3 ms·
Honestly, I wouldn't put it behind IBM to turn it around with POWER revival. They'd been doing some cool stuff recently with their NorthPole accelerator[1], and
by tucnak 2y ago
Honestly, I wouldn't put it behind IBM to turn it around with POWER revival. They'd been doing some cool stuff recently with their NorthPole accelerator[1], and using 12nm process while at it, indicating there's much room for improvement. It could eventually become a relatively open, if not super affordable platform. There's precedent with OpenPOWER! And not to mention RISC-V, of course, championed by Jim Keller et al (Tenstorrent) but it's yet to blossom, all the while pppc64el is already there where it matters.
I say, diversity rules!
[1]: https://research.ibm.com/blog/northpole-llm-inference-results https://research.ibm.com/blog/northpole-llm-inference-result...
- __d 2y agoPPC’s likely last hope died when Google didn’t go ahead with OpenPower. Talos is the exception that proves the rule, sadly.
- classichasclass 2y agoIBM did lay an egg with Power10, though. They cut corners and used proprietary IP and as a result there are few (are there any?) non-IBM Power10 systems because the other vendors stayed away. Raptor workstations and servers are a small-ish part of the market but they're comparatively highly visible - and they're still on POWER9 (no S1 yet). They did realize the tactical error, so I'm hoping Power11 will reverse the damage.