Iluvatar CoreX Tianyao 300 Benchmarked Against H200: Training Throughput Reaches 1.33x That of H100
Iluvatar CoreX held its first live demonstration of the Tianyao 300 at a media open day on October 28, claiming that in comparative tests on large language models such as GLM-5.3 as well as text-to-image and text-to-video scenarios, it outperformed the H200 overall. On the training side, in an 8-GPU full-system FP8-precision demonstration based on Qwen3-32B, the Tianyao 300 achieved per-GPU effective throughput 1.33 times that of the H100. The Tianyao 300 was unveiled at this year's World Artificial Intelligence Conference in July, features 144GB of memory, and its IXL144 supernode integrates 144 GPUs with roughly 10.1TB of memory in a single standard rack, reducing the cost per million effective tokens by about 40% compared with mainstream international solutions.
Why read this
AI score 71
The report compiles multiple sets of comparison data between Iluvatar CoreX's Tianyao 300 and the H200 in training, inference, and delivery cost, offering a basis for understanding where domestic GPUs currently stand against their benchmarks.
Topics
Body · AI translation
Domestic GPU takes a big step forward! Iluvatar CoreX's Tiangai 300 beats NVIDIA H200 in real-world tests
Fast Technology Download the app for exclusive early access October 29, 2026, 09:55:55 From Henan Updated October 29, 2026, 10:02:25
Fast Technology news, October 29: According to reports, at the "From Models to the World" 2026 media open day held by Iluvatar CoreX on October 28, its new-generation AI chip, the Tiangai 300, was opened to the media for a live machine demonstration for the first time.
At the event, a live demonstration of the Tiangai 300 was opened, accepting comparative tests against the H200 in scenarios including large language models such as GLM-5.3, text-to-image, and text-to-video; the Tiangai 300's overall performance was superior to the H200 in all of them.
On the training side, in an 8-card full-machine FP8-precision training demonstration based on the Qwen3-32B model, the Tiangai 300's effective throughput per card reached 1.33 times that of the H100.
The Tiangai 300 was officially released at this year's World Artificial Intelligence Conference in July. It adopts the Iluvatar Tianshu architecture, integrates a new-generation Tensor Core, is equipped with 144GB of memory, natively supports three types of computing modes—scalar, vector, and tensor—covers multiple data precisions including FP4, FP8, FP16, and BF16, and supports high-speed inter-chip interconnect.
Third-party testing completed by Shanghai Instrumentation & Electronics in July this year showed that under the typical PD-disaggregated deployment for large models, the Tiangai 300's converted single-card overall end-to-end throughput reached 1.05 times that of the H200—the first general-purpose GPU in China whose measured performance reaches the H200 level.
Testing by the Iluvatar CoreX Performance Lab further covered typical business scenarios:
Under a 64K long context, the Tiangai 300's Attention performance is 1.05 times that of the H200;
At the 64K scale of the GLM-5.3 model, while ensuring the same user experience, MoE throughput is 1.15 times that of the H200;
In 8K to 128K long-context scenarios, Agent first-token latency is reduced by about 10% to 30% compared with the H200;
Multimodal generation inference throughput is improved by about 2% to 5% compared with the H200;
In online businesses such as search, advertising, and recommendation, recommendation inference throughput is increased by 21%, and lightweight computing latency is reduced by 26%.
On the large-scale delivery side, the Tiangai 300 has formed a complete system solution ranging from accelerator cards and 8-card-per-machine servers to super nodes. The Tiangai 300 IXL144 super node integrates 144 GPUs and about 10.1TB of memory within a single standard rack, and the cost per million effective tokens can be reduced by about 40% compared with mainstream international solutions.
"Special statement: The above content (including videos, pictures, or audio if any) is uploaded and published by users of "Dafeng Hao," a self-media platform under Phoenix Net. This platform only provides information storage space services.
Notice: The content above (including the videos, pictures and audios if any) is uploaded and posted by the user of Dafeng Hao, which is a social media platform and merely provides information storage space services."