What /g/ thinks of The Machine?
https://www.youtube.com/watch?v=2VG59FYkPdM
>>57822461
I think intel's inability to deliver the photonic interconnect chips is probably intentional sabotage.
They didn't want a new memory standard in enterprise unless they were getting profits from it.
Funny how they bought into Micron's phase change memory and branded it as their own Xpoint.
>>57822461
How am I supposed to have an opinion on it if they didn't actually tell you how it worked?
>>57822461
It's just a server?
>>57822672
Its the new version of their prior generation Moonshot server platform.
They originally planned to use their own "Memristor" memory as both permanent storage and system RAM. Less time wasted swapping data around means more clock cycles are spent delivering performance. HP partnered with intel to develop a super high bandwidth and low latency photonic interconnect, but it kept getting delayed. Its been delayed for years. A couple years ago there was a news story about intel not being able to deliver enough of the driving chips for the interconnect, so that left HP without a way of connecting their memory to the systems.
They put the Memristors on hold, and decided to just have ECC DRAM take its place. They programmed their own OS specifically for it and everything.
I forget if they were using ARM or Atom cores in most of the old Moonshot SKUs, but I think they're just standard Xeon E5 or E7s now. Its been a while since I last looked into it.
>>57822461
>an optical interconnect
>revolutionary
>blade computers talking to each other
>clapping from women project managers
its nothing. i dont even know what the fuck this is but i can only assume its a competitor to something like a SGI UV 3000 in which case again its nothing new
>>57822461
>>57822672
https://www.youtube.com/watch?v=S--Kgseuy0Q
>>57822798
Its memory-driven computing. You do not need IO-operations. So it makes the perfect big-data NSA computer.
>>57822849
>Its memory-driven computing.
NVDIMMs are nothing new
>>57822799
Looks interesting.
Might be worth keeping tabs on to see where it will go.
>>57822793
The original project was way too ambitious. Not even IBM, Intel, Nvidia, Samsung or other giant hardware corporation have tried to make something similar. They don't have the resources to do that.
https://www.youtube.com/watch?v=S--Kgseuy0Q
>Presenter knows it's shit
>Has to try and avoid using corporate speak
I feel sorry for him.
>>57822461
>evil white people do it again
>>57822896
This isn't NVDIMM. You should look into things before making stupid posts.
>>57822798
>>an optical interconnect
>>revolutionary
An operatically interconnect that serves as the main bus between memory and also other cpus tho...
>>57822896
They plan on eventually using memeristors :^).....
>>57826727
>implying memristors are bad.
>>57822461
HP becoming HP again....
wow we must of jumped timelines.
>>57822798
God that obnoxious woman in the green shirt. You can tell as soon as she opens her mouth that all she understands is marketing speak and has no idea what the fuck they are actually doing.
>>57822461
As i understand it, they use oxygen (the first most abundant element in the earth) rather than silicon, thus building memristors rather than transistors. Storing data in much smaller spaces and saving a lot more power than the current servers.
https://www.youtube.com/watch?v=jcmsby8jDKE
>>57829804
It's because HP split into two companies a bit ago. One of them is still pumping out pure garbage.
>>57826108
Normies don't know technical jargon, anon.
>>57822461
It's dead because they can't into memristors.
>>57822798
>clapping from women project managers
What?
>>57822461
Why don't they just take a UniDIMM PCB and take a 4 core ARM CPU equivalent in processing power to maybe 2 Xeon cores and 4GB of LPDDR and 128GB of UFS 2.1 and then roughly place 32 of them vertically on a carrier board connected via multiple gigabit ports and an onchip switch and then take 3 of those carrier boards and place them into a 1U chassis and also use the top side of the chassis to have another three carrier boards for 192 nodes in total and 24TB of flash and 7 million IOPS and a power consumption of 2Kwh. Why don't they do that instead?
>inb4 mysql can't scale that well
It can.
>>57822461
I mean, it looks really cool, but I'd lie if I'd say that I have any idea how it works or what it does or what it's used for
ITT: people understanding shit about r&d.
We get these in all the time; they're impressive in the sense they can hold a ton of RAM, but when you look at how much it costs to get this and the entire cabinet running, you start to wonder just how vast the enterprise market is compared to the consumer level.