Read more.Quote:
Designed to fit the needs of the enterprise market.
Printable View
Read more.Quote:
Designed to fit the needs of the enterprise market.
So what would an example Overdrive 3000 system typically cost, in terms of general ballpark figures.
What are the USPs over competitor products?
Odd start to that write-up, saying that RISC comes from mobile. RISC was forged in high end servers and workstations when the performance was leagues beyond what a PC could manage. Even today, if I wanted the best server money can buy I would have to go for an IBM Power based machine, and before cheap AMD64 machines became available if you wanted a compute server you were probably looking at MIPS, PA-RISC or SPARC.
What is new here is that traditionally Intel cut into the server market with cheap "good enough" servers, and now ARM has the ability to do the same again, cutting market share out from the cheap end of the market.
Shame the timing didn't give us A72 cores in this, but 2GHz A57 should be interesting enough.
S/A thinks the top end APU is $150. Won't be turning up in a raspberry Pi any time soon then.
http://semiaccurate.com/2016/01/14/a...s-for-servers/
This is interesting from the PoV that ARM Opteron development was previously believed to have been abandoned.
AMD's x86s originally came from a x86 interpreter wrapped around their RISC CPU (486, k5 and k6 generations).
From the PoV of someone who spends $2-500k/year on server hardware this is interesting. We're not hamstrung by binary compatibility issues and mostly need better bang per buck/watt than x86 is offering, however there's a lot of single-threaded coding still going on where individual CPU speed matters (gotta love university researchers - most of them can't code their way out of a wet paper bag, let alone generate threaded code.)
Moving away from being intel/microsoft's bitch.Quote:
Originally Posted by sykobee
Most large have been doing this for years.
NEED.....MORE......CORES......
Half serious. I have compute servers here with 48+ x86 cores onboard that are being thrashed by researchers. If I'm going to rollout ARM boxes (RHEL environment), then I need to match the performance of the existing stuff or noone will use them.