How to make one powerful from two PCs?

Question

linux

Hello!
Advise how to make one supercomputer out of two powerful computers.
Computer configs:
1)
CPU Type HexaCore Intel Core i7-4960X Extreme Edition, 3600 MHz (36 x 100)
Motherboard Name Asus Rampage IV Black Edition (2 PCI-E x1, 4 PCI-E x16, 8 DDR3 DIMM, Audio, Gigabit LAN, WiFi)
Motherboard Chipset Intel Patsburg X79, Intel Ivy Bridge-E
System Memory 65475 MB (DDR3-1600 DDR3 SDRAM)
Video Adapter NVIDIA GeForce GTX 650 (2 GB)
2)
CPU Type HexaCore Intel Core i7-4960X Extreme Edition, 4000 MHz (40 x 100
) Asus Rampage IV Black Edition (2 PCI-E x1, 4 PCI-E x16, 8 DDR3 DIMM, Audio, Gigabit LAN, WiFi)
Motherboard Chipset Intel Patsburg X79, Intel Ivy Bridge-E
System Memory 65475 MB (DDR3-1600 DDR3 SDRAM )
Video Adapter NVIDIA GeForce GTX 650 (2 GB)
There is also an old server (it makes a lot of noise), you can use it.
CPU type 2x QuadCore Intel Xeon E5405, 2000 MHz (6 x 333)
Motherboard Intel Alcolu S5000PAL (1 Adaptive Slot, 1 PCI-E Riser Slot, 8 FB-DIMM, Video, Dual Gigabit LAN)
Motherboard chipset Intel Blackford 5000P
System memory 8185 MB
Both are used for calculations in ANSYS HFSS, recently there were not enough computer resources (from the latter, the simulation did not start because it required more than 64 GB of memory).
1) I would like to get a configuration in which 12 cores and 128GB of RAM for applications were displayed in the task manager. Applications used these resources, how is it possible for one PC?
I have basic skills in working with Linux and Win server.
2) Googled about Open Mosix. Is it possible to deploy a cluster using Mosix, and, for example, install Proxmox + Windows on top.
I would like to load all the cores with one task. In the future I plan to render 3D on it.
Has anyone experienced this and what options do you recommend?

Answer 1

nothing will work out.
The communication speed between two computers, even in terms of gigabit, is several times lower than the speed on the bus, and even more so between the cores of the same processor, so for such computers you can separate tasks, not processes (compressed one video on one computer, another on another).

Answer 2

I semi-agree with the previous speaker: you can’t pull anything out on copper (unless it’s an infinband), so we take two two-headed optical cards from Intel for 10 gigabit and connect the two boxes with a bond.
Then we get ten in one direction and ten in the other on different fibers, or 20, but with a large number of losses for service traffic + increased latency.

Answer 3

rus-linux.net/lib.php?name=/MyLDP/server/rocks.html
cluster.linux-ekb.info/specdistr.php

Answer 4

3D renderers are already able to parallelize tasks over the network