titan x pascal 7x Titan X Pascal 1x CPU

[Carbix]

So I need to render complex 3DS MAX stuff over night. Normally would take a week to render. The idea is that by using the NVLink chip on the Asus x99 WS 10G, vRay RT will be able to stack the ram and render massively complex environments. We have been limited to CPU only because of the memory constraints.

 

NVLink requires 16x PCIe (my understanding) the MB comes with 4 16x and 3 8x. This MB has an NVLink chip up converter left of the north bridge allowing for 8x to interact with each other at 16x levels. NVLink in theore alowes video cards to share each other assets entirely. Where SLi is like hand shaking and helping each other out studying after school, NVLink is like Goku and Vegeta Fusion. I should name this build Gogeta. An 84gig GDDR5 video card sounds really cool. 

 

I'm going with the i7-6850k because it's one of the cheapest 40 lane chip.

 

ram I'm not hung up on. 

 

Anyone won't to vet this idea before I drop $16k+ to make this build. 

 

CPU: Intel Core i7-6850K 3.6GHz 6-Core
CPU Cooler: Ek water buildout for GPU's and CPU. 
Motherboard: Asus X99-E-10G WS SSI CEB LGA2011-3
Memory: 2x G.Skill TridentZ Series 64GB (4 x 16GB) DDR4-3200
Storage: Samsung 960 PRO 512GB M.2-2280
Storage: 3x Seagate Barracuda 3TB 3.5" 7200RPM
Video Card: 7x NVIDIA Titan X (Pascal) 12GB
Case: Case Labs SMA8 Gun Gray. 
Power Supply: 2x Corsair 1200W 80+ Platinum (I may have to upgrade this)
Operating System: Microsoft Windows 10 Pro.

[Electronics Wizardy]

You can stack ram. Thats not how cuda works.

 

NVLink isn't pcie and a different connector, it is only found on POWER chips if it connects to the gpu, and you can get the nvlink cards in the pci card formfactor.

 

Id probably just get a worstation or server made for gpu rendering. If you really want nvlink, they have some systems with dual POWER8 chips and 8x p100's.

 

 

[IHirs]

go for 4 hhd so you can get a raid 10 setup 

[Carbix]

24 minutes ago, Electronics Wizardy said:

You can stack ram. Thats not how cuda works.

 

NVLink isn't pcie and a different connector, it is only found on POWER chips if it connects to the gpu, and you can get the nvlink cards in the pci card formfactor.

 

Id probably just get a worstation or server made for gpu rendering. If you really want nvlink, they have some systems with dual POWER8 chips and 8x p100's.

 

 

Yeah we have stuff like that and the $40k price that came with it. This was to be a "cheap" work around. After your comment I found on some white papers that NVLink will be a new connector type and at that only the Tesla lines will come with NVLink. That makes Titan X (P) just a big tease. At least to me.

 

damn I really wanted to do this build. It still would be fast and can be done. Just will need to make the renders less complex.  T-T

 

Thanks!

[Electronics Wizardy]

2 minutes ago, Carbix said:

Just will need to make the renders less complex.  T-T

Or make a render farm? Idk about power and space, but your can get something like a dell r5500 for about 300 and it takes 2 dual slot gpus in a 2u. Stack them up and that should chew through almost anything and is fairly cheap.