New Server Chip Niagara

* * Beatles-Beatles writes "Sun recently announced their latest release in server technology. The UltraSparc T1 processor, code-named Niagara, has eight computing engines on a single chip, with each core capable of handling up to four tasks at once." With this new processor Sun hopes to get a leg up on the competition. The Niagra chip is being billed as an "eco-friendly" chip because of its low power requirements. From the article: " [...] removing the world's Web servers and replacing them with half the number of UltraSparc T1-based systems would have the same effect on carbon dioxide emissions as planting 1 million trees."

Comments

Sun has the fix for global warming!

[sgtboost]

Just buy their new processor...it's equivalent to planting a megajillion trees!

Re:Sun has the fix for global warming!

[LordoftheLemmings]

Are people still going on about loggin? Shortsited logging companies huh? And what happens after a logging company clears out part of a forest? The replant the trees? Yes they do which is why we have more trees today in america then we did over 100 years ago.

nasty stuff

[Anonymous Coward]

What about all the real nasty chemicals that go into the manufacturing process of chips .. eg arsenic and acid !!!

Re:nasty stuff

[youngerpants]

Arsenic = Naturally occuring substance (you know when potatoes grow a green scum... thats arsenic)

Acid = Oh no, not acid, which type, I really hope it isnt ascorbic acid, nasty that one, I try to avoid it at all costs... hey, my teeth have all fallen out

Re:nasty stuff

[wytcld]

Chips will be manufactured anyway. The question is what the useful lifetime of the chips that are manufactured is, and what the power consumption of those chips over their lifetime will be, in ratio to the work they perform.

Sparc's strength in the early Internet days was always throughput - even under load - rather than speed. Sun also built more reliable hardware. I switched from Sun to AMD/Linux for Webservers early on, but with energy costs rising quickly, I'll be taking another look at Sun. Where these probably can matter most is for large Web farms, which currently tend to be commodity Linux boxen. But those are throw-away machines - chips headed onward to the landfill after just a couple years.

Re:nasty stuff

[moro_666]

indeed, you have acid's all over the place.

alcohol is acid, quite a weak one, but still having the effects and reactions as the rest of acids have.
most of the batteries that you throw into your ipod contain acid, so does the battery in your car.
if you let CO2 and water together, they create a nice (and a bit unstable) carbonic acid, and that is basically all over the place.
even your body contains so many acids that i would get a ban on slashdot for even naming all of these.
compared to the acids created by gasoline engines and powerstations in the atmosphere, the electionics production isn't even worth mentioning.

sun is making the right move, more computing power for less watt-per-hour, and if they can spare the energy used during producing too, it's even better (and more profitable for them, since they pay for that). having a 200W P4 screaming under your table just to play solitaire is really wicked from the energetic point of view. so is driving an engine overbloated suv just to get one fat butt from one place to another. regular swedish buses that carry 30 people have the same size of engines as hummers or corvettes, sniffing the word 'wasted' anywhere ?

while this cpu will be nogood for playing doom3, it will be a very good chip for handling many many many threads'n'processes at once and therefor be ideal for running webservers and mailservers and other type of multiple client handling services. way to go sun, i hope amd will do an amd athlon 64 X32 some time soon too :)

too bad i can't afford this stuff anytime soon :(

sadly when you are interested in the price of the latest server, you're not rich enough to buy it

Re:nasty stuff

[AKAImBatman]

I like Sun hardware, always have. But what happens when in "just a couple of years", these Sun chips aren't all that fast anymore? Do you keep them around just because you paid a lot for them?

I don't know about you, but I can always find a use for a Sun machine. They're built to last, and can often still be useful for a decade after their manufacture. The worst case is that you can resell your old machines to a refurbisher like AnySystem so that it can gain new life in someone else's possession. I know of plenty of companies where the AnySystem servers are powerhouses for the work they need to do. I also know of a lot of developers and sysadmins who would like a Sun Workstation, but can't afford new. Again, AnySystem (or Ebay, take your pick) can provide them with a system that meets or exceeds their needs. :-)

Re:nasty stuff

[InvalidError]

"Sparc's strength in the early Internet days was always throughput - even under load"

Are you implying that you can have useful throughput under no load? How do you measure this idle throughput advantage?

The Intel/AMD architectures are historically single-threaded desktop-centric where the most important thing usually is to run one thing really fast. Sun, however, was always in the HPC/workstation game where overall throughput matters most, latencies and single-thread performance be damned. These two groups were playing pretty different games up to recently.

But now, Intel/AMD have hit a GHz and complexity brick wall. They are forced to promote multi-threading multi-core at the desktop-level and optimize their future desktop chip designs for multi-threaded application throughput rather than single-threaded performance. Imagine what would happen if AMD and Intel could afford to quit competing on single-threaded performance overnight: goodbye complex deep out-of-order execution, goodbye branch-prediction and speculative execution - those transistors would be much better spent on implementing quad-threading cores to keep every pipeline filled with useful instructions that will retire cleanly on every clock.

Sacrificing single-thread performance for simultaneous multi-threaded throughput in the above-described way has been the name of Sun's game for the last few years.

Obsession with single-threaded performance is what costs current x86 CPUs the most power. Of course, in the P4/HT case, there is the added power and transistor costs of trying to be a jack-of-all-trades who predictably turned out as a master-of-none. (The P4's uOP replay engine is a neat idea... but re-executing the same stupid uOPs until they meet retirement conditions is woefully wasteful, whoever designed and bothered to patent this should be fired.)

Easy

[Laz10]

Let's go plant some trees then.

It is a pretty safe bet that 1 million trees are way cheaper than Sun technology.

Re:Easy

[Moby Cock]

I doubt they considered the enegy expended in creating this new design and manufacturing this new design. I'd like to see how they came up with such nonsense.

Re:Easy

[AvitarX]

First, they bent over. Then they reached in their ass. Lastly they published it as fact.

Better link

[Anonymous Coward]

Better link here.

Right.

[Skye16]

" [...] removing the world's Web servers and replacing them with half the number of UltraSparc T1-based systems would have the same effect on carbon dioxide emissions as planting 1 million trees."

I'm sure it would. But then, after replacing all these servers with the UltraSparc T1-based systems, we'd have to cut down 1 million trees just to print the money Sun would be making back in profit. Weeee!

Re:Right.

[mwvdlee]

"[...] removing the world's Web servers and replacing them with half the number of UltraSparc T1-based systems would have the same effect on carbon dioxide emissions as planting 1 million trees."

Removing the world's Web servers and not replacing them at all would have an even better effect on carbon dioxide emissions.

I'm curious as to how they calculated it, though. Are they talking 1 year running time or 100? Are they taking into account the energy required to build those new systems? Do they supply the new hardware's manuals on paper? How does it compare to a similar scenario using other replacement hardware?

Re:Right.

[Anarke_Incarnate]

At least in the US, money is printed on cotton pulp and not wood pulp.

raw power

[emptybody]

Niagara systems take the concept of dual core processors (with which most of you are familiar), and goes to an absolute extreme - building 8 cores, each capable of running 4 jobs simultaneously (4 threads), onto a single chip. Doing the math, we'll be delivering a 32-way chip, running 9.6GHz, which sips power (about 70 watts). , JonathanSchwartz BLOG.

This is why I got into Sysadmin 15 years ago.
To play with big honkin fast machines and new technology that makes your head spin.
Just musing about the name. Think of your kitchen sink faucet.
Now think of all the faucets in your house turned on at once.
Now think of all the faucets on your street turned on too.
Add all the faucets in your community.
Keep on thinking of how many faucets in how many communities it would take to equal the raw power behind something so large as Niagra falls.

Am I hooked?
You bet.

Re:raw power

[rkhalloran]

For a lot of uses today, a swarm of processors like this makes more sense than driving one CPU hellishly fast so it can task-switch quick enough to get around to everything.

Sun traditionally has been very good at engineering the interconnects so I expect the actual throughput on this is pretty good.

Will be interesting to see how well this does.

Re:raw power

[anOminousCow]

Good luck to Sun and their new Viagra chip. I'm sure it will help them get their third leg up on the competition.

What about I/O?

[Max Romantschuk]

Eight cores at four threads is 32 simultaneous threads. Nice, but what about memory bandwith? Each thread needs proper I/O if this is actually going to do any good... Anyone have any real info on this marchitecture?

Re:What about I/O?

[SQL Error]

Specs here. Four 144-bit DDR2-533 interfaces. That's more memory bandwidth than a quad-Opteron system.

Re:What about I/O?

[HaydnH]

Sun always builds their systems to be balanced and avoid bottlenecks, it's the first thing you learn about on the internal training courses so needless to say the I/O is fast enough.

Haydn.

Re:What about I/O?

[the eric conspiracy]

The 4 threads per core is designed exactly for that issue. If a thread is waiting for memory, execution can proceed on a different thread.

Re:What about I/O?

[Paul Jakma]

The whole point of Niagara is to get higher throughput *despite* memory latencies.

Re:What about I/O?

[fitten]

I haven't looked at the actual timings, but the vast majority of memory operations are on cache lines. So, you have to use some math to determine how fast a cache line is filled using each technology to see what the difference is. It isn't all about the time to get the first byte in, it's about getting the cache line in (which is 256 bits - 2x128-bit reads on our favorite x86 machines). DDR will still be a little faster, I'm guessing, but the initial read latency is amortized over the two reads in a dual-channel memory configuration.

Besides, I think Sun is saying that they will be using faster than DDR2-533 memory *and* the system is more tollerant of latency issues than the AMD and Intel parts that we are more familiar with. Of course, I'm taking their word with a grain of salt because even four threads on a single core can easily all be stalled given that at 2GHz, a single cache load may be the equivalent of over 200 instructions. So, it's quite imaginable that all four contexts can be stalled for main memory accses simultaneously given the rather small caches it has. Also, IIRC, even the L1 caches have a 3-cycle penalty which seems to me like that alone will guarantee many short stalls (like every instruction/bundle fetch).

Re:What about I/O?

[loose_cannon_gamer]

Ahhhhh, I find that parent very informative indeed. So this is really like 8 cores and quad level hyperthreading, where the first thread will be running 70%+ of the time, and the second thread will get scheduled when the first thread blocks, and run for 25% of the time, and the third thread will get scheduled when the first two are both blocked, and run for 4.8% of the time, and that fourth thread, well, 0.2% is all it gets. (And yes, I just pulled these numbers out of the air).

Just out of curiousity, shouldn't there be some kind of 'quality of service' notion when claiming how many threads you run simultaneously? I just seem to recall that hyperthreading on the P4 gives about a 15% throughput increase with a second thread, and I can't imagine that you get anything other than similarly diminishing returns if all four threads are fighting for the same memory bandwidth and execution units.

Just to explain, it is my understanding that hyperthreading is little more than opportunistic hardware level thread scheduling -- while one is blocked on I/O or something, two can use the same execution units *carefully* to accomplish work. If my analysis is right, and I could be out in space, you might as well build a chip that can run 1024 or 1048576 threads simultaneously (sure, you need a little extra hardware per thread), but you're still only going to see work get done by the first 2, maybe 3, although I doubt even that many.

Dumb

[pubjames]

Dear Sun,

RE: your statement:

Removing the world's Web servers and replacing them with half the number of UltraSparc T1-based systems would have the same effect on carbon dioxide emissions as planting 1 million trees.

Please engage brain before opening mouth.

Thanks.

Comparisons to the cell?

[Anonymous Coward]

These processors are a step in a different direction. Like the cell processor, they lack features like branch prediction, have small, very simple pipelines, etc. However, that isn't really all that bad, esp. on some tasks where your CPU is mostly just idling waiting for IO to finish anyway. I wonder if these "simple but gets the job done" CPUs will see an even wider market in the future. As the article said, they are cheaper and consume less power than their competitors

Re:Comparisons to the cell?

[Quicksilver]

Not really the same thing at all. The cell uses one master to control several specialized units. Niagara is just what is sounds like. 8 cores on 1 piece of silicon. They all are the same and the all can run any Sparc code.... unlike the Cell which isn't compatible with anything and each unit can only work on what it is specialized in.

And...

[Moby Cock]

removing the world's Web servers and replacing them with half the number of UltraSparc T1-based systems would have the same effect on carbon dioxide emissions as planting 1 million trees

"And it has the added benefit of lining my pocket."

Re:And...

[popeyethesailor]

And probably bankrupt atleast a million small-medium web hosts ;)

Re:Apple need this

[Anonymous Coward]

low power requirements != low enough for a laptop.

Niagra = 70 watts
G4 = 19 watts

Impractical

[Anonymous Coward]

I don't know many people who have a server room large enough to hold a million trees.

(twiddles thumbs for the remaining 17 seconds. Lahdy dahdy hum dum dum)

Massively multi-core x86s

[rbanffy]

AMD or VIA would build a cheap multi-core x86 based on VIA's or Geode cores... Sun could sell systems with them as developer boxes running Solaris 10.

BTW, what would happen to performance if you started with a Geode core and spent the rest of your wafer-area budget with Itanic-size caches?

For now, I have no hope to have one of these on my desktop anytime soon.

Re:Massively multi-core x86s

[tomstdenis]

Dude, a PPC 440 is 6mm^2 and consumes 700mW of power at 667Mhz. You could easily fit a dozen on a die the size of a P4 and still take FAR less than this Niagra core. At a rated 1200MIPS per core a die with eight of them would net you close to 9600MIPS max. Of course you'd need some form of L2 cache and high speed internal bus.

Similar the new ARM cores Cortex it takes roughly the same power at 1Ghz which gives it apparently 2000MIPS. The area is about the same as PPC 440. So in theory you could hook 4-8 of these up as well and get a killer chip too..

Point is Suns quotes of being "2 possibly 3 generations ahead" is totally bullshit. They're at most one generation ahead. It takes one multi-core ARM or PPC to totally destroy this.

Tom

Re:Massively multi-core x86s

[tomstdenis]

Law of diminishing returns?

Compare a 2MB L2 cache on a P4 to a box with 1MB of L2 or 512K ... the cache ends up contributing less and less to the overall performance.

What is also important is associativity. If you have a low-assoc cache, meaning a given address has few places in the cache it could reside you end up wasting more space. That's why [iirc] the AMD processors have high associvity L2 caches. They make good use of the 512K available.

At my previous job we built Gentoo distros on 128 and 256K semprons and the time to build wasn't really that different even though we were building 100s of packages.

So you could have a relatively high performance web server or file store [or whatever] without the 2MB of cache stuck on the back of the thing.

Tom

The lowdown

[gtoomey]

http://www.sun.com/processors/UltraSPARC-T1/specs. xml

Since the story is devoid of content:
- up to 8 cores, 4 threads per core
- integrated RSA
- 3MB L2 cache
- 90nm process
- 1.2 GHz

Not appropriate for all types of workloads

[XNormal]

You need at least a dozen concurrent threads or processes before you can make good use of this CPU's power. Certainly not a good idea for desktops. An excellent match for web servers. Other server-type workloads (e.g. database, file server) may need some tuning to make the best of this architecture.

Re:Not appropriate for all types of workloads

[juancn]

Not necessarily, at the time of this writting, my desktop machine has 510 threads belonging to 48 processes (I'll grant you that is not a typical setting, it is a development workstation).

Single threaded applications are pretty rare, most modern applications have more than one path of execution (you cannot afford to freeze the screen while saving for example). Network I/O is much easier to program with threading, in oposition from asynchronous I/O (think a browser with several tabs/windows open).

I wouldn't so lightly assume that multi-threading support is only desirable for server systems.

Anyway, this has been designed mostlly for servers, consider that a typical Java server will have around 50-100 threads (maybe more than 300). This is where this technology makes more sense.

The processor after this

[Xerp]

Will obviously have the name "C1ala1s"

A new measure of CPU performance...the tree (T)

[ip_freely_2000]

We can add this performance criteria to our system selection process...how many trees are saved. Featuring the Sun chip which is the world's first Megatree (MT) performer.

Why I remember when I was a lad we had Kilotree performers, and we were glad for it!!!

Sounds like an anti-slogan

[suso]

How about: I'd rather plant a tree than run on a Sun.

Misread Code Name

[FJ]

Am I the only one who read the code name as Viagra?

I knew sun was having troubles but not THAT kind of trouble.

Re:Misread Code Name

[ceeam]

Well, Suns are quite well-known for their decent uptimes!

Re:I don't know about the rest of the world

[ray-auch]

In that case the old inefficient systems would still be running, using power, hence no environmental gain.

The only way to get the claimed environmental gain would be if the old systems were never used again - which then does raise the landfill etc. issues

Ultrasparc III Cores

[the eric conspiracy]

The Niagara uses Ultrasparc III type cores which have limited single thread performance. This limits this design to certain applications that are highly concurrent in nature. More interesting is the Next Gen Rock CPU which will have highly parallel Rock CPUs.

Re:"The Niagra chip"

[ettlz]

That's because "with this new processor Sun hopes to get a leg over on the competition" (e.g., fuck Intel?).

The Spelling of NIAGARA

[CrazySpence]

It is not Niagra. Thank you

Why is it named Niagra?

[Unnamed Chickenheart]

...is it good at memory leakage? ^_^

Or is it perhaps not as low-power as they clame: maybe it require a huuuge current? ^_^

One should carefully name ones product. Its fate may stand or fall on it ^_^

New server chip Viagra

[ardchoille]

Whoa, talk about "uptime".

Mandatory sed quote

[fm2503]

s/n/v/

Re:So that's no effect at all, then

[Yartrebo]

Right on.

To be pedantic, planting trees (unless it's done on soil that was used for industrial agriculture, which has pretty much giving up its carbon already), will generally cause a release of CO2 from the ground. Even once the forest becomes mature, the net release of CO2 is positive in many cases (especially if the land used to be grassland).

But assuming that is ignored, a million trees:
  - Is nothing. Assuming they're Christmas trees, it's about a square kilometre. It's also about 1/100th of the annual harvest in the USA.
  - Is meaningless. Tell me in megatonnes of CO2 or gigawatts how much this will save, and if it doesn't equal a megatonne/yr or gigawatt, then it is just a drop in the bucket. Probably less of an effect than eradicating all spyware (thus causing less PCs to be replaced by lazy or ignorant or rich PC owners).

In other news

[frostfreek]

Sun announced today, that they will be chopping down one tree for every new system sold!

Am I the only one...

[Cow Jones]

... who misread the topic "New Server Chip Niagara" as "New Super Cheap Viagra"?
That would at least be an honest slashvertisments for a change.

Re: appropriate for all types of workloads

[LeninZhiv]

Um, yeah, but given that your two most strenuous processes are using 4% and 2% of your CPU respectively, with your machine posting mighty load averages of 0.33, 0.30, 0.18, I think you can safely hold off on upgrading to an 8-core processor any time soon--unless some of those threads aren't sleeping fast enough for you :-)

Arbor Day Foundation

[jbeaupre]

If you're willing to do the planting, the National Arbor Day Foundation will send you 10 trees for $10. http://www.arborday.org/shopping/Memberships/membe rships.cfm Get 10% of the registered users on Slashdot to sign up for this (and plant them) and you're close to a million.

Re:Except...

[Pollardito]

reminds me of this Newsweek article from a few months back :

That boom comes from buyers like Roberta Gray, who threw away most of her old clothes to make room for a new wardrobe from Greenloop, a boutique in Portland, Ore. A newly converted vegetarian and yogi, Gray, 43, wanted clothes that matched her healthier lifestyle. "When I buy things there, they last," she says. "They're of good quality, and I feel good about that."

it seems like her old wardrobe was lasting just fine until she threw them out!

The low Sparc of high heeled boys...

[argent]

The Sparc architecture has always had a problem getting good straight-line performance, because of the way the register windows constrain the compiler by exposing a small register set to the optimiser while still having the large context switch overhead of a large register set. Sun has long used multiple register contexts and microthreading-like techniques supported by the OS to get good multitasking performance despite these shortcomings, so this is a natural extension of the Sparc family.

And, yes, I'm sure straight-line single-processor performance will be nothing exceptional. But don't knock the impact of lower power... if it uses half the power of a comparable dual-core Opteron then you can fit twice as many processors in a rack, just because of the cooling requirements.

I also wonder what "handling up to four tasks at once" means. This could simply mean they have four traditional Sparc register contexts per core. TFA doesn't go into detail there.

Good Chip; Bad Angle

[MidKnight]

Sun has been talking about this puppy for a while now, and it's good to seem them deliver it. It does round out their processor strategy pretty nicely: AMD on the low end, and if you want obscene performance per-CPU at the high end you get this guy. I'll be interested to see some performance numbers.

Typical Sun though: crap-tacular marketing. What's the deal with the "eco-friendly" angle? See Sun's front page. Which CTO's actually care about that again? It's just stupid; saving the planet is a great corporate goal, but hopefully Sun is a bit more concerned with their bottom line, where they haven't consistently made a profit in 5 years.

Re:Or even easier....

[nelsonal]

Please mod this up, recycling paper has zero impact on cutting of forests, paper pulp "trees" are grown on farms.

Re:Apple need this

[level_headed_midwest]

That G4 is also far less powerful than the Niagra. We're talking about using these chips in servers, not laptops. Server manufacturers do want to reduce power consumption and heat output, but they need a lot more porcessing power than five-year-old laptop chips (such as the G4) can provide. 70W is quite low for a server as most server chips are at least twice as power-hungry as that.