Intel readying 15-core Ivy Bridge Xeon E7 processor

O

octoberasian

2[H]4U
Joined
Oct 13, 2007
Messages
4,082
http://www.anandtech.com/show/7753/intel-readying-15core-xeon-e7-v2

Information from the original source on the top end CPU is as follows:

  • 4.31 billion transistors
  • Will be in the Xeon E7 line-up, suited for 4P/8P systems (8 * 15 * 2 = 240 threads potential)
  • 2.8 GHz Turbo Frequency (though the design will scale to 3.8 GHz)
  • 150W TDP
  • 40 PCIe lanes
Click to expand...
Expected price is somewhere north of $5000, and will contain 3 blocks of 5 cores each with 37.5MB of cache total. Each of the 3 blocks have their own clock domains and is backwards compatible to existing Socket 2011 boards that can use Xeon processors.
 
Last edited:
Interesting... I can't imagine there will be a great deal of demand for this CPU, that is, over the current E7 offerings.

But, who knows.
 
Last edited:
I wish the top model was unlocked....I'd give my left nut if it was. You'd think that >$5000/CPU would buy you that ability...:rolleyes:

From what I've read, the top model will be $7-7.5k...

I can understand why Intel would want to avoid the Westmere days of cheap low bin Xeons being overclocked to high heaven in single CPU boards (and the SR-2). But the top model of each Xeon line should be fully unlocked...E3, E5 and E7.
 
lutjens said:
I can understand why Intel would want to avoid the Westmere days of cheap low bin Xeons being overclocked to high heaven in single CPU boards (and the SR-2). But the top model of each Xeon line should be fully unlocked...E3, E5 and E7.
Click to expand...

The yields may be low enough that they don't want to tempt enthusiasts with them. I'm guessing that locked chips have a lower return rate, too.
 
We're already planning upgrading on of our 4P hosts with these. 1TB of RAM and 120 vCPU's, YES PLEASE!
 
Scalable to 3.8 GHz...schwing!

Speculating here, but: These new E7's could very well be the final stepping stones to near-future 3.5+ GHz 8C/16T consumer i7 parts.
 
DeathFromBelow said:
The yields may be low enough that they don't want to tempt enthusiasts with them. I'm guessing that locked chips have a lower return rate, too.
Click to expand...

Overclocking voids your warranty...if a chip was overclocked, died and was returned, you can bet that Intel'll know that you killed it. Most of the time, though, Intel doesn't care, as the dead chip was a reject/low margin chip to begin with. To the best of my knowledge, all high end Xeons that are returned due to failure are fully dissected to determine why. Damage due to overclocking/overvolting will be a very clear case to deny warranty on a processor.

Also, the total number of these chips that would end up in systems that permit overclocking would likely be <1000 total, with 95% of those being sold by a reputable OEM builder along the same lines as Appro, that would overclock the CPU safely, validate it and warranty it themselves, ensuring that the increased thermal requirements are met with a comfortable margin of safety.

Most enthusiasts who would be able to buy one for their own use would be unlikely to do anything to risk the hurting the chip (considering its very high price) and the ones that do end up dying due to abuse should be very few in number. Again, those that are killed due to abuse are inherently ineligible for replacement under warranty in any regard, so those claims (if made) would be summarily denied.

In addition, with the DP (or MP for that matter) Xeon as it exists today (fully locked), there is absolutely nothing that would prevent someone from throwing huge voltage at it in a single CPU motherboard and killing it. Throwing 1.75V into a Sandy-EP chip will kill it dead, whether it's locked or not. It's the prolonged and excessive voltage that usually kills chips, not a safe gentle overclock with the cooling to support it to use what margin exists in a CPU.
 
DejaWiz said:
Scalable to 3.8 GHz...schwing!

Speculating here, but: These new E7's could very well be the final stepping stones to near-future 3.5+ GHz 8C/16T consumer i7 parts.
Click to expand...

Scalable to 3.8GHz? Maybe the ultra-neutered edition with 4 cores. I doubt very much that the fully-fledged version will have that kind of turbo speed, and if it somehow does, will be the 1 Core Turbo that we never actually see. The "all core" Turbo, if it exists at all, will likely be a maximum of two bins (if we're very lucky)..
 
Interesting product, but I'd wager that if you're at the point where you're able to parallelize across 15 threads (or is it 30 on this guy?), you're probably better-served in Xeon Phi territory.

Dat Knight's Landing can't be far off...
 
DejaWiz said:
Scalable to 3.8 GHz...schwing!

Speculating here, but: These new E7's could very well be the final stepping stones to near-future 3.5+ GHz 8C/16T consumer i7 parts.
Click to expand...

Isn't haswell E supposed to bring 8 cores extreme chips ?
 
lutjens said:
Scalable to 3.8GHz? Maybe the ultra-neutered edition with 4 cores. I doubt very much that the fully-fledged version will have that kind of turbo speed, and if it somehow does, will be the 1 Core Turbo that we never actually see. The "all core" Turbo, if it exists at all, will likely be a maximum of two bins (if we're very lucky)..
Click to expand...

I'm not so sure about a lower-binned model with only 4 cores being scalable to 3.8 GHz... 8 and maybe up to 10 cores, I'd guess. Especially with that level of TDP.


Michaelius said:
Isn't haswell E supposed to bring 8 cores extreme chips ?
Click to expand...

Yes, very likely (I will be shocked if there isn't an 8C HWL-E). Sorry, I should have specified "mainstream" 8C/16T i7 instead of "consumer", since the Extreme Editions are technically consumer segment parts, as well.
 
It would be interesting if all 15 cores can hit 3.8 GHz at Turbo with that high of a TDP. I doubt it, but we can't expect miracles here on a server chip.
 
Some more info has surfaced...$6841 for the E7-8890 V2.

http://en.wikipedia.org/wiki/List_of_Intel_Xeon_microprocessors#Xeon_E7-48xx_v2_.28quad-processor.29

A turbo of 3.4GHz is quoted...I can't imagine that it's the all core turbo number, though, considering the E5-2687W V2 has half as many cores, nearly the same TDP and has an all core turbo multiplier of 36x. Considering the intended market, it's easy to assume that Intel will be conservative with the E7.

I found a benchmark sheet from Huawei, stating a Turbo speed of 3.2GHz (no way of knowing how many cores this speed is for though).

http://enterprise.huawei.com/ilink/cnenterprise/download/HW_326110

Between the two of them, I'd guess a maximum one core Turbo of 3.4GHz and maybe (if we're VERY lucky) an all core turbo of 3.2GHz. I suppose it's possible Huawei had an older chip with a lower Turbo. Tough to say for sure. Either we have to wait for a more detailed leak, or wait for launch, which shouldn't be too much further off.
 
As most have heard by now, the launch was today...

The Bad News: Not compatible with the existing LGA2011 infrastructure, as the existing socket is 2011-0 and the new one is 2011-1. The packages are the same size, but very different land pattern, and the socket seems rotated 90 degrees in its narrow ILM vs the old 2011-0. Also, only a total of 32 PCI-E lanes per processor (which probably won't be a huge issue in 4P and 8P configs).

The Good News: A new, previously unknown performance model, the E7-8895V2 with a max turbo of 3.6GHz was part of the announcement. Still no word what the all core turbo on any CPU is.
 
lutjens said:
As most have heard by now, the launch was today...

The Bad News: Not compatible with the existing LGA2011 infrastructure, as the existing socket is 2011-0 and the new one is 2011-1. The packages are the same size, but very different land pattern, and the socket seems rotated 90 degrees in its narrow ILM vs the old 2011-0. Also, only a total of 32 PCI-E lanes per processor (which probably won't be a huge issue in 4P and 8P configs).

The Good News: A new, previously unknown performance model, the E7-8895V2 with a max turbo of 3.6GHz was part of the announcement. Still no word what the all core turbo on any CPU is.
Click to expand...
So, Anandtech was wrong on the compatibility and there is actually a FORK (!!!) in the SB/IVB socket 2011 now?

That's probably going to anger someone with a Xeon Socket 2011 board hoping they could pop these in and hope to have 15 core/30 threads of action for their server.

My only reaction to Intel is this: /headdesk. /headdesk.
 
I don't think there will be an all cores turbo, if you are using these CPUs then they will be at full load 100% of the time, so the base clock is likely as fast as they can go on all cores.
 
octoberasian said:
So, Anandtech was wrong on the compatibility and there is actually a FORK (!!!) in the SB/IVB socket 2011 now?

That's probably going to anger someone with a Xeon Socket 2011 board hoping they could pop these in and hope to have 15 core/30 threads of action for their server.

My only reaction to Intel is this: /headdesk. /headdesk.
Click to expand...

They technically weren't wrong, as IIRC they stated the Xeon E7 is LGA2011, which it is. They never directly stated backwards compatibility with the existing LGA2011-0 sockets, although there is a subtle implication, given the lack of a differentiator.

I can understand why Intel did modify the socket, likely so that Haswell-EX and Broadwell-EX will work with it as well, as the Brickland platform is supposed to support them all. They probably could have used the existing LGA2011-0 socket for Ivy-EX, but it may not have worked for the later chips. The new systems based on the E7's Socket R1 all seem to use memory cartridges that have the Jordan Creek buffer and DDR3 DIMM sockets on them. This seems very intuitive, as upgrading to a DDR4-based CPU in the future should therefore be a simple matter of exchanging the cartridges for a new DDR4 version that has DDR4 DIMM sockets and compatible memory buffer and replacing the CPU itself.

The fact that the E5-2697 V2 exists is proof that this die can be used with the LGA2011-0 socket. What would be nice to see (although we are highly unlikely to) would be one top-end 15 core E5 Xeon SKU, that shares the high 155W TDP. Having it unlocked would be a nice touch...:D. Intel should have done this rather than waste time with the E7-2800 V2, which will likely see poor sales due to the fact that most people who are looking at an E7 Xeon are looking for a 4P and greater CPU count. I've haven't seen a 2P E7 motherboard yet. The E7-2800 V2 will be likely be a CPU with a VERY limited selection of motherboards, which is a shame, as it would probably see notable demand as an LGA2011-0 compatible chip. It wouldn't surprise me to see the E7-2800 V2 used only in a handful of premade workstations and to never see a DIY 2P Xeon E7 motherboard ever made.
 
I took a closer look at the data sheets to get my head around some of the differences between the LGA2011-0 and LGA2011-1. LGA2011-1 appears to be a more "fleshed out" version of LGA2011-0, with the voids present in LGA2011-0 filled in with pins. Looking at the socket size and dimensions, they are identical. It appears that the two processors will physically fit in each other's sockets as the lockout slots are the same between the two of them. Land size is the same as well.

This is where whatever hope I had for backwards compatibility began to be erased.

Looking at the actual Land Number listing in the Datasheet for both the E5-2600 V2 and the E7 V2, the pinouts are VERY different, with almost no lands being the same between the two of them.

It also appears that Intel was quite lazy when they created the Thermal/Mechanical sheet for the LGA2011-1 socket, with many of the pictures and references clearly showing the LGA2011-0 socket. The two sockets are clearly different as shown in the Datasheet drawings. Given the stark difference in pinout, it's likely that attempting to use one CPU in the other's socket would result in a very bad (and very expensive) day.

I do find it strange, however, that Intel would allow a situation to exist where a CPU would physically fit in an incompatible socket. Usually the lockouts on a new socket are changed to prevent insertion of an incompatible CPU.
 
lutjens said:
I took a closer look at the data sheets to get my head around some of the differences between the LGA2011-0 and LGA2011-1. LGA2011-1 appears to be a more "fleshed out" version of LGA2011-0, with the voids present in LGA2011-0 filled in with pins. Looking at the socket size and dimensions, they are identical. It appears that the two processors will physically fit in each other's sockets as the lockout slots are the same between the two of them. Land size is the same as well.

This is where whatever hope I had for backwards compatibility began to be erased.

Looking at the actual Land Number listing in the Datasheet for both the E5-2600 V2 and the E7 V2, the pinouts are VERY different, with almost no lands being the same between the two of them.

It also appears that Intel was quite lazy when they created the Thermal/Mechanical sheet for the LGA2011-1 socket, with many of the pictures and references clearly showing the LGA2011-0 socket. The two sockets are clearly different as shown in the Datasheet drawings. Given the stark difference in pinout, it's likely that attempting to use one CPU in the other's socket would result in a very bad (and very expensive) day.

I do find it strange, however, that Intel would allow a situation to exist where a CPU would physically fit in an incompatible socket. Usually the lockouts on a new socket are changed to prevent insertion of an incompatible CPU.
Click to expand...
So, long story short: The new Xeon E7 processor is literally a 15-core IVB-EX CPU that fits into a Haswell-EX server socket 2011, but without DDR4 support.

That's... rather interesting.

Man, I feel bad to the IT folks who see this and want to upgrade their servers to it only to find out that they need an all new Xeon Socket 2011 board. :(

That'll be a very expensive decision for a CTO or someone in charge of server hardware at a company. It honestly makes me wonder who are Intel marketing this processor to given it'll be a very low volume market and possibly limited production.
 
octoberasian said:
So, long story short: The new Xeon E7 processor is literally a 15-core IVB-EX CPU that fits into a Haswell-EX server socket 2011, but without DDR4 support.

That's... rather interesting.

Man, I feel bad to the IT folks who see this and want to upgrade their servers to it only to find out that they need an all new Xeon Socket 2011 board. :(

That'll be a very expensive decision for a CTO or someone in charge of server hardware at a company. It honestly makes me wonder who are Intel marketing this processor to given it'll be a very low volume market and possibly limited production.
Click to expand...

The 4P and 8P configurations are definitely a step up from the old Xeon E7's and they make sense in many applications. The 8P configurations in particular should sell quite well, and considering the lengthy upgrade path laid out for the Brickland platform, it's a good business decision in many cases. Once Haswell-EX comes out, it should be a simple matter of exchanging the DDR3 memory cartridges for DDR4 versions and replacing the CPUs. Broadwell-EX should be even easier, with merely a CPU swap being needed.

It's the E7-2800 V2 configurations that are totally pointless, unless the market decides to roll out a large quantity of 2P Socket R1 motherboards. This in itself is incredibly unlikely, as the existing LGA2011-0 boards are well established, much cheaper, and offer only modestly inferior performance with dual E5-2697V2s that together will cost less than one E7-2890V2. Also, the more advanced Haswell-EP setups that will hopefully show up later this year will erase almost every reason why anyone would consider a 2P E7 V2 system. Haswell-EP will offer a newer chipset, and a newer architecture with almost the same number of cores, not to mention far more widespread adoption at a much more reasonable price.
 
lutjens said:
The 4P and 8P configurations are definitely a step up from the old Xeon E7's and they make sense in many applications. The 8P configurations in particular should sell quite well, and considering the lengthy upgrade path laid out for the Brickland platform, it's a good business decision in many cases. Once Haswell-EX comes out, it should be a simple matter of exchanging the DDR3 memory cartridges for DDR4 versions and replacing the CPUs. Broadwell-EX should be even easier, with merely a CPU swap being needed.

It's the E7-2800 V2 configurations that are totally pointless, unless the market decides to roll out a large quantity of 2P Socket R1 motherboards. This in itself is incredibly unlikely, as the existing LGA2011-0 boards are well established, much cheaper, and offer only modestly inferior performance with dual E5-2697V2s that together will cost less than one E7-2890V2. Also, the more advanced Haswell-EP setups that will hopefully show up later this year will erase almost every reason why anyone would consider a 2P E7 V2 system. Haswell-EP will offer a newer chipset, and a newer architecture with almost the same number of cores, not to mention far more widespread adoption at a much more reasonable price.
Click to expand...
I'm holding out on a major computer upgrade until Haswell-E comes out this year. I want to see how that processor is compared to the older SB-E/IVB-E generation. That and the fact it'll have a newer chipset compared to the limited X79 Express chipset hopefully with more SATA and USB 3.0 connections.

Yeah, it is curious why Intel released a 15-core behemoth and charge $6000 for it when Haswell-EP/EN/EX server processors are coming out this year. I'm still questioning the reasoning behind it. That and the fact it's using a differently configured Socket 2011 board means it won't find a high adoption rate among enterprises and datacenters, or maybe educational and research institutions if budgets allow. However, if they're smart enough money-wise, they'd hold out for Haswell server processors instead.
 
octoberasian said:
I'm holding out on a major computer upgrade until Haswell-E comes out this year. I want to see how that processor is compared to the older SB-E/IVB-E generation. That and the fact it'll have a newer chipset compared to the limited X79 Express chipset hopefully with more SATA and USB 3.0 connections.

Yeah, it is curious why Intel released a 15-core behemoth and charge $6000 for it when Haswell-EP/EN/EX server processors are coming out this year. I'm still questioning the reasoning behind it. That and the fact it's using a differently configured Socket 2011 board means it won't find a high adoption rate among enterprises and datacenters, or maybe educational and research institutions if budgets allow. However, if they're smart enough money-wise, they'd hold out for Haswell server processors instead.
Click to expand...

Normally, I'd agree and say waiting for Haswell-EX is the way to go. However, many businesses who can take advantage of these machines are desperate for new hardware, and are still running the first generations of E7. I'd imagine many will buy Ivy-EX now and skip Haswell-EX instead of waiting, simply opting to wait until Broadwell-EX. The benefits to upgrading now are too huge to wait for.
 
lutjens said:
Normally, I'd agree and say waiting for Haswell-EX is the way to go. However, many businesses who can take advantage of these machines are desperate for new hardware, and are still running the first generations of E7. I'd imagine many will buy Ivy-EX now and skip Haswell-EX instead of waiting, simply opting to wait until Broadwell-EX. The benefits to upgrading now are too huge to wait for.
Click to expand...
You make a very good point.
 
450 said:
Good point. 15 core Mac Pro it is!
Click to expand...

Still don't think so, they would have to change a lot to get this working in the cylinder. They are already at the limits of the thermal cooling capabilities plus the extra power usage.
 
You must log in or register to reply here.
Back
Top