Sign in with
Sign up | Sign in
Your question

Nehalem Comes to Xeon 5500 Series Server and Workstation Processors

Last response: in CPUs
Share
March 21, 2009 12:34:50 PM

What do you think about the Xeon line, which is finding its way not only into servers, but also into workstations such as Apple’s Mac Pro.

On Thursday April 2nd from 10am to 12 noon PDT we will be presenting a special Chat on Intel’s Xeon 5500 Series Server and Workstation Processors. Due to popular demand, a bonus chat has been added on Friday April 3rd from 10am to 12 noon PDT! Join both the Chats at http://communities.intel.com/community/server
March 24, 2009 5:35:12 AM

All of the chips will be drop-in compatible with Intel's current dual-processor platform, the company INTEL said.
The processors are built on a 45nm process, using hafnium gate transistors and a halogen-free design.
March 24, 2009 7:19:34 PM

112inky said:
All of the chips will be drop-in compatible with Intel's current dual-processor platform, the company INTEL said.


you mean, Socket 771, or a dual-proc version of socket 1366 ?

Bring it on !

i have a Q9550 & if i ever upgrade again, i will want MORE SPEED !!!

Related resources
March 25, 2009 3:41:28 AM

hmmm.. sure you will have.... :D  :D  :D 
March 25, 2009 4:46:32 PM

Despite the lack of any public information available from Intel, with enough Googling the character of the Xeon 5500 series is out there. Digging into the specs on Wiki, and comparing the 'roadmap' offerings of the Nehalem line, my conclusion is Intel has some shrewed marketing going on that most consumers will not catch. If you have looked at the i7 variants, you know the difference comes down to whether the Multipliers are locked or not. It's the same 'engine' called different things depending on how much you pay to have all its features enabled. If you are an Enterprise customer your Nehalem chip will be called Xeon W55xx instead of i7-xxx. According to the specs if you turn up an i7-965 you get a Xeon W5580, except... you can't use an i7-965 in a multi-processor board because the second QPI on the Nehalem "i7" is not enabled. You can have an i7-965 now for $1000. and the posted release price of the Xeon W5580 is listed at $1600. So, if you would like to build a multi-processor rig using the latest Tylersburg capability, you get to pay a $1200. premium for the pleasure of buying two chips. Fair?
March 25, 2009 6:44:57 PM

Skywriter7 said:
You can have an i7-965 now for $1000. and the posted release price of the Xeon W5580 is listed at $1600. So, if you would like to build a multi-processor rig using the latest Tylersburg capability, you get to pay a $1200. premium for the pleasure of buying two chips. Fair?


i don't know if it's "fair", that's kind of a philosophical question.

it helps pay for Intel R&D, among other things.

c'mon, AMD, let's see the 2P Opteron version of the Phenom 2 940 !!!

and, there's always 2P versions of Harpertown, for example
dual 771 Supermicro motherboard
http://www.newegg.com/Product/Product.aspx?Item=N82E168...

Xeon 5420 @ 2.5 GHz for $419
http://www.newegg.com/Product/Product.aspx?Item=N82E168...
a b à CPUs
March 26, 2009 9:43:41 PM

Actually, the architecture is different. The i7-965 has a single QPI link, while the Xeon W5580 has a pair of QPI links. This is needed for the dual processor hookup. Therefore, the fact that the Xeon is more expensive makes perfect sense.
March 27, 2009 2:38:51 PM

What you are implying is that Intel produces a new master die for each version of a chip to be released. Specificly, that they started with a full Nehalem chip and then deleted the second QPI to make an i7, or started with one QPI and went back and added another to make the W5500. My understanding is that there is only one Nehalem 'engine', it starts with all the capability, and is dumbed down ie disabled proportionately depending on the market target it is aimed for. Please, if you have engineering references that show different, let's see them. My inquiry to Intel has yet to be answered. But, let's assume you are right. It costs $600.00 to add a second QPI? More fundamentally, when was the last time you heard of buying more of a product (microchips) costs you more? Xeon chips should actually be cheaper to encourage multiprocessor machines so they could sell more chips!
March 27, 2009 5:53:31 PM

I had heard the release date for these chips is March 29th...is that not correct?

I've seen the bulk price of the E5520 listed at $373, what does that mean for the actual price off store shelves?

Also, have there been any price quotes for the applicable dual 1366 motherboards from Asus and Supermicro?

I'm looking to build a workhorse scientific computer in the next couple of week, and I'm waiting for this new line to come out. Any help/info would be greatly appreciated

~Lyuokdea
March 28, 2009 9:26:13 PM

This is what the architecture was designed for, guys. Lots of memory bandwidth, twin QPI links on the 5500-series CPUs, Hyper-Threading, Turbo Boost, etc...

I'll be back to discuss further once the NDA breaks at the end of the month!
March 29, 2009 4:10:00 PM

Skywriter7 said:
What you are implying is that Intel produces a new master die for each version of a chip to be released. Specificly, that they started with a full Nehalem chip and then deleted the second QPI to make an i7, or started with one QPI and went back and added another to make the W5500. My understanding is that there is only one Nehalem 'engine', it starts with all the capability, and is dumbed down ie disabled proportionately depending on the market target it is aimed for. Please, if you have engineering references that show different, let's see them. My inquiry to Intel has yet to be answered. But, let's assume you are right. It costs $600.00 to add a second QPI? More fundamentally, when was the last time you heard of buying more of a product (microchips) costs you more? Xeon chips should actually be cheaper to encourage multiprocessor machines so they could sell more chips!


It doesn't cost $600 in silicon real estate to add a second QPI interface, of course. But it has significant costs in R&D to get them working in parallel, to characterize them, and to get them to yield appreciably. This is likely not the case with i7, but what if, in order to get good yields on 2 QPIs per chip, the fabrication process has to be skewed so far in one direction that it yields poorly (though each chip that IS good delivers on the QPIs)? Since you're throwing away more of the wafer, the per-unit cost goes up.

Any company will seek to recoup this investment by charging a higher price for line items which take advantage of features which are harder to develop or yield.
a c 100 à CPUs
March 30, 2009 9:31:18 PM

Skywriter7 said:
What you are implying is that Intel produces a new master die for each version of a chip to be released. Specificly, that they started with a full Nehalem chip and then deleted the second QPI to make an i7, or started with one QPI and went back and added another to make the W5500. My understanding is that there is only one Nehalem 'engine', it starts with all the capability, and is dumbed down ie disabled proportionately depending on the market target it is aimed for.


Yes, that is almost certainly what's going on. The fact that Intel is packaging them for the exact same socket is pretty much a dead giveaway that they're the same silicon. AMD did exactly the same thing with the Opteron roll-out in 2003, where the 1xx, 2xx, and 8xx chips all used the same die and same socket, but the number of enabled HT links varied. AMD still uses the same die for all of their current Opterons, except now they put the single-socket ones in a different socket than the multi-socket ones. It just makes sense as making different masks and a different fab line for the relatively low-volume DP and MP parts is very expensive rather than just using one common die for those and the higher-volume UP parts.

Quote:
Please, if you have engineering references that show different, let's see them. My inquiry to Intel has yet to be answered. But, let's assume you are right. It costs $600.00 to add a second QPI? More fundamentally, when was the last time you heard of buying more of a product (microchips) costs you more? Xeon chips should actually be cheaper to encourage multiprocessor machines so they could sell more chips!


I don't think we're ever going to see multi-socket-capable CPUs ever be even close to the same price as equivalent single-socket chips ever again. 10-15 years ago, normal desktop chips were multi-socket capable- you could just go buy two normal Pentiums, PIIs, Mendocino Celerons, or PIIIs and stick them in a dual-socket/slot board and they'd work. Technically you could do the same with Athlon XP-Ms, but it was an unofficial hack. After those days, Intel got the great idea to hose people who wanted SMP by changing the new Pentium 4 Willamette-based Xeons to socket 603 and preventing normal P4s from working in SMP. Otherwise, you could run P4s in SMP if the CPUs allowed it as the desktop chipsets up through the 875 supported SMP. Intel took further steps to hose people who wanted SMP setups by intentionally crippling the 915 and 925 chipsets to not support SMP operation because mobo makers dared to use less-expensive 865 and 875 chipsets to drive dual socket 603 Xeons rather than the officially-blessed E73xx/E75xx units. A noted side effect of this is that no 915 or 925-based system supports dual-core CPUs as the MCM Pentium Ds act exactly like two Xeons in two different sockets to the chipset. About the only situation where the vendor could even make much of an argument for needing to charge higher prices for multi-socket-capable parts than single-socket-only parts is AMD after it migrated all DP and MP parts to Socket F and UP parts were AM2/AM2+.

archibael said:
It doesn't cost $600 in silicon real estate to add a second QPI interface, of course. But it has significant costs in R&D to get them working in parallel, to characterize them, and to get them to yield appreciably. This is likely not the case with i7, but what if, in order to get good yields on 2 QPIs per chip, the fabrication process has to be skewed so far in one direction that it yields poorly (though each chip that IS good delivers on the QPIs)? Since you're throwing away more of the wafer, the per-unit cost goes up.


QPI, like AMD's HyperTransport, is a feature that takes up only a small amount of die space but greatly changes the functionality of the chip. I'll be willing to bet that very few HT/QPI links are disabled because they are actually defective- they are disabled because AMD and Intel would rather customers pay $2000 for a 4/8-socket-capable or $500 for a dual-socket-capable version of a $200 single-socket chip rather than just $200.

Quote:
Any company will seek to recoup this investment by charging a higher price for line items which take advantage of features which are harder to develop or yield.


I think it's more "any company will seek to recoup this investment by charging a higher price for line items which take advantage of features which the customer is willing to pay more for." The general public does not buy multi-socket systems, nor would paying $800 for a bottom-of-the-line CPU sit well with them. But an IT guy working for a company doesn't flinch at paying that much for a 4-way-capable unit as he absolutely needs a four-way server and it's not his own money. If the general public started clamoring for four-socket systems, you can bet that prices of those CPUs would come down a *lot* due to competition, and probably be not much higher than single-socket-only CPUs.
April 1, 2009 7:35:56 PM

this is the motherboard i want

http://www.newegg.com/Product/Product.aspx?Item=N82E168...

"ASUS Z8PE-D12X(ASMB4-IKVM) Dual LGA 1366 Intel 5520 SSI EEB 3.61 Server Motherboard", about $500.

it has PCI-X slots, and i have an Adaptec U-160 PCI-X controller sitting around doing nothing.

right now the Nehalem Xeon's are a little pricey. no hurry !

April 1, 2009 7:40:41 PM

What's the difference between the Z8PE-D12 and the Z8PE-D12X?

~Lyuokdea
April 5, 2009 7:54:17 AM

This topic has been desticky in top of the forum by Justinblue
a c 127 à CPUs
April 6, 2009 2:21:33 AM

Raviolissimo said:
http://www.newegg.com/Product/Product.aspx?Item=N82E168...

the D12 is about $50 cheaper & doesn't have the PCI-X connectors.

i'm not sure if PCI-X is a "legacy" part at this point or not.

http://www.adaptec.com/en-US/products/Controllers/Hardw...

looking at that low-profile 320 speed SCSI controller, which uses PCI-Express 1 lane (the small PCI-E connector), it looks like PCI-X is past its prime.


From what I remember PCI-X was supposed to be the next step above PCI. But it never really made its mark in the DT market but did make a mark in the server market, very small mind you.

Then cane along PCI-Express which was cheaper and able to do more bandwidth. So in the end it won out.
!