Originally posted by phoron
View Post
Announcement
Collapse
No announcement yet.
Libre/Open-Source POWER10 Hardware Systems Unlikely Until At Least 2022
Collapse
X
-
-
I'm watching the Power10 HotChips presentation right now. It's an interesting architecture that clearly is trying to do some of the same things as Sapphire Rapids since it introduces a matrix unit similar to AMX.
Interesting tidbit from the talk: IBM confirmed that Power 10 is being fabbed by Samsung, not TSMC.Last edited by chuckula; 17 August 2020, 01:36 PM.
Leave a comment:
-
It's a shame though that due to whatever is going on at IBM that there will seemingly not be any It's a shame though that due to whatever is going on at IBM that there will seemingly not be any fully-open platforms at launch for these next-gen 7nm CPUs. platforms at launch for these next-gen 7nm CPUs.
Leave a comment:
-
Bitter sweet news to me... I have to wait longer before I can get my hands on a P10 system... But it means it will be longer before I drop that kind of cash.
I loved my Talos II so much I ended up also buying a Blackbird. Definitely forwarding to P10.
- Likes 3
Leave a comment:
-
Originally posted by pgeorgi View PostEven if Raptor gets the engineering Even if Raptor gets the engineering samples to work with, they might not be the first customer to get production chips for logistical reasons:
"Worst case" the POWER10 design requires signed blobs and Raptor can never work with it or some such. But we do not have such evidence either.
samples to work with, they might not be the first customer to get production chips for logistical reasons:
With a new CPU design there's always the risk that something goes wrong: Yield might be lower than expected (which leads to delays). There may be last minute issues, or problems that only show in deployment and require to replace the chips with a fixed revision.
Which customer is more likely able to deal with that: A comparably cash strapped computer boutique with customers (nearly) everywhere or some public research network with practically infinitely deep pockets that deploys all those chips in a single data center?
In the meanwhile Raptor can hopefully keep growing with POWER9 so that they are not such a comparatively small IBM customer in the future...
some public research network with practically infinitely deep pockets
Leave a comment:
-
Originally posted by phoron View PostIf I read it right IBM says it'll have POWER 10 CPUs by second semester 2021, so in December 2021.
Raptor says they'll have systems in 2022 or later, best case January 2022. That would not be such a huge difference.
Consider two kinds of customers (not saying that IBM has those, but hypothetically, also taking the extreme positions): One, some HPC project that will put several thousands of those CPUs in a single data center. The other, Raptor, putting hundreds, maybe thousands of those CPUs in workstations that are then shipped all around the world.
With a new CPU design there's always the risk that something goes wrong: Yield might be lower than expected (which leads to delays). There may be last minute issues, or problems that only show in deployment and require to replace the chips with a fixed revision.
Which customer is more likely able to deal with that: A comparably cash strapped computer boutique with customers (nearly) everywhere or some public research network with practically infinitely deep pockets that deploys all those chips in a single data center?
- Likes 1
Leave a comment:
-
There is an open source advocate now in charge of the OpenPower foundation and IBM still comes up short of expectations. Great to know that Intel isn't the only big blue company to mess things up.
- Likes 2
Leave a comment:
-
Let's hope it's more a IBM-Raptor relationship problem (like disagreement on conditions for early access to prototypes or something like this) than an inherent blob-encumbered design or something more serious.
If I read it right IBM says it'll have POWER 10 CPUs by second semester 2021, so in December 2021.
Raptor says they'll have systems in 2022 or later, best case January 2022. That would not be such a huge difference.
In practice it's possible it'll be much later, but anyway, POWER9 is good enough for some time. It just means the efforts of people porting things to POWER9 will be useful longer. And let's hope in the meantime there's more support and community for POWER9 and adapting to POWER10 doesn't prove difficult.
I'm not sure what Raptor meant with new hardware for POWER9 (LaGranje? Nimbus?, more Sforza boards with different peripherals? a POWER9 laptop if there's some way to handle the battery/heat?). I just hope they keep selling Talos II and Blackbirds and expand their retail channels.
Good luck Raptor, and keep the compromise with software freedom.Last edited by phoron; 17 August 2020, 12:17 PM.
- Likes 4
Leave a comment:
-
Libre/Open-Source POWER10 Hardware Systems Unlikely Until At Least 2022
Phoronix: Libre/Open-Source POWER10 Hardware Systems Unlikely Until At Least 2022
While Raptor Computing Systems has been making fabulous 100% open-source/libre hardware systems based around POWER9 with the likes of their Talos II and Blackbird systems, don't hold your breath on quickly seeing fully-open POWER10 systems even with "OpenPOWER" being trumpeted in recent years and similar for being more open-source friendly than the likes of Intel and AMD...
Tags: None
Leave a comment: