Have you seen the specs? Consumer RTX 5090 is faster and cheaper than the Furiosa RNGD Gen 2. You gotta be mad stupid to buy something that performs worse and it is 5 times more expensive.
They just declined Meta's $800 million offer. What are they smoking? I just saw the specs and nothing is special about the Furiosa RNGD Gen 2 card compared to the RTX 5090. Sure, it has more SRAM, but that is not a deal breaker. The same goes for power consumption, data centers have incentives for power.
If each Furiosa RNGD Gen 2 card costs $10k while an RTX 5090 costs $2k, and the RTX 5090 has better performance for LLMs, you have to be mad stupid, have a personal grudge against Nvidia, or just want to burn cash for no good reason to rack up your data centers with Furiosa.
The value of their company is going to diminish and their next offer won't go over $1.5 billion. It will actually be less than $800 million since every year Nvidia, Intel, and other AI hardware startups introduce a better and faster card.
If Furiosa cards magically became cheaper than Nvidia's similar hardware, Furiosa might be worth a quarter billion dollars. I highly doubt this would ever happen because making AI compute with cutting edge lithography is hella expensive and involves heavy politics.
No wonder cutting edge LLMs themselves are so confused with DX12 APIs... I guess I should stick with CUDA, they have so good documentation and support on everything all in one place.
Only reason I need to use DX12 is because the swapchain uses DXGI textures. I hope Microsoft gives NVIDIA great priority and leverage to use CUDA native textures at the swapchain directly in Windows.
DX12 functions ported directly into CUDA will be a big plus, programming stack is literally next to the NVIDIA hardware, do everything under one language and don't need to mess with two language interoperability.
Deepseek is indeed better than Mistral and ChatGPT. It has tad more common sense. There is no way they did this on the “cheap”. I’m sure they use loads of Nvidia GPUs, unless they are using custom made hardware acceleration (that would be cool and easy to do).
As OP said, they are lying because of export laws, they aren’t allowed to play with Nvidia GPUs.
However, I support DeepSeek projects, I’m here in the US able to benefit from it. So hopefully they should headquarter in the States if they want US chip sanctions lift off since the company is Chinese based.
But as of now, deepseek takes the lead in LLMs, my goto LLM.
Sam Altman should be worried, seriously, Deepseek is legit better than ChatGPT latest models.