IBM Power10 CPU will have 2.6x performance per watt and the 7nm CPU will be fabbed by Samsung. It should become available in late 2021. This is another report from Sander Olson.
It will have up to 60/30 cores/socket (240 HW threads) Modular Building Block Die New Core uArch AI-optimized ISA Energy Efficiency Focus HW Enforced Security Enterprise Focus PowerAXON 2.0 PCIe G5 Memory Clustering.
POWER10 Design Focus
Data Plane Bandwidth, Capacity, Composability, Scale
Terabyte/second sockets, Petabyte system memory capacities, 16-socket SMP → Clusters
Powerful Enterprise Core
New Core Architecture, Flexibility, Larger caches, Reduced Latencies
End-to-end Security
Hardware enabled and co-optimized with PowerVM hypervisor
Energy Efficiency
3x improvement over POWER9
AI-Infused Core
10-20x matrix-math performance / socket compared to POWER9
SOURCES – Hot Chips 32, IBM, Sanders Olson
Written by Brian Wang, Nextbigfuture.com
Brian Wang is a Futurist Thought Leader and a popular Science blogger with 1 million readers per month. His blog Nextbigfuture.com is ranked #1 Science News Blog. It covers many disruptive technology and trends including Space, Robotics, Artificial Intelligence, Medicine, Anti-aging Biotechnology, and Nanotechnology.
Known for identifying cutting edge technologies, he is currently a Co-Founder of a startup and fundraiser for high potential early-stage companies. He is the Head of Research for Allocations for deep technology investments and an Angel Investor at Space Angels.
A frequent speaker at corporations, he has been a TEDx speaker, a Singularity University speaker and guest at numerous interviews for radio and podcasts. He is open to public speaking and advising engagements.
Outsourcing the Fab was a good idea. But it still cause a lot of money to design and test each new generation. Eventually, it will die.
IBM isnt shouldering the fab, its outsourcing, so same boat as other fabless companies. So what you say is purely a software support question. There are enough institutional IBM users and open source is massively bigger than yesterday, can be recomplied easily enough, and go-betweens live java, javascript make for much portability. It may be pure technical mindset but I enjoy the totally fresh approaches of Power10, and I think it will make a mark. Besides, there is also a trove of patents and while a single arch probably won’t give enough bang, after a few iterations this arch can be effectively revived.
After PowerPC swapped endian-ness to match x86, the big cloud boys started sniffing around, testing them with recompiled linux software. The improvements over Intel at least were attractive enough that Google at one point was seriously thinking of building one datacenter-worth of them. Though how much things have changed after AMD (and now to a certain extent Intel) have embraced chiplets and heterogenous chips threw a spanner in that. At least the PowerPC folks have been rather embracing of new interconnects, which again the big cloud guys find interesting for internal needs.
I am surprised the POWER CPU isn’t dead yet. I predicted two decades ago that most CPUs would died because of the increasing cost of FAB as you shrink the geometry. You have to be able to sell a certain minimum number of CPU to justify the cost of designing, testing and manufacturing the CPU. So year after year CPUs joined the dustbin of time. But I guess IBM can afford to waste money keeping POWER alive when they should let it die. IBM need to let marginal products like POWER die when they can see there is no future for them.