936 karma · joined August 27, 2012
If you look at the big graph later on in the article, even for similar _physical_ core counts the Graviton does better than the Skylake - this implies there's another bottleneck.
It's worth pointing out as well that ARM's weaker memory model gives it many more opportunities for aggressive optimisations that can make better use of the available bandwidth.
It could be another "free for everyone" where everyone means "US only for now". Pretty frustrating when that happens.
The servers themselves come with unlimited transfer.
We've moved hundreds of thousands of dollars through them over the last couple of years with no problems on their end (we've had problems with banks failing to transfer correctly though..)
Luajit is a brilliant project.
That remain-backing and second-ref parties actually got the majority of votes makes it even worse. First past the post is not fit for a modern democracy.
That a single £100k salary is enough to maintain a household of two kids is neither here nor there (and also depends very much on where you live).
22 hours of free childcare a week is worth ~£9k a year post-tax. If either you or your partner earn a penny over £100k then you're no longer eligible for any of your children. So with two kids in nursery a penny pay rise might lead to a £30k pre-tax pay cut.
There is little to no reason why high performance RISC-V implementations can not achieve performance comparable to modern "CISC" cores.
As other commenters have said, VLIW doesn't suddenly make the problem go away and no amount of compiler magic is going to fix the fundamental issue that memory access on a modern processor is highly non-uniform and scheduling is not something that can be done statically ahead of time.
RISC-V is designed to be extended with complex instructions and accelerators and that's the scalable part.
In the extreme case, if all firms use a single model that underestimates the risk of an event then that event occurring poses a threat to the entire industry.
With more models, each making different errors, only parts of the industry are exposed to each error. The more models, the smaller the number of affected firms and with reinsurance between firms the risk to the whole industry is much smaller.
Worth pointing out for anyone else that it seems PlaidML is AGPL licensed - so maybe not worth getting too excited about if you have any commercial applications in mind.
I wonder how realistic running analytics over stored data is.
https://www.scaleway.com/pricing/ has 16GB just over $20
A 16GB of ram server could basically serve responses sub-millisecond at line rate for like $25/month.
The usability side you mention is what Brendan's suggesting is worth porting from DTrace, just the front-end that the users sees. That would just back onto eBPF and the user would know little different.
If your needs are met by the low end, you're forced to pay a premium.
If you want equivalent specs then sure, the prices are going to be similar but that's the whole point, with Android (and a PC) you've got a wide range of _choices_ on many parts of the specifications/price spectrum. With Apple you've got far fewer.
The same goes for laptops where you listed three different manufacturers.
However, OS X gets excluded because it immediately constrains my hardware purchases.
https://www.ovh.com/us/public-cloud/storage/object-storage/
It seems to be more expensive for storage but nearly half the cost for transfer.
Alienation of our nearest neighbours and the likely dissolution of the Union.
It's also looking likely that Brexit will cause problems for the Northern Ireland peace arrangement.
It's a real shame, it fits the bill perfectly but Dell will almost certainly get my money for an XPS 13 instead.
We have a large amount of free Azure credit and yet our experiences at actually trying to use the platform have been met with frustration.
When we tried about six months back, there were two different administration webapps that both offered different subsets of functionality. The IaaS offering seemed to still be tied very much in to the PaaS offers.
After a few hours of frustration I managed to spin up 8 instances where only 6 had connectivity with each other. I gave up the other 2 instances for lost.
When I finally got Spark running on the remaining 6 instances, local disk IO turned out to be a joke.
I gave up at that point.
Anyone have more recent experiences?
Also, once you're paying upfront for reserved instances you've lost one of the major advantages in your cashflow scaling with your usage.