanes bottlenecks performance for 3D rendering & houdini open cl Cuda ? yes i know it bottlenecks for games & frame rates. will i get its memory pooling & 20K Cuda & rtx & tensor pooling advantage somehow than a single card for 3d works. Please do answer for question 2.1.

 

 

[jaslion]

3 minutes ago, apjith said:

question 1.1 ; does rendering efficiency is improvised than 2080 so thats similiar to quadro in this regard, and in houdini open cl or cuda supports calculation of simulation, is enough for artists like me. if that handles similar to quadros thats great. so if the card supports this case in particular is really good for me. But as said, for some one use it for scientific calculations and so on, then they should go for quadro am i right?

 

question 2.1 for all ; please answer my post question. does pci lanes bottlenecks performance for 3D rendering & houdini open cl Cuda ? yes i know it bottlenecks for games & frame rates. will i get its memory pooling & 20K Cuda & rtx & tensor pooling advantage somehow than a single card for 3d works. Please do answer for question 2.1.

 

 

1: it's a fair bit better than a 2080 but it's pretty random in which case a similar price quadro is better and when a 3090 is better. You really have to look at it in a case by case basis and I cannot answer it for all your tasks as I simply do not know it for all of those and the data for it is currently scarce due to there just simply being very little 3090's available and out there.

 

2: for rendering and such don't worry. Pcie 3.0 x16 is plenty good. Even for gaming it's not an issue really. Memory pooling of cards over nvlink is a feature that is disabled on the 3090 to prevent the 3090 from eating away in the quadro space. Same for everything else in terms of hardware pooling. The usual use 2 cards on their own for rendering and such works as long as the software supports it. But this is software wise then not hardware wise which is different and cannot pool the resources like I think you want them to.

[apjith]

On 10/21/2020 at 4:09 AM, jaslion said:

The usual use 2 cards on their own for rendering and such works as long as the software supports it. But this is software wise then not hardware wise which is different and cannot pool the resources like I think you want them to.

Thanks, Im now thinking of not buying 3090 if it doesnt scale / pool it hardware resource. Please Let me know when you know some updates on it. ill check this forum & read any updates available every now and then.