Building a PC from components gives you tremendous leeway, and over the years you’ve seen computers that range from humble or compact to monstrosities of performance at the CPU, GPU, and other component levels. This is exactly what a user named Faerco did when he shared his experience on Reddit after investing a whopping $35,000 in his computer.
What’s Inside a $20,000 PC. A quick look at the specs makes it clear that this PC is not designed for gaming, as many of its components are clearly aimed at the server and workstation market. In particular, the most important components are the following:
- Motherboard: SuperMicro X13SWA-TF ($1,400)
- Processor: Intel Xeon W9-3495X: 56 cores, 112 threads, 4.8GHz, 350W base TDP ($9,500)
- RAM: 1 TB Kingston DDR5-4800 ECC in 16 x 64 GB modules ($9,000)
- Boot Drive: 2 x 8TB (16TB) PCIe 4.0 NVMe SSD in RAID0 ($1,500)
- Storage Drive: 4 x 8TB (32TB) PCIe 4.0 NVMe SSD in RAID0 ($3,000)
- Video Card: NVIDIA RTX 6000 Ada Generation 48GB GDDR6 ($9,000)
- PSU: 1600W ($300)
brutal PC, brutal price. The user who set up this equipment indicated that “I’m not at liberty to say the price” but compared it to “the price of a new car” without specifying what type of car. We made a (very) rough estimate by looking at the prices of these components on Google Shopping in the US. Since it doesn’t explain some components in detail and doesn’t include others, this is only an estimate, but in the calculator the total cost is $33,700 according to our data, which could easily go up to $35,000 with these other additional items included. assemblies.
this is not to play. The hardware was not a whim, and certainly not designed for video games. The investment was driven by professional necessity, and Faerco explained that it had built it “to handle massive amounts of LiDAR data.” In particular, to conduct “interference studies in industrial environments where accuracy is critical.” He gave an example like “is it possible to put this motor through this hole or do I need to remove this pipe?”. As Faerko pointed out, this allows you to “plan and reduce downtime because you reduce the number of things that don’t need to be removed, or you find that more things need to be removed. Much cheaper to solve in a simulation than in the field where every hour counts.”
giant projects. Faerco talked about a project that already takes up 300 GB of raw data, but in the “composition” and generated scan points, the project takes up 1.3 TB … and it’s half done.
Advanced settings. He explained that the choice of RAID0 configuration is based on his need to prioritize read and write speed: with this configuration, “terabytes of data can be downloaded at once,” he pointed out.
LiDAR becomes more and more important. It’s a common technology in autonomous driving systems, but Apple has also integrated it into some of its iPads and iPhones. LiDAR sensors or scanners allow you to “clone reality” with amazing accuracy, which is very useful in the field of augmented reality – it’s not for nothing that this scanner is part of Apple Vision Pro – and photography also helps at night.
too much cpu. Faerco said the hardware is so powerful that some applications can’t even use all those resources. In his work, he uses a tool for capturing, processing and registering 3D point clouds called FARO Scene, but this application is not optimized for this number of cores.
But great evil needs great means. However, all this performance is necessary because, as he pointed out, its working accuracy is less than 1.5 mm, and “to be able to do it in a reasonable amount of time, you have to have the best of the best.”
Image | fireco
In Hatake | I assembled a beast with six RTX 3090s, but not for cryptocurrency mining, but for artificial intelligence research