With Advanced Micro Devices admitting that it's getting ready to launch its "Istanbul" Opteron six-shooter, Intel can't afford to let AMD monopolize all the talk about x64 processors. And that is why Intel hosted a conference call today about its eight-core "Nehalem EX" processor for four-socket and larger servers, even though …
I thought AMD's Magny-Cours was supposed to do 12 cores w/ 2 6 core dies in the same package.
Either way, all this news just makes me sad that I'll probably never need anything more than a dual core in my personal machines...
MCA is the final puzzle piece
The only benefit Itanium had left over Xeon was its Pelliston cache self-healing and its Machine Check Architecture. If Nehalem-EX has both, it is game over, Itanium. Solaris has the RAS infrastructure to take advantage of MCA, and that would provide the alternative for an HP-UX environment.
Machine check architecture ?
Is that the same kind of "machine check architecture" as AMD were talking about back in 2001 with "Hammer" (and delivering soon after)?
What took Intel's AMD64 clone so long to catch up?
What, now, are the value added Itanium features (RAS or otherwise) which will make it worth Intel's while to continue investing in Itanium?
64 cores is as big as big iron gets??? Fool.
See HP Superdome, Sun M9000, (Fujitsu M9000) and a host of other servers...
The m9000 can, for example take up to 4TB of memory *now*, and 64 *sockets*... NOW.
With quad core CPUs, that's 256 cores, 512 threads - All in one OS instance (or split among many if you like.)
The Hp's are similar... (But - Who wants to run Itanium... I hear ya...)
Apples .vs. Oranges
--"Last December, Fujitsu tested a PrimeQuest Itanium box with 32 of Intel's dual-core "Montvale" Itanium 9150M processors spinning at 1.66 GHz, and it delivered 2.38 million transactions per minute (TPM) of throughput on the TPC-C online transaction processing test. IBM's eight-socket, 48-core Dunnington box, a System x3950 M2, was able to hit 1.2 million TPM."--
So in other words you're comparing a 64 core (32 socket) PrimeQuest Itanium box, the size of an entire rack that deliveres 2.38 million TPM to a 48 core (8 socket) Dunnington box in a 4U package that delivers 1.2 million TPM. I strongly suspect that I/O bandwidth between these two very different systems had a bigger part to play than the CPU performance. In any case I wonder how many 8 socket IBM System x3950's you can buy for the price of the PrimeQuest box.
What I want to see next year is a matching core for core TPM comparison between a Nehalem EX system and a Tukwilla Itanium system, together with the price for both. They must have the same amount of memory, I/O cards and disks.
Surely what matters when comparing systems is the performance for given required load, and the total cost i.e. hardware cost, software and support costs, energy consumption etc. Details like core count, amount of memory, disk etc. - who cares*? If system A performs twice as well as system B, it doesn't matter if it has 2x as much disk or half as much. Unless of course disk space is part of your performance criteria.
* ok I care, but I'm just interested, I'm not in purchasing :)
The eight socket system is a dog
Intel does not have enough QPI's on the chip to scale past 4 sockets so they are doing a hoaky interconnect thru another chip. Getting rid of the front side bus finally is good news but Intel still has no clue how to create systems past four sockets without another vedors chipset. Looking for a windjammer to solve? Anyone know if windjammer will support Nehalem or is it just for Itanium?
RE: MCA is the final puzzle piece
Oooh, a Sunshiner that actually knows his own name! Shame Magellan is just as wrong as all the AC Sunshiners.
"....The only benefit Itanium had left over Xeon was its Pelliston cache self-healing and its Machine Check Architecture....." So the ability to scale in the Superdome to 128 cores (256 when Tukzilla lands) wasn't an advantage then? Or the massively larger IO bandwidth and memory space? Then again, seeing as Slowaris is your toy of choice, you've probably never worked in enterprise computing anyway, so how would you know?
"..... If Nehalem-EX has both, it is game over, Itanium....." <Yawn> Didn't we last hear that garbage about quad-core Xeon and Opteron? I'm sure you Sunshiners were sprouting it about UltraSPARC V (care to remember what happened to that prediction?). Please, try some reading and get a clue - us customers buy the hp Integrity kit for a whole lot of tasks Xeon can't do.
"....Solaris has the RAS infrastructure to take advantage of MCA, and that would provide the alternative for an HP-UX environment." So why has Slowaris been losing marketshare so massively then? You do remember all those years of losses and shrinking Sun marketcap. Those happened beacuse customers found out hp-ux offers better RAS, performance, management, partitioning, consolidation and virtualisation features, whilst Linux like RedHat just does it all faster and a lot cheaper than Slowaris. Which is why Sun is dead. Go get some training on new OS, Sunshiner, your Slowaris gravytrain has reached the end of the line.
Re: Comparing systems
--"Details like core count, amount of memory, disk etc. - who cares*? If system A performs twice as well as system B, it doesn't matter if it has 2x as much disk or half as much. Unless of course disk space is part of your performance criteria."--
Those details do matter. It's not about disk space it's about IOPs ( I/O operations per second). A single 10K rpm SCSI disk can do (on average) 125 IOPs and Database performance is very sensitive to that. Performance scales considerably the more IOPs you can throw at it. Take a look at the specs of some of the configurations the vendors put together to achieve high TPM benchmark results. Many of them have thousands of disks. In real life of cause very few people (if any) need to build single systems as big as that, and where there is a need other solutions are available that are better suited to provide a good balance between cpu performance, disk performance and scalability.
It was not that long ago when...
It was not that long ago when MCA was Micro Channel Architecture...
The reuse of these abbreviations is pretty annoying.