Too Many Servers Could Mean No New Homes In Parts of the UK (gizmodo.com) 133
An anonymous reader quotes a report from Gizmodo: Data centers have caused skyrocketing power demand in parts of London. Now, new housing construction could be banned for more than a decade in some neighborhoods of the UK's biggest city because the electricity grid is reaching capacity, as first reported on by the Financial Times. The reason: too many data centers are taking up too much electricity and hogging available fiber optic cables. The Financial Times obtained multiple letters sent from the city's government, the Greater London Authority (GLA), to developers. "Major new applicants to the distribution network... including housing developments, commercial premises and industrial activities will have to wait several years to receive new electricity connections," said one note, according to the news outlet.
The GLA also confirmed the grid issue to Gizmodo in an email, and sent along text from one of the letters, which noted that for some areas utilities are saying "electricity connections will not be available for their sites until 2027 to 2030." Though the Financial Times reported that at least one letter indicated making the necessary electric grid updates in London could take up until 2035. [...] "Data centres use large quantities of electricity, the equivalent of towns or small cities, to power servers and ensure resilience in service," one of the GLA letters seen by the Financial Times reportedly said. [...] Developers are "still getting their heads round this, but our basic understanding is that developments of 25 units or more will be affected. Our understanding is that you just can't build them," said David O'Leary, policy director at the Home Builders Federation, a trade body. Combined, those sections of London contain about 5,000 homes and make up about 11% of the city's housing supply, according the Financial Times. A spokesperson for the London Mayor told Gizmodo in a statement: "The Mayor is very concerned that electricity capacity constraints in three West London boroughs are creating a significant challenge for developers securing timely connections to the electricity network, which could affect the delivery of thousands of much-needed homes...The increased demand for electricity capacity in the area is believed to be largely due to a rapid influx of batteries and data centers."
The GLA also confirmed the grid issue to Gizmodo in an email, and sent along text from one of the letters, which noted that for some areas utilities are saying "electricity connections will not be available for their sites until 2027 to 2030." Though the Financial Times reported that at least one letter indicated making the necessary electric grid updates in London could take up until 2035. [...] "Data centres use large quantities of electricity, the equivalent of towns or small cities, to power servers and ensure resilience in service," one of the GLA letters seen by the Financial Times reportedly said. [...] Developers are "still getting their heads round this, but our basic understanding is that developments of 25 units or more will be affected. Our understanding is that you just can't build them," said David O'Leary, policy director at the Home Builders Federation, a trade body. Combined, those sections of London contain about 5,000 homes and make up about 11% of the city's housing supply, according the Financial Times. A spokesperson for the London Mayor told Gizmodo in a statement: "The Mayor is very concerned that electricity capacity constraints in three West London boroughs are creating a significant challenge for developers securing timely connections to the electricity network, which could affect the delivery of thousands of much-needed homes...The increased demand for electricity capacity in the area is believed to be largely due to a rapid influx of batteries and data centers."
wat (Score:4, Insightful)
The increased demand for electricity capacity in the area is believed to be largely due to a rapid influx of batteries and data centers.
Batteries can relieve grid load problems by letting people draw power from them when grid capacity is unavailable, or even more directly by being used for grid stabilization. In what way are they supposed to be responsible for an increase in demand? Data centers yes, batteries... buh?
Re: wat (Score:2)
Batteries can relieve grid load problems by letting people draw power from them when grid capacity is unavailable, or even more directly by being used for grid stabilization. In what way are they supposed to be responsible for an increase in demand? Data centers yes, batteries... buh?
Batteries allow people to evade rolling blackouts. When demand exceeds supply, utilities decrease demand by implementing rolling blackouts: periods of several hours per day power is turned off to a group of homes. When that period is finished, power is restored and turned off for a different group of homes. In this way, the utility is never providing service to 100% of homes at once. By varying the duration and frequency of the rolling blackouts, the utility can adjust that service percentage to bring i
Re: wat (Score:2)
Re:wat (Score:4, Informative)
Well there's an initial load spike at least if you're charging a bunch at once.
The obvious solution is to not all charge at once.
If the grid is overloaded, there will be a slight drop in voltage. The battery charger can detect this with a device that costs about 2 cents. It can then back off on charging and try again later.
Re: (Score:3)
The obvious solution is to not all charge at once.
The trick is to get people to actually do that.
Re: wat (Score:2)
Re: (Score:2)
The trick is to get people to actually do that.
The obvious way to do that is to raise the price of electricity when demand is high and lower the price when demand is low.
My electricity is already sorta priced this way. I pay 7 cents between 2 and 4 am (when my EV is programmed to charge), 20 cents from 2 pm to 7 pm, and 12 cents any other time.
Switching from clock-based to actually measuring the load isn't a big leap.
Re: (Score:2)
Switching from clock-based to actually measuring the load isn't a big leap.
if you actually ever had thought it thought, you would know: yes it is. Every meter wod neex a data connection. If you want them to be smart meters, able to acti ate and deactivate devices, it gets even more complex.
Internet does not exist ? (Score:2)
Complex, oh my god.
Like, Internet does not exist ?
Re: (Score:2)
You need a second internet connection. As I pointed out already. I called it "data", as in real life it usually goes via GPRS. Base cost for my power plan is $5,00 per month. The cheapest data plans - regardless if landline or GPRS are minimum $10. So my base cost for power would triple ... either I have to pay it: or the power company has to provide it.
That was the start - any more problems you do not foresee? Good, that you are not managing such problems then.
I already answered to another idiot above, so
Re: (Score:2)
Switching from clock-based to actually measuring the load isn't a big leap. if you actually ever had thought it thought, you would know: yes it is. Every meter wod neex a data connection. If you want them to be smart meters, able to acti ate and deactivate devices, it gets even more complex.
Is the UK not using smart meters? They are prevalent in the US. As for demand billing, it's already possible and in use in the US (see Griddy in Texas for how that can end up when taken to extremes). Ditto for controlling smart devices. Again, see Texas last winter. The power company just needs to hook the appropriate APIs from the smart device makers. Doesn't even need to happen at the meter, it happens at the power company datacenter.
It's not flip-a-switch easy of course, but it's not an unsolved proble
Re: (Score:2)
It's not flip-a-switch easy of course, but it's not an unsolved problem either.
It is an unsolved problem. However it is not unsolveable if that is your point.
Many things need to come together to make smart meters work. And one of the big hurdles are data centers as you named them. Aka: the software in power companies. Power companies are notoriously bad in making software.
You have five main problems:
a) reading data from the smart meters (which basically includes the requirement to store/buffer enough to cov
Re: wat (Score:2)
Re: (Score:2)
Smart meters are already common.
I have one. So do millions of other people.
The meter doesn't deactivate devices. The devices control their own power consumption. So far, I have only two devices that do this: My car charger and the compressor in my AC.
Re: (Score:2)
Smart meters are already common.
Nope they are not.
I have one. So do millions of other people.
That is not common, that is rare. Hint: a million is 1/1000th of a billion. The planet has 8 billion people.
The meter doesn't deactivate devices.
Gosh nitpicking?
The SmartMeter sends the signal to the device that it is time to activate - oops.
Re: wat (Score:2)
Re: (Score:2)
If they are datacentres it might be that they all want to charge their batteries at the same time, when energy is cheap. For big energy consumers it is worth buying a battery to time shift consumption from the grid.
Re: (Score:2)
Energy is cheap at those times because demand is lower.
Re: (Score:2)
It is now, but we are trying to shift demand there such as EV charging and heat pump use, so building a load of new houses with those technologies built in before the grid is ready for it doesn't sound like a good idea.
Re: (Score:2)
"These are backup batteries. The batteries are not being drained and charged all the time like a car. A good data center will also do a battery test at least once a month which creates negligible drain during a short test."
And they usually take the load until the backup generators are running.
Re: wat (Score:2)
Re: (Score:2)
If the grid is overloaded, there will be a slight drop in voltage. The battery charger can detect this with a device that costs about 2 cents.
I'm really digging this decentralized approach. Simple 8-bit program graphing voltage on any given power system over a 24hr period.
Re: (Score:2)
There can be, depending on the charge rate and technology. It's not unusual for the maximum charge rate to be much less than the maximum discharge rate. This is typical for SPSes for example.
Re: (Score:3)
The batteries come with their own machines which consume that power. The power to charge the batteries and run these machines is additional demand. It's a bit daft to pin it on the batteries and not the devices which are battery-powered though. They mean cars, by the way, and the whole thing is just a ruse to keep London's real estate market from tanking. You can't brexit and then also add more houses. Are you crazy? Who would want prices to go down?
To give people and idea of why this isn't just crazy AC talking and might actually be fully real, the UK has, since previous financial crises, had strict rules about only selling people mortgages that they could "afford" ("afford" is a pretty loose term - they need to be able to pay it and have a little left to eat; clothing, heating, transport and so on optional). Even that fairly weak protection has just been eliminated.
Re: wat (Score:2)
The only bit of mortgage affordability tests scrapped is the interest rate stress test that required the mortgage to remain affordable if interest rates rise by 3%.
Other requirements, including loan to income ratios remain.
Re: wat (Score:2)
Ban datacenters instead. (Score:5, Insightful)
Datacenters will then have the incentive to actually be efficient. Lots of energy wasted doing things in shitty ways, like running things written in interpreted or VM languages because the programmers are too stupid and lazy.
Or at least stop doing unnecessary shit with all that unnecessary data.
Re:Ban bitcoin as well! (Score:3, Insightful)
Re:Ban bitcoin as well!
Re:Ban bitcoin as well! (Score:4, Informative)
Re:Ban bitcoin as well!
No sane person mines bitcoins in the UK, where the cost of power is about three times what Americans pay.
Re: (Score:2)
Google the UK energy price cap. Look at ofgem who will tell you the actual rate instead of some bullshit "average household total cost" that means nothing.
28p/kWh.
Weep for your limey cousins.
Re: (Score:2)
That won't happen. The government is very pro big business and won't tax it. In the general hierarchy of needs, datacentres come way above housing for plebs.
Re: (Score:2)
To afford their houses, the plebs need jobs. They won't have those jobs if the businesses move elsewhere.
Re: (Score:2)
The old "don't you dare tax us or we will leave" gambit.
The reality is that if a shitty business environment was enough to make companies leave, brexit would have done it already.
Re: (Score:3)
Re: (Score:2)
Data centers are landlords. They sell space, cooling, physical security, and power.
They do not have any say in what language or CPU's their customers use.
What the city can do is charge data centers more for power. The data centers then pass that cost to their customers who are then incentivized to find ways to reduce power.
That won't happen. The data centers would become uncompetitive and shut down which would mean someone in the city planning office would lose out on their kickbacks for allowing the gri
Re: Ban datacenters instead. (Score:2)
Are the data centres actually in London, or are they further up the Thames valley towards places like Reading? It seems to me that land costs in London would put them off building there, let alone open themselves up to the city's regulations.
Re: (Score:2)
The better thing to do is to regulate/tax the power usage of datacenters.
They already kind of are.
Businesses, ergo datacentres are not given the energy price cap that ordinary consumers gets, so they have to pay full market rate for energy in the UK. This encourages them to use as little power as they can because the cost of energy has risen significantly in the UK over the last few years.
An additional tax wont change anything, it'll just make running a datacentre more expensive, ergo forcing them to raise prices and making overseas datacentres in Germany, the Netherlands,
Re: (Score:2)
Lots of energy wasted doing things in shitty ways, like running things written in interpreted or VM languages because the programmers are too stupid and lazy.
You're begging the question. First you should prove that data centres are actually wasting energy, then you need to quantify the energy, and then compare the savings to alternatives such as changing hardware.
Power is a cost and datacentres already have incentives to minimise it, which is precisely why so many go out of their way to design custom hardware to improve efficiency. However, the switch to ARM would make far more sense efficiency wise than your hate for interpreted languages.
Re: (Score:2)
Always let the best be the enemy of the good.
Good thinking, can we ban Python in London? :) (Score:5, Interesting)
Datacenters will then have the incentive to actually be efficient. Lots of energy wasted doing things in shitty ways, like running things written in interpreted or VM languages because the programmers are too stupid and lazy.
Good thinking. Let's ban Python and node.js in London! While we're at it, on the browser end, can we stop using 4000 frameworks and a megabyte of HTML per page? (I'm half joking)
I've long worried about the ecological practice of bad software engineering and devs' obsession with throwing as many frameworks into the mix as possible. It's already annoying that everything gets slower every year despite hardware getting faster....but even worse, that slowness is due to wasted electricity and generates a lot of heat. A certain amount of frameworks are beneficial, but any more just seems like waste. I've seen many simple CRUD REST apps take minutes to complete saves that could take 1/2 of a second if they just stopped misusing ORM. I am amazed at how slow modern web pages are and then I view source and I see why. A simple landing page has megabytes of nested span tags for no reason from what I can tell. The layout is simple and static, but the devs are so used to fancy frameworks a simple page that should be a few kilobytes is megabytes of nested containers.
I thought the move to cloud would make companies more cost sensitive and stop writing new production apps in node.js and Python to reduce costs when a Java/scala/Go version would be twice as fast and efficient. However, I was mistaken. I've seen from personal experience, most managers would rather waste millions of dollars in extra checks to AWS/Azure than spend time thinking about already deployed apps. One of our biggest expenses is a simple app written in Flask that needs about 10x more nodes than its Java siblings and all it does is simple CRUD, nothing that NEEDS python (it was written by a team we acquired in a merger)
I spent years advocating removing misused ORM and replacing it with native SQL in a customer-facing application. 1/100th of the number of lines of codes and moving an operation users complain about from 5 min wait time to 1/2 second wait time (really bad misuse of ORM). I wrote the code and demo-ed the 600x performance improvement...seems like a slam dunk, right?....nope....the managers would rather write huge checks every month than risk a refactor (and this was extremely low risk).
I'll wager half of the electricity consumed in a modern datacenter is waste....especially from lazy devs who, rather than learning a backend language that's fit for the task, would rather use their favorite scripting or browser language....the equivalent of me writing a UI in Java applets because I don't want to learn JavaScript.
Re: (Score:2)
the managers would rather write huge checks every month than risk a refactor
Been there. The problem is that too many companies have a zero-tolerance approach to failure. If you proposed a thing and it wasn't 110% successful then you don't just lose your bonus, you're out. O-W-T, out. So you don't propose anything remotely risky until the gun is at your head because at best you'll get a pat on the head for being right, but more likely you'll get your books for being slightly wrong.
I blame shareholders. Short term profits and "line goes up" leads to short term thinking. Biden's chipm
Re: (Score:2)
"The better thing to do is to regulate/tax the power usage of datacenters. "
The Tories are all on deregulation and detaxing.
Modern fission power (Score:3)
Doesn't risk environmental catastrophes, doesn't have high weaponization potential even in countries less stable than UK. Just build it out and have both datacenters and homes without carbon emissions or bowing to Russia.
Re: Modern fission power (Score:5, Informative)
The problem here isn't a general lack of energy, it's a lack of transformers and HV power lines. It's a distribution issue.
We have the same issue in NL. Too much power being generated in the NE parts, too much being required in the NW parts near Amsterdam. And not enough capacity on the grid to bring it from A to B.
I think the local utilities ordered about 100 main high voltage transformation installations but they are all built to order. Takes a long time.
If someone has a mass produced and cheap alternative, they'll make a lot of money.
Re: (Score:2)
The problem here isn't a general lack of energy, it's a lack of transformers and HV power lines. It's a distribution issue.
So it sounds like, rather than banning homes... or data centers... they should be concentrating on upgrading their grid to handle the increased usage, yes? No small feat, and not cheap by any means, but if you really want to anger voters, try telling them "sorry, you can't buy a home in this neighborhood because the server farm gets priority".
Re: (Score:2)
You're making an assumption that they aren't. They are. That's precisely why a timeline is mentioned. This shit takes time, a LOT of time. You don't just magic a few hundred MW in another direction on a whim.
The UK didn't tell them no new homes because server farm gets priority. The server farm is already there. The grid is already full. You want to really screw your voters, then shutdown industry (right as a recession is already looming) just so you can build houses for the resulting poor. While this kind
Re: (Score:2)
"The problem here isn't a general lack of energy, it's a lack of transformers and HV power lines. It's a distribution issue."
I guess Electricité de France, the second biggest provider in the UK just got nationalized by Macron and they'll invest in the EU instead of the contract-breaking Brits.
Re: (Score:2)
The UK government is trying to build new nuclear plants, but it's not going very well. The current one being built, Hinkley C, is costing several times the original budget and won't be ready for at least another decade. There is another new one that just been approved, Sizewell C, but the builder (EDF) is saying it will take at least 20 years to come online.
A couple of weeks ago the UK auctioned off-shore wind licences. The energy produced costs about a quarter of natural gas, and an even smaller fraction o
Re: (Score:2)
Re: (Score:2)
The real culprit is austerity (Score:4, Informative)
In the past when private companies could not be incentivized by profits to do something that was necessary for the public good the government would step in. But after four and a half almost five decades of non-stop austerity politics and a constant demand for tax cuts and only ever seem to go to the people at the top there just isn't any money to do that.
Re:The real culprit is austerity (Score:4, Informative)
Normally you would just build out more capacity for your power grid. We certainly have the technology to do so.
Um, that's what they're doing.
The clues are in the lines where they say things like "will have to wait several years to receive new electricity connections," and “electricity connections will not be available for their sites until 2027 to 2030.”
Re:The real culprit is austerity (Score:5, Informative)
Rest of UK (Score:3)
you tend to start ignoring increasingly important things
Yes, but what they are ignoring is the rest of the UK. Westminster has focused all its efforts of economic growth on London and the South East and directly created this problem. There is plenty of grid capacity elsewhere and growth would be more evenly spread out if they would invest in the transport infrastructure.
Instead, HS2 has been pruned down until it is almost useless, the transpennine upgrade was cancelled and yet London still got its Crossrail and now Crossrail2. This sort of myopic view of the
Re: (Score:2)
Yes, but with appropriate incentives (e.g. the ability to charge more to provide power when it's scarce) someone would have invested in this ahead of time. However, it's a difficult problem as power companies are natural monopolies so it's hard to set the incentives correctly. Given a large enough power grid it's possible to have a wholesale market which allows this kind of profit incentive but it's a complex problem since you also need certain levels of regulation to prevent issues like the TX disaster a
Re: (Score:2)
Re: (Score:2)
Were austerity in effect, nobody would be building out datacentres. Clearly there's a growth motive.
Re: (Score:2)
Re: (Score:2)
Re: The real culprit is austerity (Score:2)
That includes Scotland, which is very different to the SE of England.
Real question is? (Score:2)
Re: (Score:2)
What bunch of over paid incompetents are managing the grid? And what needs to be done to replace the leadership in whole, until things run right! Is it the government?
They are called the 'Tory Party' and they are very fond of drawing the magic cost cutting-sword from the stone and swinging it about.
Re: (Score:2)
Re:Real question is? (Score:4, Insightful)
Is it the government?
NO
It's been privatised because that makes it so much more efficient (not).
Instead of investing in growth for the benefit of all, profits (from a monopoly supply of an essential service) have been skimmed off to investors (who were / are mainly Tory party donors & chums), directors etc.
The same outcome as the privatisation of the generation capacity [now mostly in foreign hands] and the middlemen selling electricity to the end users with virtually no competition as the 'big six' energy suppliers control everything [but we can't say 'cartel'].
Still, all this must be an illusion, as Slashdot contributors keep telling me that public ownership is bad; private enterprise is good.
Re: (Score:2)
Well it was privatised in part due to being a mismanaged public utility. With politics there's a strange phenomenon where you are completely forbidden from breaking any eggs to make an omelette but for some reason you can throw all the eggs out and buy in pizza instead.
Batteries? (Score:2)
Batteries I can understand, as they buy power at cheap times, and use it during the expensive parts of the day, depriving the power company of all that juicy extra profit.
But Data Centres? Unless they are getting free power I would have thought they'd be paying a handsome sum for all the power they use.
Or is this just another case of the power company creating a problem so the government gives them cash to "fix it".
Re: (Score:2)
But Data Centres? Unless they are getting free power I would have thought they'd be paying a handsome sum for all the power they use.
What does that have to do with available capacity?
Re: (Score:2)
But Data Centres? Unless they are getting free power I would have thought they'd be paying a handsome sum for all the power they use.
What does that have to do with available capacity?
The fact that this handsome sum they are paying does not seem to have been invested in building extra capacity?
Re: (Score:2)
The fact that this handsome sum they are paying does not seem to have been invested in building extra capacity?
Why would it be, that would only lower the prices!
Re: (Score:2)
The fact that this handsome sum they are paying does not seem to have been invested in building extra capacity?
Why would it be, that would only lower the prices!
Well the UK public voted for a Tory Government, they are now getting what they wanted.
Re: (Score:2)
The fact that this handsome sum they are paying does not seem to have been invested in building extra capacity?
I'd rather they use it to subsidize poor people's electricity than build new houses for rich people in the center of London.
(this "housing shortage" is in a very specific place, it's not the whole of the UK as the headline might suggest)
Dupe? (Score:2)
Isn't this basically a Dupe [slashdot.org] of this earlier article?
Re: (Score:3)
Isn't this basically a Dupe [slashdot.org] of this earlier article?
Think of it as a hot fail-over server?
#FAILoverANDover
new BBC serial (Score:4, Funny)
"Are You Being Servered?"
The problems just keep piling up... (Score:2)
Electricity too expensive to heat buildings, not enough electricity for the data centres to turn into massive amounts of heat and dump it into the atmosphere, it is all so depressingly unsolvable.
Tell the far-right (Score:2)
Could have seen this coming (Score:3)
I wouldn't think there are that many data centers in Hounslow. It's next to Heathrow yet but most of that Infrastructure is Richmond a different borough. Hounslow also isn't particularly densely housed by UK city standards, it's a suburb.
It's very, very difficult to upgrade capacity in these areas as nearly all cables are buried and run under streets plus there isn't much room to put in a new substation feeder.
Also if you take a patch of land that had say 6 houses on it and put in 40 apartments you have just multiplied by 6 times the electricity draw. And a lot of these housing developments are on brownfield or old sites so it's an additional draw not a replacement.
It's common in other parts of the world for developers to get nailed for the cost of new electricity infrastructure beyond just running some feeds down the road.
The core problem is the UK, especially London and it's surrounds, has seen a population explosion over the last 10-20 years that the politicians said wouldn't happen and stuck their head in the sands when it was blindingly clear it was happening and didn't spend a $ on expanding housing, health systems, education etc etc to cope with it.
And that bastard Blair was in charge then not the Tories.
Re: (Score:2)
There's at least one: the Equinix DC opposite the north entrance to Heathrow. Note that Heathrow itself is a massive energy user.
There are also datacentres at Stockley Park, not far away.
It's not a new problem. We were told 10-15 years ago that power capacity in the area was very limited and that more power connections to datacentres nearby were going to be very difficult to get.
In particular that Equinix DC got an additional 1MW connection about a decade ago after several years' wait and was told that was
Re: Could have seen this coming (Score:2)
A decade where nothing was done? Don't be silly, there will have been tens of millions of pounds spent on meetings in nice conference centres all over the world to discuss the problem.
Poor planning (Score:3)
Regulate what data can be stored first? (Score:2)
If homes for people are being delayed, because a data centre is storing completely unessential data, then that needs regulation!
"Yeah, sorry home developers, you can't build here right now because Acme Data Ltd. needs all that power for an NFT blockchain solution."
I am joking - to a degree - on that premise.
But there's so TOTALLY some mileage here - what if that data centre is involved with cryptocurrency mining?
But the way data storage is going, where we start to push more and more computation processes at
Crowded housing (Score:2)
Combined, those sections of London contain about 5,000 homes and make up about 11% of the city's housing supply.
So all 9 million Londoners live in a housing supply of 46,000 homes? How do they fit all 200 people in each home?
Re: (Score:2)
Families of 4 are created by lottery and assigned for 1 week of housing per year.
Why is a problem ? (Score:2)
2) mkdir
3)
4) Profit!
Not a problem (Score:2)
The UK has twice as many homes in the UK than their are households. There is no requirement to build new homes there is twice the housing stock required.
Iceland! (Score:2)
Move the datacentres to Iceland, where's there's plenty of space and electricity is cheap.
What processors do servers run on? (Score:2)
I just wonder what CPUs all of these servers run on and why isn't anyone mandating efficiency improvements in them just like they do with cars.
Re:Intel CPUs are the most power hungry (Score:5, Insightful)
If we are going to do a big switcheroo from x86 to more efficient processors in the datacenter, we should go to RISC-V, not ARM.
RISC-V is more efficient, and a more modern architecture. It also has no licensing fees since it is open source, so the cost and admin overhead of new designs are lower.
Many universities use RISC-V for their computer architecture classes, so expertise is growing.
Re: (Score:2)
Is RISC-V really more efficient than ARM? The only benchmark I've seen substantiating that was using an embedded suite that isn't representative of typical server loads, or even running a full OS.
In any case, moving to RISC makes sense for some loads, where CPU performance is not critical. But for areas where heavy processing is needed then neither RISC-V nor ARM can compete with x86. It's not clear that they ever will be able to match x86, as for pure performance all the expected gains of RISC have turned
Re:Intel CPUs are the most power hungry (Score:5, Informative)
Is RISC-V really more efficient than ARM?
There are many reasons that RISC-V should be more efficient, and it actually is more efficient in the embedded space which (so far) is the only area where it is used commercially.
neither RISC-V nor ARM can compete with x86.
That is only because Intel has thrown hundreds of billions in R&D at the x86. There is no reason that x86 is inherently more performant. Apple has made huge progress with ARM by spending far less than Intel.
ARM code, because it's much less dense
RISC-V has code compression built-in for the non-embedded version. The code is more compact than either ARM or x86. So the cache can hold about 30% more code.
Re:Intel CPUs are the most power hungry (Score:4, Interesting)
I'm arguing that despite x86 being horrible, it turned out to be a better intermediate stage than RISC machine code did.
RISC-V compressed instruction set is similar to ARM Thumb. It reduces the size of some instructions, but doesn't do much for code density. The instructions are still RISC, you still need more than one of them to replicate the functions of many x86 op-codes.
Although nobody has built really high end RISC-V systems, I expect they will have the same issue as high end ARM: over reliance of massive memory bandwidth (and by extension caches).
Re: (Score:2)
RISC-V's compressed instructions are not the same as ARM's thumb mode.
Thumb mode is an independent but incompatible instruction set. You are either using all thumb instructions or none. ARM's thumb mode is so lame that thumb mode has been dumped from most modern ARM processors.
RISC-V compressed instructions work very differently. They can be intermixed with full-width instructions. In a typical program, about 60% of the code can be compressed, resulting in a 30% improvement in code density over what an AR
Re: (Score:2)
They can be intermixed with full-width instructions.
So it is more like a CISC style instruction set? Perhaps RISC-V was not the greatest name for this architecture.
Re: (Score:2)
So it is more like a CISC style instruction set?
No. It is still a load-store architecture with minimal ISA bloat. But in the very common case where the source register and destination register in an instruction are the same, there is an opportunity to reduce code size. You can quibble and say that makes it less RISCy, but ideological purity is not a good CPU design philosophy. It is better just to do what works.
Code size is one of the few advantages of CISC over RISC, and with RISC-V, even that is gone.
Re: Intel CPUs are the most power hungry (Score:2)
Weâ(TM)re still a long way from this being possible, despite a lot of improvments recently. GPU-based encoding tends to suffer from lower quality, poor rate control/HRD compliance and a BD-rate disadvantage. Transcoding pipelines also tend to lack 4:2:2 support, which is important for professional content and work-flows. Furthermo
Re: (Score:2)
When debating what we should be doing it's always important to remember what "CAN" we do. Much in the same way as nuclear is the solution to global warming (which we can't build in time) RISC-V sounds great on paper, which is about the limit to what is widely available.
Do a search for ARM and you get products. Do a search for RISC-V and you get statements of promise from SiFive along with a few demonstration board level products. Your server can't run on hopes and dreams.
Re:Intel CPUs are the most power hungry (Score:5, Interesting)
Re: (Score:2)
Getting more efficient devices will never result lower consumption, but will end up using more of those devices until they meet the limits.
Jevons paradox [slashdot.org]
Re: (Score:2)
Getting more efficient devices will never result lower consumption, but will end up using more of those devices until they meet the limits.
No, you're assuming that a datacentre is constrained exclusively by power use. People don't build datacentres until no power is available, power is cost centre not a limitation, the limitation is customer requirements and in some cases network capacity.
Re: (Score:2)
That's kind of true, when looking at the overall picture of the planet. But an organization that needs 100 desktops won't buy 200 desktops just because the new ones have lower power budgets. They will still only buy 100 desktops.
Reducing power consumption reduces the bottom line, which is all that really matters. Even uptime only matters insofar as it affects the bottom line, and it comes down to just doing the math of what it costs to swap hardware, how long it takes, blah blah blah.