Never one to cower in the face of hyperbole, Sun Microsystems has come out touting the new eight-core UltraSPARC T2 - aka Niagara II - chip as the world's fastest microprocessor. We've written so much about Niagara II that there's not much left with which to surprise you. As promised, the chip has just as many cores as Niagara …
"Niagara II does bring some fancier features. For one, Sun has included an on-chip 10GbE NIC. In addition, you'll find eight crypto acceleration units and eight lanes of PCI Express I/O, along with four memory controllers."
SPEC or no SPEC, this is one fast chip. There's enough beef on that chip to eliminate most of the bottlenecks in PowerPC and Intel systems right away - double the clock speed on a P4, for example, and you'll see a 10% to 20% overall speed increase. Double it on Niagra II and (aside from the puff of very expensive smoke, you should see at least a 40% speed increase.
Someone hand me a towel, I'm drooling!
Wow, 1.4GHz Niagara II faster than 4.7GHz POWER6
Wow, just looking at the SPECint and SPECfp of the Niagara II, Sun has been able to generate a chip that is faster than the 4.7GHz POWER6, that IBM just announced the other day.
And not just by 1 or 2%, but a good margin as well..
Niagara2 is Absolutely the fastest CPU in WORLD
Ashlee, how do you get off saying " But, er, lots of chips have leading benchmark scores and even more chips run at faster speeds than Niagara II, so we think the world's fastest claim is a bit too bold".???
Tell me, which CPU chip is faster on SPEC?
Niagara2 (1x1.4GHz chip) SPECint_rate2006 78.3 SPECfp_rate2006 61.5
Looking at SPEC.org, I see:
Power6 (1 x 4.7GHz) SPECint_rate2006 60.90 SPECfp_rate2006 58.0
Itanium2 9050 (2 x 1.6GHz) SPECint_rate2006 N/A SPECfp_rate2006 58.1
Xeon X5355 (2 x 2.66GHz) SPECint_rate2006 60.50 SPECfp_rate2006 60.9
Opteron 2222 SE (2 x 3GHz) SPECint_rate2006 60.40 SPECfp_rate2006 52.5
HyperVisor Driver Heavy MetAI.......
Bravo Sun...... Something for the Dedicated Great Games Player who, of course, needs to have a finger in every pie/spy in every port.
And re Victoria Falls, are we to Imagine AI Breakthrough akin to "Dr Livinstone, I presume?" IT does seems as if the computer and its Programs are merely there to provide Windows and Support Services to the Information and Intelligent Streaming from the Core Chip via ITs Internal Proprietary Networking Protocols....... which is very much Sun territory..... The Network is the Computer.
Is InterNetworking 42WinWin, ITs UltiMate Game? There is, of course, no need to answer that silly question as everyman and his dog knows the answer to anyone's Dream is "Yes, yes, ..... oh Please do not stop, don't stop just now, we're almost there". The Beginner's have an easier, although no less pleasurable for them, slope to climb..... "Yes, yes, ..... oh Please do not stop, don't stop just now, I'm almost there" :-)
Seventh Heaven GCHQ Territory for the Apache Brave and Lion Hearted Knight........ Virgin Fields 42 Virtualise Real and as Imagined.
Are you following this, Sir Richard, or is Media letting you down rather than giving you ITs Lift?
cc. Virgin Media via Registered Post.
"Give aman or a woman what they want and need, and you will Control them" for at least as long as they can accept IT .....which is a Beta Order Protocol than any Random Arbitrary Denial for Chaos instruction set
Niagara II is still a light thread niche unscalable chip
Niagara is good for light threads only. If you have any heavy threads with your web services applications it will slow to a crawl.
It still has 8 Keep It Simple (aka minimal function) cores, so now having 8 threads vs. 4 sharing one core exasperates the inability to handle heavy threads. P6 scales past NigaraII 6X on SPECint and 7X SPECfp.
The real got ya is the inability to scale and do virtualization beyond one chip. Sun's Hypervisor is very rudimentary by only assigning threads to an OS vs. priorities/mins/maxs/entitlements...etc. It takes more than one chip or a simple SPEC benchmark to get past niche.
The only time it looks like a fast chip is read only workload or if you are comparing against the end of life SPARCIV. If Sun actually cares about the environment maybe they should stop forcing customers to buy 32GB so they can get more revenue/profit.
10 % faster with 4x the # of cores
or in plain english, 1 core equals +/- 27.5% of a p6-core
that is not very impressive. Luckily, it was not meant to be .....
At least, Sun's people do not see the niagara line in the same segment as itanic, power, rock ( that's what the SUN-Andy's explained to me & some other folk present )
I think they are very right about that.
Fine chip for what it is built for, albeit perhaps a little expensive compared to an AMD / Intel chip or similar.
Hopefully Rock will put Sun back on the map for high-end computing. this chip is not meant for that.
please remember that your software is usually charged per core; quite a consideration on an 8-core chip...... (IBM will give you a discount of 70% on Niagara I per core, and there is a good reason for that....... the P6 on the other hand gets an extra charge of 20% )
"Most Powerful Chip" more accurate than "Fastest"
Niagara 2 is without a doubt the most powerful processor for many workloads. "Fast" is now up for debate. Is a 2.66 GHz Clovertown "faster" than a 3 GHz Woodcrest? In addition, I would note the numbers Sun has released are preliminary estimates, so actual numbers are likely to be higher. Also, Niagara 2 is likely to perform better on Web, Java, and Database benchmarks than CPU intensive benchmarks. So expect the relative performance difference to be much better than 10%, especially if networking or SSL performance is part of the benchmark.
Regarding IBM's claims Niagara and Niagara 2 are for "niche" workloads, I would point out Niagara and Niagara 2 are designed for Internet workloads. From these comments one can only assume it is IBM's official position the Internet is a niche market.
The "light thread" vs. "heavy thread" debate is for the most part, irrelevant. In Internet workloads, user think time and network response times induce more latency than processor performance. The one exception being processes which are inherently single-threaded, like integer linear programming, which is common in supply chain management. Maybe Niagara is not the right processor to run the SAP Central Instance, but it may be the right processor for SAP's front-end processes. And the ultimate "light thread" database workload is TPC-C, which IBM seems to promote as relevant to any customer application.
Regarding virtualization, POWERFAN seems to assume Niagara 2's virtualization capabilities are identical to Niagara 1. I doubt that is the case. The other advantage a radically multithreaded processor has in virtualization is it does not need to time-slice individual threads among multiple VMs/LPARs. IBM's POWER5 Micropartitions incur significant overhead as Micropartitions per processor increase. This is because each time the processor is switched between LPARs, it incurs a context switch, and likely a flush of the L2 cache. This is not the case when the individual threads are not time-sliced between OSs. Instead, in Niagara, the hardware handles the switching of individual threads, and thread contexts are not changed because the thread remains within an OS instance. You may not be able to weight Sun's LDOMs by percentages, but you get 64 threads per processor to weight into your LDOMs, compared to 200 percent (100% per core) with POWER5. 64 is probably enough.
Regarding, the comment about software licensing per core. This is true for platform software like RDBMS and J2EE Application Servers. It is not true for enterprise application software like ERP and CRM (which is licensed per user). So while the Oracle database is licensed per core, the Oracle Financials app riding on top is not. Also, open source web software like Apache is not licensed per core, and that is the heart of the Niagara market. And the platform vendors like IBM, Oracle, and BEA have already adjusted their licensing models for Niagara style processors.
Finally, Niagara represents where the processor market is going. POWERFAN, I suggest you look at what is planned for POWER6+ and POWER7. Rumor has it POWER6+ will be simply a four-core version of POWER6 with little increase in clock rate. POWER6 will likely never hit the 6GHz originally planned. I hear POWER7 is modeled on the Cell Processor rather than POWER6 and will be at least four cores and likely more than two threads per core. I also hear POWER7 may have a lower clock rate than POWER6. AMD's future processor, Sandtiger follows the Niagara path of more, simpler cores. Intel plans to bring back Hyperthreading in a future Xeon so eight or sixteen threads per socket may be the norm for x86 processors in two years.
Sun has shown the way, and today's IBM's POWER vs. Sun FUD is going to look hypocritical compared IBM's own System X messaging and its System P messaging. And once IBM's starts disclosing POWER7 details, IBM will simply look foolish.
T2 is a Niche and Rock is really only 4 cores (not 16)
A "major problem with Sun’s UltraSPARC T2 architecture is its poor single threaded performance. Since Sun isn’t being helpful in answering this question, I asked CPU analyst David Kanter of Real World Technologies and asked him what he thought the single thread performance of a T2 would be like. Kanter replied that it would likely be better than 1/16th of the total performance of the T2 chip. The number 1/16th was derived by looking at the number of pipelines (2 per core with 8 cores is 16 pipelines) in the UltraSPARC 2 by assuming that a single pipeline is the smallest execution engine per thread. Since a single threaded application would have a monopoly on the CPU cache, it is reasonable to postulate that single threaded performance would be better than 1/16th of the total aggregate performance of the UltraSPARC T2.
Intel’s architecture has a worst case single threaded performance of 1/8th of an 8-core server. But the fact that a single Intel core has full monopoly on the cache, memory controller, and FSB (Front Side Bus) means single threaded performance is as high as 1/5th of the total 8-core server. In fact a single core test on a 3 GHz Xeon 5160 gets a whopping score of 21 on SPECint_2006. If we hypothesize that an UltraSPARC T2 1.4 GHz processor might get 1/14th of the best 8-core score, we’re looking at a best case SPECint_2006 score of 5.6 per thread which isn’t very good at all.
Single threaded performance is so important in the server world because it essentially allows co-hosting logical or virtual servers to borrow each other’s idle time and complete tasks so much faster. Unfortunately, single-thread performance is where Sun’s UltraSPARC T2 architecture fails badly. That doesn’t mean the UltraSPARC T2 is a failure by any means because it fills its niche role in the proprietary SPARC world very well. Just don’t expect it to steal market share from Intel or AMD any time soon."