Skip to content
View original post on X: ZyphraOfficial
Published
22/100

Zyphra's ZOE makes GLM-5.3-Flash on MI300X profitable at 64 sessions

AIZyphra says its ZOE serving system makes GLM-5.3-Flash on MI300X profitable, earning $2.95 per hour per GPU after $2.02/hr rent at 64 sessions per node. Stock SGLang earns $0.13 per hour under the same conditions.
ZyphraVerified on XOfficial
@ZyphraAI
View on X

Part of a thread · earlier post

ZOE makes serving GLM-5.3-Flash on MI300X profitable. At 64 sessions per node and full utilization, each MI300X makes $2.95 an hour in profit after its $2.02/hr rent. Stock SGLang makes $0.13.

Running 96 sessions earns more, but p90 TTFT rises from 2.9s to 21.8s, so we run 64.

Source details

Zyphra · x.comPublished · added here

Original title: ZOE makes serving GLM-5.3-Flash on MI300X profitable. At 64 sessions per node and full utilization, each MI300X makes $2.95 an hour in pr...

Topics

Deployment & engineeringChina AIProduct updates