Tianshu Zhixin's Tianjie 300 matches H200 with 1.05x throughput in real-device tests
10/28/2026 — 10/30, 19:20·2 sources (all)·2 reports (all)
Key takeaways
- At a 2026 media day, Tianshu Zhixin showed its Tianjie 300 general-purpose GPU running against Nvidia's H200. The company says it is China's…
Background & analysis
On October 28, 2026, Iluvatar CoreX held its 2026 media open day in Shanghai and publicly demonstrated the Tiangan 300 general-purpose GPU running on real hardware for the first time. The company described Tiangan 300 as China's first general-purpose GPU whose measured performance reaches the level of NVIDIA's H200, citing third-party testing that puts single-card aggregate throughput at 1.05x that of H200. IT Home noted, however, that the specific test methodology, model loads and workload conditions had not been disclosed. Until this event, most public information on the product had been limited to architecture and roadmap details rather than live operation.
The open day also showcased the companion IXL144 super-node solution, which had appeared earlier at the China International Industrial Fair. The design packs 144 GPUs and 18 CPUs into a single cabinet with roughly 10.1TB of aggregate memory capacity, which Iluvatar CoreX called the highest-density single-cabinet GPU super-node in the industry. Based on the company's own calculations, the solution can reduce cost per 1M effective tokens by 40% compared with mainstream international alternatives. A separate advance disclosed at the event was that China's first AFD-split deployment combining GPUs with XPUs completed a successful run, cutting inference latency by 60%. In this context, AFD splitting generally refers to partitioning computation stages such as attention and feed-forward networks across heterogeneous compute units.
On October 30, Leiphone reported more detail on the testing. Iluvatar CoreX stated that in a third-party test completed this year, under a typical PD-split deployment, Tiangan 300's single-card end-to-end throughput reached 1.05x that of H200. Leiphone observed that competition among domestic GPUs is shifting toward system efficiency, software adaptation and actual delivery, and noted that the software stack now incorporates Agent and Skills to address the engineering challenges customers face in deployment, adaptation and optimization. The report also mentioned that Iluvatar CoreX is broadening its lineup from the cloud-oriented Tiangan 300 to the edge-focused Tongyang series. As of this point, the public disclosures stand at measured performance figures and companion system designs, with full test conditions still unreleased.
AI-generated from 2 reports · updated 6 days ago
Latest turnAt this week's Iluvatar CoreX media day, the company showed its Tiangai 300 GPU in a live comparison with the H200. Iluvatar says third-party tests this year put single-card end-to-end throughput at 1.05x that of the H200 in a typical PD-disaggregated deployment. The report says competition among domestic GPUs is shifting toward system efficiency, software adaptation, and delivery.
- Heat index
- 25
- All sources
- 2
- All reports
- 2
- First seen
- 8 days ago
Reports on this story headlines open the original
At this week's Iluvatar CoreX media day, the company showed its Tiangai 300 GPU in a live comparison with the H200. Iluvatar says third-party tests this year put single-card end-to-end throughput at 1.05x that of the H200 in a typical PD-disaggregated deployment. The report says competition among domestic GPUs is shifting toward system efficiency, software adaptation, and delivery.
雷峰网AI score 58
Tianshu Zhixin publicly demoed its Tiangai 300 general-purpose GPU for the first time at its 2026 media open day, claiming it is the first domestic GPU to match Nvidia's H200 in measured performance, with third-party tests putting per-card throughput at 1.05x H200 — though test methods and workloads were not disclosed. The company also said its IXL144 super-node cuts cost per million effective tokens by 40% versus mainstream international designs.
IT之家AI score 54
Other stories people are talking about
- Heat index 840OpenAI ships Images 2.6 with GIF and SVG output15 sources
- Heat index 675OpenAI says its cyber model pipeline confirmed 1,724 Linux kernel flaws and generated 948 privilege-escalation exploits11 sources
- Heat index 642RisingOpenAI GPT-5.6 Brings Major Agentic and Computer Use Upgrades13 sources
- Heat index 633RisingCat Wu, product lead for Claude Code, joins OpenAI9 sources
- Heat index 630NewSurgeOpenAI releases GPT-6 Cyber cybersecurity model7 sources
How is heat calculated?About the methodHide
Heat counts how many independent sources covered a story in the last 48 hours: one source counts once no matter how many posts it published, decaying with a 24-hour half-life. What ranks first is what many people are talking about.
This page aggregates public feeds. Headlines and summaries are machine-organized and remain the property of the original authors; verify important facts at the source. If you believe a headline or summary infringes your rights, email the contact address in the footer with the page URL and basis of your claim, and we will remove or replace it after review.
- Surge
- Discussion rising fast
- New
- First report within 6 hours
- Rising
- Still gathering discussion

