Artificial Intelligence thread
- Thread starter 9dashline
- Start date
This legit? whats the catch, mich slpwer tps?
Tsinghua University's open source project breaks through the bottleneck of large model computing power: RTX 4090 single card achieves full-blooded operation of DeepSeek-R1
It is very difficult for ordinary users to run the real full-blooded version of DeepSeek-R1 on ordinary hardware, and the cost of renting a server is stressful even for developers
It is very difficult for ordinary users to run the real full-blooded version of DeepSeek-R1 on ordinary hardware, and the cost of renting a server is stressful even for developers
Some have shown that you can run the full version on a $2000 rig with a lot of system RAM, it's just pretty slow. Looks like this rig also has a lot of system RAM, so the question is how much does a single 4090 speed up the process?
Sugar has added support for DeepSeek model in its machines. It also supports V3, Janus Pro, Qwen 2.5, Llama and ChatGLM. Uses domestic x86 CPU and GPGPU
WeChat now testing with DeepSeek-R1. Again, this is going to be huge for AI search adoption with R1
ByteDance CEO said they did not move fast enough with reasoning model and Deepeek moved faster and got there first.
I wish they would prioritize some domestic hardware for this kind of efforts rather than improving efficiency for more Nvidia products. Since no one other than Chinese researchers are looking to squeeze this much of performance from lower-end hardware, if they just do this with domestic products they could have more leveled the playing field between domestic cards and much higher-end Nvdia cards. Kind of like how Huawei has been matching performance of western 4nm chips by optimizing their designs on 7nm node beyond what western chipmakers, with access to leading edge fabs, are looking into.
Last edited:
Seeing so many companies in China attaching and customizing DeepSeek to fit them.
Maybe that's the real thing that's important. Not how many benchmarks it can beat but the open source nature that allows the entire ecosystem to adjust based on the new technology. Something like wouldn't happen with stuff like OpenAI because they aren't gonna give away their stuff for free especially when these companies have so much money.
Maybe that's the real thing that's important. Not how many benchmarks it can beat but the open source nature that allows the entire ecosystem to adjust based on the new technology. Something like wouldn't happen with stuff like OpenAI because they aren't gonna give away their stuff for free especially when these companies have so much money.