And Samsung just launched 36 GB packages of HBM3E: <a href="https://www.anandtech.com/show/21278/samsung-launches-12hi-36gb-hbm3e-memory-stacks-with-10-gts-speed" rel="nofollow">https://www.anandtech.com/show/21278/samsung-launches-12hi-3...</a><p>Combined with this[1] interesting paper from summer 2023 on HBM combined with Xeon processors which would now allow for 144 GB on a single CPU. In theory at least.<p>1: <a href="https://lenovopress.lenovo.com/lp1738-implementing-intel-high-bandwidth-memory" rel="nofollow">https://lenovopress.lenovo.com/lp1738-implementing-intel-hig...</a>
How about higher capacity GDDR6X?<p>48GB consumer cards (or 96GB pro cards) would sell like hotcakes if AMD/Intel dare to break the artificial VRAM segmentation status quo.
The linked press release is a better source, anandtech didn't add much.<p><a href="https://investors.micron.com/news-releases/news-release-details/micron-commences-volume-production-industry-leading-hbm3e" rel="nofollow">https://investors.micron.com/news-releases/news-release-deta...</a>
This is an empty article:<p>- CAS latency?<p>- Wattage?<p>Also: <a href="https://piped.video/watch?v=2G4_RZo41Zw" rel="nofollow">https://piped.video/watch?v=2G4_RZo41Zw</a> (is this memory same size as 5nm?)
Would this explain why I started seeing so many ads from HBM manufacturers with no product an individual like me can purchase? (I'm not even close to hardware, professionally)<p>Not naming the company but seems like HBM manufacturers might be going all-in to benefit from Nvidia's stock surge.
By the article (March production) new GPU eng. samples should be flying in H1 which means new high memory GPUs coming end of the year. 2025 will be extremely exciting for ML.