AMD’s CEO, Lisa Su, decided to intervene in an issue faced by the AI startup TinyCorp, which is currently integrating AMD’s Radeon RX 7900 XT into a dedicated machine.
TinyCorp’s Journey To Integrate RDNA3 GPUs Into an AI Machine Hits a Barrier, Lisa Su Joins In To Get The Job Done
We recently reported about the venture TinyCorp is currently featuring AMD’s flagship RDNA 3 GPUs, where they decided to integrate six AMD’s Radeon RX 7900 XT, stacked into a 12U rack and interconnected using PCIe 4.0 x16 lanes. This AI-focused machine seemed like something great considering the price it was supposed to come at. Still, TinyCorp reportedly faced a roadblock with the development, primarily involving issues with the firmware of the Radeon RX 7900 XT.
Thanks for the collaboration and feedback. We are all in to get you a good solution. Team is on it.
— Lisa Su (@LisaSu) March 6, 2024
TinyCorp’s official X account frequently posted the troubles they encountered with AMD’s libraries and requested the firm to at least open source relevant GPU firmware, pledging that the firm would fix it by themselves. AMD’s CEO, Lisa Su, answered almost immediately, claiming that her team was working on resolving the issue. The developers’ community welcomed her response, and only professionals saw AMD as a cooperative firm. However, it isn’t sure whether AMD would open-source GPU firmware or fix the issue themselves, but that is yet to be seen.
However, TinyCorp believes that AMD could potentially do the open-sourcing for the GPU’s scheduler and memory hierarchy management code in the firmware. In a follow-up post, the firm said the “call went well,” claiming they are 70% confident that things will go their way. However, the firm did buy up the “Acer Predator” Intel Arc A770 as an alternative, so we could potentially see a shift to Team Blue in the end, but that remains a lesser possibility.
Call went pretty well.
We are gating the commitment to 6x7900XTX on a public release of a roadmap to get the firmware open source. (and obviously the MLPerf training bug being fixed)
We aren’t open source purists, it doesn’t matter to us if the HDCP stuff is open for example.… https://t.co/Zh6df2utIK
— the tiny corp (@__tinygrad__) March 6, 2024
TinyCorp is on the verge of accomplishing something great, so they have received massive industry interest. Utilizing consumer GPUs for AI applications is the way to go for low-to-mid intensity tasks, considering that they have decent compute power packed into themselves and provide a great performance per dollar value. We do hope that TinyCorp gets rid of the existing GPU firmware problems to see the “TinyBox” AI solution in action.