whaaaaaaaaaaaaaaat \AWS now offers a virtual machine with over 4TB of memory

[piemadd]

Source: https://techcrunch.com/2017/09/14/aws-now-offers-a-virtual-machine-with-over-4tb-of-memory/?ncid=rss

 

Quote

Earlier this year, Amazon’s AWS group said that it was working on bringing instance types with between 4 to 16TB of memory to its users. It’s now starting to fulfill this promise as the company today launched its largest EC2 machine (in terms of memory size) yet: the x1e.32xlarge instance with a whopping 4.19TB of RAM. Previously, EC2’s largest instance only featured just over 2TB of memory.

"Only 2TB of memeroy." Oof. So, some of us remeber from CES in 2016 (i believe) when LTT made a video on a 4tb of theoretical memory being a thing. Well, now you can do that.

Quote

These machines feature quad-socket Intel Xeon processors running at 2.3 GHz, up to 25 Gbps of network bandwidth and two 1,920GB SSDs. There are obviously only a few applications that need this kind of memory. It’s no surprise, then, that these instances are certified to run SAP’s HANA in-memory database and its various tools and that SAP will offer direct support for running these applications on these instances.

Lot o' Xeon. 4tb of ssds as some sort of cache for any extended storage (HDDs).

Quote

These new instances are now available in four AWS regions: US East (Northern Virginia), US West (Oregon), EU (Ireland) and Asia Pacific (Tokyo). Running these instances doesn’t come cheap, of course. On-demand pricing in the US East region is $26.688 per hour and in Tokyo it’s $38.688, for example. That makes them AWS’s most expensive machines yet.

 

It’s worth noting that Microsoft Azure’s largest memory-optimized machine currently tops out at just over 2TB and that Google already calls it quits at 416GB of RAM.

Lot o' money too. From ~27 USD/hour to ~37 USD/hour depending on the region. Microsoft only has 2t ram machines, and google has 416gb ram machines.

[Lurick]

Psh, only 25Gbps host network speeds, that is just so 2016.

50Gbps or bust Amazon, gosh!

[themctipers]

24 minutes ago, CUDA_Cores said:

If you were the very small group of companies who need that much RAM, it would probably be cheaper to buy the hardware outright and do the hosting yourself. At 24 hours a day, 365 days a year, you would be blowing over $236,520 at a MINIMUM just for a single year. If you were to buy a 24/7/365 subscription at maximum cost, that would cost $324,120. On top of that, you can't resell the hardware later to recoup some of the cost, so to me this just seems kind of stupid. 

Something something IT, maintenance, costs, internet, whatever.

[PLME888]

wait what, 26 dollar per hour or 26k dollar per hour? damnit EU/US punctuation!

[Lurick]

4 minutes ago, PLME888 said:

wait what, 26 dollar per hour or 26k dollar per hour? damnit EU/US punctuation!

26 dollars per hour to 37 dollars per hour.

We use commas for thousands  

[Bananasplit_00]

well, i feel like renting one for an hour and seeing how many chrome tabs i can bring up before it crumbles...

[dexT]

10 minutes ago, Bananasplit_00 said:

well, i feel like renting one for an hour and seeing how many chrome tabs i can bring up before it crumbles...

Rent it for an hour and legally pentest it with a group.

[ScratchCat]

But can it run Crisis

I'm impressed with the value, for the cost of a day out you can use one of the most powerful servers made available. This will most likely be used by people who need the entire dataset in RAM, use the server to generate/process the set to a more manageable size. I can't think of a use case where you would need 4TB 24/7 and not be able to spread it across multiple machines. 

[Drak3]

2 hours ago, Bananasplit_00 said:

well, i feel like renting one for an hour and seeing how many chrome tabs i can bring up before it crumbles...

Two, two tabs in Chrome, pyo.

[piemadd]

guys, we're forgetting something here

 

"virtual machine"

 

THAT MEANS THAT THE ACTUAL MACHINE HAS MORE RAM OR SOMETHING LIKE THAT

[Drak3]

Just now, Implosivetech said:

guys, we're forgetting something here

 

"virtual machine"

 

THAT MEANS THAT THE ACTUAL MACHINE HAS MORE RAM OR SOMETHING LIKE THAT

Page Files and swap partitions exist, pyo.

[piemadd]

Just now, Drak3 said:

Page Files and swap partitions exist, pyo.

but what if

 

what if the machine had more than 2tb of memory

 

oh fuk i missed 1,111 posts

[Drak3]

Just now, Implosivetech said:

but what if

 

what if the machine had more than 2tb of memory

 

oh fuk i missed 1,111 posts

Alternatively, what if it had swap partitions that totaled 3.5TiB and only 512GiB of physical RAM dedicated to the VM, pyo?

[piemadd]

4 minutes ago, Drak3 said:

Alternatively, what if it had swap partitions that totaled 3.5TiB and only 512GiB of physical RAM dedicated to the VM, pyo?

wouldn't it just be simpler to lend out each machine at a fraction of the cost at that price, and let the users of the machines put them into virtual machines? That way, companies only needing that fraction of ram and bandwith could use it.

[James]

On 9/17/2017 at 9:36 AM, CUDA_Cores said:

If you were the very small group of companies who need that much RAM, it would probably be cheaper to buy the hardware outright and do the hosting yourself. At 24 hours a day, 365 days a year, you would be blowing over $236,520 at a MINIMUM just for a single year. If you were to buy a 24/7/365 subscription at maximum cost, that would cost $324,120. On top of that, you can't resell the hardware later to recoup some of the cost, so to me this just seems kind of stupid. 

Thing is, these companies likely don't need it 24h/day, 365 days a year. They just need to be able to handle peak usage.

[WMGroomAK]

Reading this, I'm reminded of a Phoronix Article from a couple of days ago where they tested some dual socket Xeon Gold Servers and Single Socket AMD Epyc Server against some of Amazon's middle tier EC2 cloud instances.  Actually showed some really interesting results where all the solutions were trading blows based on different tasks...  Of course the highest Amazon Service tested was only the r4. 16xlarge which consists of 64 vCPUs & 488 GB of RAM for 4.256/hour so it would be interesting to see how this all stacks up to the x1e services.

 

https://www.phoronix.com/scan.php?page=article&item=amd-epyc-ec2&num=1