Author: Li Dan
Source: Hard AI
NVIDIA, known for pioneering artificial intelligence (AI), officially delivered the AI supercomputer DGX H200 to OpenAI, giving OpenAI a head start.
On Wednesday, April 24th, Eastern Time, Greg Brockman, President and Co-Founder of OpenAI, revealed on social media X that NVIDIA handed over the world's first DGX H200 to OpenAI. He also shared a photo taken at the delivery site with OpenAI CEO Altman and NVIDIA CEO Huang Renxun. Brockman stated that this device, created with the full support of Huang Renxun, &#34;will advance AI, computing, and human civilization.&#34;
NVIDIA previously introduced that the DGX H200 is a new AI supercomputer built by NVIDIA by interconnecting multiple Grace Hopper super chips into the same GPU. It can handle large recommendation systems, generative AI, and TB-level models in the field of graphic analysis, providing a large shared memory space with linear scalability for large AI models. The Grace Hopper super chip combines NVIDIA's Grace CPU with Hopper GPU on the same package, eliminating the need for traditional PCIe CPU to GPU connections, increasing bandwidth by seven times, and reducing interconnect power consumption by more than five times. With the one-stop deployment of DGX GH200, users can build large models in a matter of weeks, rather than months.
Netizens on the X platform summarized some key parameters and highlights of the DGX H200:
32 Grace Hopper super chips interconnected via NVIDIA's NVLink, massive shared GPU memory space of 19.5TB, GPU to GPU bandwidth of 900 GB per second, and AI FP8 performance of 128 petaFLOPS.
Some netizens quoted Wikipedia to show how expensive a DGX H200 is, with a launch price of £379,000 or $482,000.
One netizen cleverly asked Brockman if there were any former board members and Chief Scientist of OpenAI, Ilya Sutskever, inside the box delivered by NVIDIA
<img src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/e56198f4-31d1-4097-a4d5-751183f89c72.jpeg?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/e56198f4-31d1-4097-a4d5-751183f89c72.jpeg"/>
Some netizens directly modified the group photo provided by Brockman by adding an arrow pointing to the NVIDIA logo on a box, implying that Ilya is inside the box. The caption of the image reads, &#34;Finally found him (Ilya).&#34;
<img src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/77374ce7-3609-45f5-9072-ee12498c8b1e.jpeg?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/77374ce7-3609-45f5-9072-ee12498c8b1e.jpeg"/>

SOXL

NVDL

NVDA

DGX H200 is a new AI supercomputer created by NVIDIA, which interconnects multiple Grace Hopper super chips to the same GPU. It has a massive shared GPU memory space of 19.5TB, GPU-to-GPU bandwidth of 900GB, and AI performance of 128 petaFLOPS

OpenAI "Tries Out": Receives the world's first DGX H200 AI supercomputer from NVIDIA