A customer is buying a powerful computer for AI content. The machine has more video memory (VRAM) than RAM for a good reason. In AI, CPU and RAM play a lesser role, while the graphics card and its VRAM are much more important.
A user reports on Reddit that a friend of his works at a PC company that professionally assembles PC systems for buyers. A customer is said to have ordered a special PC system: a machine with 4 RTX-4090 graphics cards and 64 GB of RAM.
The special thing about the system: The 4 graphics cards alone have more video memory than the entire computer has RAM. A standard RTX 4090 has 24 GB of VRAM, which totals 96 GB of video memory.
How much did the system cost? The user explains that the customer paid about 16,000 US dollars for the PC, including assembly, shipping, and warranty. This is approximately 14,000 euros.
What is not known? It is not known which CPU, which motherboard, how much system memory, and what kind of case are installed. According to another comment on Reddit, the case is said to be a specific custom model from a Chinese manufacturer. The motherboard offers 8 slots for RAM, so it is likely to be a specific motherboard.
Generative AI relies on fast graphics cards and readily available video memory
And what is the expensive hardware used for? According to the user’s statement, the customer wants to use the system to deploy AI models locally. This at least explains the need for the graphics cards.
Why is VRAM more important than RAM? The computer is primarily intended for generative AI use. A lot of RAM does not bring significant advantages in generative AI. RAM only becomes important, as a user explains on Reddit, when you want to save intermediate steps, called checkpoints.
Video memory has the advantage that you can already fit the entire AI model into this memory. The important and large models are then at 20 GB and more. Processor and RAM play a less important role here because VRAM is significantly faster than normal RAM and is much more directly connected to the graphics card and its graphics processor.
Neural networks, for example, require many parallel calculations simultaneously, which graphics cards are particularly good at. The more video memory is available, the faster and easier it goes. Running these calculations on a CPU that then accesses normal RAM would be significantly slower.
Community considers the computer a fair offer when observing high prices
Why is this a good offer? Several users explain that 14,000 euros for such a system is quite fair, considering the current market situation. Particularly in light of the rising prices for hardware. Because it is still unclear when prices will stop rising.
Just the four graphics cards and 64 GB of RAM are expected to be significantly more expensive in a few months than they are currently. One user writes that the offer is better than he thought. Another adds: “With the 5k cards that are coming soon, it sounds almost like a good deal.”
A user bought 1 TB of RAM for around 1,400 US dollars. A few months later, the hardware is worth 10,000 US dollars to his surprise. He wanted to sell it much earlier but forgot about it: A user bought 1 TB of RAM and 20 TB of storage for 1,200 euros a few months ago: Today the hardware is worth more than 8,500 euros
