Unfortunately, the NVIDIA 5090 GPUs are not available, but the real star of CES 2025 is—Jensen Huang's Tom Ford leather jacket, now as a LoRA.
Trained with OSTRIS AI-Toolkit @ runpod.io
Description
Comments (9)
LOL that jacket looked so rebelious yet so funny also when he wore it for the presentation. For someone who is a futurist or visionary, he sure has 1980's fashion sense. Just as long as he doesn;t ever wear skivvys with denim jeans. Steve Jobs had that image exclusive.
Supposedly his wife and daughter dress him, or did. The 80's fashion sense doesn't surprise me. We were born in the same year.
You made me check his wiki. He started from nothing, possibly middle to lower class background. Changing subject, when the company recently lost a huge chunk of stock in 1 day everyone reported it was due to DeepSeek's models released using not expensive high end GPU's to train LLM but that is only half of the equation. DeepSeek signed up with another GPU tech company that developed a new type of GPU architecture that is so much faster than nvidia GPU (which has bottlenecks) to run these latest LLM models. Thats the real reason, nvidia is not the marlet leader for running LLM. But when they design these new GPUs the other company is doing, then nvidia is safe from losing stock.
@xlr8td What other mysterious company is it supposed to be?
Isn't it more likely that the rumors are true—that it's simply bullshit and that China violated the sanctions by smuggling around 50,000 Nvidia H100s for DeepSeek’s training via the black market?
Besides, it’s just not true. No other company besides Nvidia has CUDA, and without CUDA, AI only works to a very limited extent.
That’s why 98% of GPUs in all data centers are Nvidia, not AMD or Intel or that mysterious china company.
@denrakeiw All wrong what you said. This will make you retract. Wafer-Scale Engine-3 by Cerebras Systems, the gpu is a single wafer with no bottlenecks like nvidia etc
I viewed a very good youtube video that explained everything but dont have the link here, but you can read a small article to get you drooling.
@xlr8td Sorry, but could you help me out here? I don’t see the connection.
The article is about the Cerebras Systems supercomputer providing the fastest DeepSeek inference. That’s to be expected from a machine that costs anywhere between $2.5 million and $3.5 million per unit.
In comparison, an NVIDIA H100 workstation ($40,000) or even an 8xH100 cluster server ($360,000) is a bargain.
Cerebras Systems is an American company that builds supercomputers—what do they have to do with training DeepSeek?
These American supercomputers were never exported to China to train DeepSeek.
Nvidia GPUs have stripped-down versions for the Chinese market, but these supercomputer machines do not.
@xlr8td BTW, the Cerebras chips are manufactured by TSMC, just like Nvidia chips. 😉
@denrakeiw Sorry we got off track a little. I was trying to make the point that when nvidia lost a big chunk of stock from a panic, almost all people reported due to DeepSeek models and that you dont really need state-of-the-art tech to develop AI, but I only found 1 or 2 people mention that the partnership with Cerebras was also a very big issue for the panic because the partnership offers a very fast service (almost realtime) and you dont need to be usng expensive bottlenecked GPU architecture by nvidia to develop AI. Cerebras wasnt used to train DeepSeek models but is the best way to run the models being released. I am not sure what price is being charged though. While countries are starting to ban DeepSeek models, going off-topic, Chinese EV vehicles are getting DeepSeek installed. Tesla is getting stiff competition and wont be installing Chat-GPT in EVs anytime soon (infact never, unless Elon buys Open AI which is why he wants to IMO, in order to compete with DeepSeek being installed in competitor EVs), while the African inventor of self-powered technology may also disrupt Tesla and its shares could drop also as there is a much much much cheaper alternative. Its only just heating up.
A lot of people also speculated (not pointing any fingers) that DeepSeek used the banned nvidia GPUs through sneaky importing tactics and DeepSeek lied about the hardware used but if you see this video you will learn that the company is capable of doing what they did and that 2 experts (Emad who started up Stable Diffusion and an Asian female engineer who has a youtube channel but cant find the link atm) have checked the numbers in the report and said they add up and dont flat out doubt what DeepSeek did. Also DeepSeek did not startup overnight or have come from nowhere etc the founder has been around for years making breakthroughs with other types of models.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.