Yeah but there is your bottleneck.
If you build a render system with three 2080ti´s,
then you also want as fast as possible nvme drives.
And sure the the first m.2 gets it’s own 4x dedicated lanes.
But you likely would wanne use as much fast nvme storage as possible.
Or in other words, more then one highspeed nvme drive.
Also yeah pci-e power might be a concern as well indeed.
So yeah in my opinion HEDT would still be the way to go here.
aBav… You have just made things a little more interesting for making the ACE an option still. The last issue for me is the possible PCIe power; I might consider under volting and clocking the cards just a bit.
MisteryAngel, I’m not rendering animations. I want to render single 6k or 8k frames. All my 3D projects and files will be stored on a nas to prevent file lost. I will connect by either ethernet (I think it’s 2.5gbps maybe 1gbps) or usb.
I’ve just joined this site a few day ago and so far its been great! Thank you all for a the help!
Yes i understand that.
But you are not rendering directly to the nas i suppose?
So yeah, that’´s what i mean by the limitations of a mainstream platform.
I would suppose you prefer to use more then one super fast nvme drive.
And then you pretty much run out of lanes.
Unless the render output files aren’t that big,
and you put them right on the nas when finished.
Then you could eventually work with just one nvme ssd.
Just out of curiousity: Are there any benchmarks how the actual used PCIe bandwith affects a GPU’s performance when using them for the rendering purposes you detailed?
The very small differences make me wonder how an X570/3950X system with 7 GPUs @ PCIe 3.0/4.0 x4 would perform (PCIe bifurcation and a contraption with active risers).
I’m planing to mostly use LuxCoreRender which uses Open CL and will likely use other GPU Renders as well. 3.0 PCIe 4x will work for Lux at a similar speeds to 16x according to one of Lux’s Developers:
If I were to bifurcate, the only closed case option to support more that 4 gpu’s I know of is the Rosswill GPU miner… That’s a maybe. Also, If I bifurcate won’t that increase the power draw on the PCIe slot?
Gold_Two
Power usage bill for 12 gpu’s is not an option me lol! Maybe 4 or 5, calm down man lol!
You AMD people have unbelievable amount of problems there is also a solution that support all of those have been ten years it is called Intel Xeon. My system is from seven years ago launched dual xeon e5-2695v2 12c/24t 2.79ghz all core turbo 3.2max hp z820 and it takes four cards and no lane is chared with some problem causing other device actually even more but then have to bother to use risers
Lol, old school Intel Xeon’s are ok; but I’m planning to do cpu video rendering too. Since the AMD 3900x 12c/24t base is 3.8, boost is 4.6, and is at 105 watts; it easily beats the 2695v2 in performance at a lower wattage.
Also I’m running an e5-2690v2, 40 PCIe lanes is nice, but I want a new motherboard and PCIe 4.0. Also with Bifurcation x4 I can run up to 6 GPUs for 3D rendering with no performance hit for what I’m doing.
Old school Intel Xeon isn’t gonna cut it this time.
An update regarding the Pro WS X570-ACE’s U.2 port. I wanted to try to use it for a regular PCIe AIC with an active riser from Delock, however the motherboard doesn’t POST as soon as the riser is connected to the U.2 port (even if nothing is installed in the riser).
Quite weird since the same U.2 cable and riser work fine if you use them with additional adapters getting the PCIe lanes from the CPU and not the X570 chipset.