Intel Releases Itanium 9500, Packing Up to 8 Cores
Intel released the 32 nm generation of its Itanium processor series, now known as the 9500-series.
The new product family, formerly code-named Poulson, succeeds the 65 nm 9300 series, code-named Tukwila, which was launched in February 2008.
Poulson represents a major shift in Intel's Itanium strategy. It is still the company's mainframe processor, but its focus is changing from legacy mainframe installations to a cloud infrastructure, both for private clouds as as well as cloud environments that are operated to enable a service model. Intel is also aligning the new Itanium much closer to its x86 architecture, making the 9500-series compatible with the Xeon E7 processor's 7500-series chipset. Intel noted that the new Itanium also shares the Xeon's memory buffer, interconnect and "industry standard" memory to make the Itanium more cost-efficient to produce.
Compared to its predecessor, the Itanium 9500 delivers up to 2.4x improved performance and 33 percent greater bandwidth. the 3.1 billion transistor chip integrates up to 54 MB on-die memory, and supports up to 2 TB of low voltage DIMMs in a four-socket configuration. The power band is also slightly improved. Whereas the 9300 series ran from 1.6 GHz to 1.73 GHz at 130 watts to 185 watts, the 9500 series is clocked from 1.73 GHz to 2.53 GHz and rated at a maximum power consumption of 130 watts to 170 watts.
At launch, Intel is offered four versions of the 32 nm Itanium. The 9520 (quad-core, 1.73 GHz), 9540 (8-core, 2.13 GHz), 9550 (quad-core, 2.4 GHz), and 9560 (8-core, 2.53 GHz). Tray-pricing starts at $1,350 for the 9520 and tops out at $4,650 for the 9560.
I think it means 4 cpu sockets (equals 4 memory controllers)... and many many dimms
http://arstechnica.com/business/2011/06/ask-ars-why-itaniumask-ars-with-xeons-improvement-why-bother-with-itanium/
http://arstechnica.com/business/2011/06/ask-ars-why-itaniumask-ars-with-xeons-improvement-why-bother-with-itanium/
Jesus, 512GB DIMMs? Or am I misinterpretting that?
I think it means 4 cpu sockets (equals 4 memory controllers)... and many many dimms
That makes more sense. (And here I had gotten my hopes up
You can already buy 32GB on a single DIMM, it won't be long before we get to 512.
"Now, this $4 billion number is a lot smaller than $30 billion, which is the size of Intel's Xeon business. But it's a lot larger than $1.6 billion, which was the revenue for all of AMD combined (CPUs, GPUs—the whole company) in the first quarter of 2011."
oops, a year is not a quarter. Ars fail. full stop.
And that "lot of DIMMs" would be buffered 32 chips quad-bank (think two dual-sided DIMMs smashed into one) DIMMs... four of them per channel, four channels per CPU, 64 DIMMs total.
Maxed out, such a monster must be an interesting thing to behold at least for the first few times around.
At least they moved it to a common socket so maybe in the long run Itanium might gain market share.
They do. It would look something like this:
http://www.brightsideofnews.com/news/2009/5/27/intel-nehalem-ex-xeon-spells-the-voice-of-doom-for-itanium.aspx
This is a quad Nehalem system from back in 2009, not Itanium, but the Itanium setup would look similar with memory riser cards.
Not really that big of a deal. IBM's POWER7+ processors hit over 300 watts
There are not, and will never be, consumer versions of the Itanium processors. There are 8 core Sandybridge processors, but they are pricey. Itanium is Intel's answer to IBM's extremely reliable POWER based enterprise servers and mainframes. The only reliability features present in x86 processors is support for ECC memory, a very tiny subset of the features present in top POWER7+ processors and Itanium processors. Enterprise grade systems do allow for CPUs to be hot-swapped but there is no failure protection on x86 chips, all workloads on that CPU will fail. Top end POWER7+ chips such as those found in the IBM zEnterprise mainframes allow for workloads to fail over from one processor to another, not something that you need in a desktop.
Intel should either go 22nm with this, or wait a year and go straight to 14nm,
Does anyone know why server users did not choose it? It wasn't because they were not available, although pretty soon they will not be. Too much buck for the bang? Bad architecture? Some other shortcoming?
If AMD hadn't
I'm not totally sure about this, but I remember reading somewhere that Itanium is designed in a way to handle server-type data workloads (maybe super-highly-threaded workloads or huge chunks of data, I don't really know), unlike x86. So even if mainstream software were made to work with Itanium, they wouldn't run as good as they do with x86. Is this false?