AMD Instinct MI325X in Q4 2024, 288GB of HBM3E

The claim that the next generation would be 35x faster, felt like an "Osborne moment" to me, but if demand is robust enough...

&gt; 35x increase in AI inference performance compared to AMD Instinct MI300 Series<p>Even for marketing claims that’s pretty wild.<p>Still lots of trajectory left in just scale up plan it seems

Such a weird &amp; cruel modernity, where these releases are purely in the abstract. No, you still won&#x27;t be able to buy a MI300X in Q4 2024. The enhanced edition will absolutely not be available.<p>(I miss the old PC era where the world at large was benefiting in tandem from new things happening (or falling behind from not adapting)).

AMD comparison:<p><pre><code> 8x AMD MI300X (192GB, 750W) GPU 8x H100 (80GB, 700W) GPU </code></pre> What would be the result against<p><pre><code> 8x H100 NVL (188GB, &lt;800W) GPU ?</code></pre>

Is the software stack working (for practical use)?<p>AMD still has to prove themselves in this.