@@WorBlux I guess it depends how mortal we're talking. The Blackbird with a 4-core or 8-core Power9 is ~2,000 without RAM or other components. Not a consumer commodity, but cheaper than the cheapest Mac Pro.
@@ericwood3709 Both the Lite, and blackbird options are more affordable. However the Lite doesn't fit into a standard case, and will having the same CPU:RAM as the full board,, it's I/O is sadly quite limited (24 lanes in 2 slots, + an option for an expensive on-board SATA controller) and the blackbird only has 1/8 the RAM bandwidth and about 1/6 of pci-e lanes for only half the cost. It's simply not a very cost-effective competitor to x86, especially thread-ripper or even Ryzen9. The full Talos II board is a lot more competitive, but it's also a lot more compute than I need. It's a full on server-class system. For the next round, I'd rather see a budget version to be ATX form factor, 4 channels of memory, 1 x16, 1x8, 1x4, and 2 x4 m.2 (either on board, or adapted from and x8 slot) of pci-e as a square competitor against x299/threadripper. I could afford to spend an extra 50% on the base of a workstation I was going to buy anyways, but paying double for what is now desktop-class performance really hurts. It's also doubtful we''l see OMI and off-board SMP interconnects on Talos's offerings. People that actually need that are buying full Racks from IBM or renting time on a cloud machine once they've developed the program on a local machine.
@@WorBlux I agree completely. I'm an enthusiast myself, and price has kept me from getting one to play around with even at the low end. If they can deliver full ATX or micro-ATX at a lower price, and if the damned CPU prices would also come down a bit (or if IBM would offer a more consumer-oriented version of it in the next iteration), I just might bite. Of course, I would also be happy to play with some kind of ATX or micro-ATX ARM system; I'm responding to you on a Raspberry Pi 4 running Ubuntu Mate right now, and it's actually kind of a nice little desktop replacement. It would be so much better, though, in a larger form factor with more power and I/O to match.
At least it made a little hype for IBM cpu´s. We must wait to the "little" 4 threath IBM power 10 cpu. It would be nice to have some products for enthusiast. Now you can buy an used power 9 cpu&memory relatively cheap but the Blackbird motherboard is really expensive and quite limited. I would like to try it.
Memory clustering (and Single System Image techniques in general) feel like removing some of the control from the programmer around performance of the network. We live with this for the cache/memory divide for historical reasons, but I'm not sure it's a pattern we want to continue, even if twenty five years ago we were all dreaming of getting our hands on a T3E or an Origin 2000.
It's a chip/processor with up to 15 cores running 8 threads each core (120 simultaneous threads per chip). It can scale in a single system (IBM POWER E1080) to 16 chips/sockets, that results in on large computer of 240 cores / 1920 simultaneous threads.
The failure to understand the PPC giga lite processor was an unbelievable failure. Matrix computing was within your grasp just as it was with the 88K for Motorola. Don't be stupid and concentrate on just a small sector of enterprise. The complementary ability of the smaller computing systems to add to the total SYSTEM is beyond what you are stating. AI, and I was a decision scientist with hueristics, modeling of every profession and quantitative mathematics, could allow you to connect possibly the power of all computers as Apple has done with Rosetta with X86 to what will be a competitor of yours. Latency is always the part of the slowest bus in the chain. GPU? Look at ARM!!! Atto technologies is working with 120GB/Sec for external and internal systems of bucket brigade storage will be capable of feeding much higher levels of processors. You think that the cloud is going to be of IoT / IoS /... is going to be capable of the ultimate speed but this has no reality unless you can work seamlessly. RISC has always had the capability of multitple level matrix data to be run with one instruction. Massively parallel execution will need to synchronize the work of all systems that will have the relational connection to a multilevel sync over all of your systems regardless of OS. This means handshakes will be not only required by TCP/IP but a data equivalent.
I look forward to seeing what Raptor comes out with based around this chip.
Indeed, not sure how they'll make this chip accessible to the realm of us mere mortals. Might have to be satisfied w/microwatt.
@@WorBlux I guess it depends how mortal we're talking. The Blackbird with a 4-core or 8-core Power9 is ~2,000 without RAM or other components. Not a consumer commodity, but cheaper than the cheapest Mac Pro.
@@ericwood3709 Both the Lite, and blackbird options are more affordable. However the Lite doesn't fit into a standard case, and will having the same CPU:RAM as the full board,, it's I/O is sadly quite limited (24 lanes in 2 slots, + an option for an expensive on-board SATA controller)
and the blackbird only has 1/8 the RAM bandwidth and about 1/6 of pci-e lanes for only half the cost. It's simply not a very cost-effective competitor to x86, especially thread-ripper or even Ryzen9.
The full Talos II board is a lot more competitive, but it's also a lot more compute than I need. It's a full on server-class system.
For the next round, I'd rather see a budget version to be ATX form factor, 4 channels of memory, 1 x16, 1x8, 1x4, and 2 x4 m.2 (either on board, or adapted from and x8 slot) of pci-e as a square competitor against x299/threadripper. I could afford to spend an extra 50% on the base of a workstation I was going to buy anyways, but paying double for what is now desktop-class performance really hurts.
It's also doubtful we''l see OMI and off-board SMP interconnects on Talos's offerings. People that actually need that are buying full Racks from IBM or renting time on a cloud machine once they've developed the program on a local machine.
@@WorBlux I agree completely. I'm an enthusiast myself, and price has kept me from getting one to play around with even at the low end. If they can deliver full ATX or micro-ATX at a lower price, and if the damned CPU prices would also come down a bit (or if IBM would offer a more consumer-oriented version of it in the next iteration), I just might bite. Of course, I would also be happy to play with some kind of ATX or micro-ATX ARM system; I'm responding to you on a Raspberry Pi 4 running Ubuntu Mate right now, and it's actually kind of a nice little desktop replacement. It would be so much better, though, in a larger form factor with more power and I/O to match.
At least it made a little hype for IBM cpu´s. We must wait to the "little" 4 threath IBM power 10 cpu. It would be nice to have some products for enthusiast. Now you can buy an used power 9 cpu&memory relatively cheap but the Blackbird motherboard is really expensive and quite limited. I would like to try it.
Memory clustering (and Single System Image techniques in general) feel like removing some of the control from the programmer around performance of the network. We live with this for the cache/memory divide for historical reasons, but I'm not sure it's a pattern we want to continue, even if twenty five years ago we were all dreaming of getting our hands on a T3E or an Origin 2000.
Seeing the development of Power it's sad to no longer have it coming down to end consumers anymore (with the death of PPC).
Thanks
Your git website is STILL not functioning properly.
is it single processor or multiprocessor, I mean is it single cpu?
It's a chip/processor with up to 15 cores running 8 threads each core (120 simultaneous threads per chip). It can scale in a single system (IBM POWER E1080) to 16 chips/sockets, that results in on large computer of 240 cores / 1920 simultaneous threads.
I would like to be owner of power 10.
I would be impressed.
The failure to understand the PPC giga lite processor was an unbelievable failure. Matrix computing was within your grasp just as it was with the 88K for Motorola. Don't be stupid and concentrate on just a small sector of enterprise. The complementary ability of the smaller computing systems to add to the total SYSTEM is beyond what you are stating. AI, and I was a decision scientist with hueristics, modeling of every profession and quantitative mathematics, could allow you to connect possibly the power of all computers as Apple has done with Rosetta with X86 to what will be a competitor of yours.
Latency is always the part of the slowest bus in the chain. GPU? Look at ARM!!! Atto technologies is working with 120GB/Sec for external and internal systems of bucket brigade storage will be capable of feeding much higher levels of processors.
You think that the cloud is going to be of IoT / IoS /... is going to be capable of the ultimate speed but this has no reality unless you can work seamlessly. RISC has always had the capability of multitple level matrix data to be run with one instruction. Massively parallel execution will need to synchronize the work of all systems that will have the relational connection to a multilevel sync over all of your systems regardless of OS.
This means handshakes will be not only required by TCP/IP but a data equivalent.
👁 🐝 MAIDEN VOYAGE
Don't be condescending. I find consumer products useful. And I'll bet your phone isn't Power 10, is it?