15 core Ivy EX?

[GermanyChris]

Computers don't really excite me like they used to but there really is a 15 core Ivy EX just imagine what we'll see in 2 socket MP's and 4 socket work stations. Just imagine 4 of these with 4 or 6 xeon Phi's you'd have a super computer in a tower under your desk

http://www.sweclockers.com/nyhet/16459-intel-databas-avslojar-ivy-bridge-e-med-upp-till-15-karnor

 

[Umbongo]

Interesting that they aren't making a separate socket for Ivy Bridge EX. End of the line for 8-way servers perhaps with AMD having already stopped making them.

This seems like a list for multi-processor servers so not aimed at the dual-processor workstation market. Which means low clockspeed and high price.

 

[deconstruct60]

Interesting that they aren't making a separate socket for Ivy Bridge EX. End of the line for 8-way servers perhaps with AMD having already stopped making them.

They are making a separate socket. That article talks about the -EP models and sockets and then transitions to -EX where it says nothing about sockets at all.

If anything the -EX is going to trend toward being socket compatible with the Itaniums. That's long, long, long overdue. The next iteration of the E7 chipset is suppose to be compatible with the next iteration of Itanium. Intel is working toward a common upper end Xeon and Itanium socket. That way if the end the Itanium product updates folks with big iron can go forward with just a software adjustment and transition to an E7 alternative.

Socket 2011 is limited to 4 socket set ups and only with E5 4000 series offerings. Those aren't workstation class set-ups nor very high end server set ups either. ( too many hops between the corners of the mesh since limited to two QPI links. ) It is an "in between" offering similar to how the E5 2400 (with fewer QPI and PCI lanes) is a reduced offering between what can do with an E3 or E5 2600 solution.

 

[Giuly]

Yes, I know... But some Mac software would benefit from 32 cores at 2.6GHz (or in that case 2.3GHz).

 

[Umbongo]

They are making a separate socket. That article talks about the -EP models and sockets and then transitions to -EX where it says nothing about sockets at all.

The package number assigned to these parts (5579-1.2) only lists LGA 1155, LGA 2011, LGA 1356. All EX parts listed are 15-core and all EP are 6 or 10. To me this says all these parts are for the same socket,

http://qdms.intel.com/cn/wJrtegsgkEKYoToFFtopPA==

I suppose they could be using EP for 6 and 10 core parts whether they are for LGA 2011 or a new socket, with 15-core EX parts only for a new socket. That still doesn't explain the lacking socket number and no mention of it yet in any form when new sockets are usually revealed well in advance.

If anything the -EX is going to trend toward being socket compatible with the Itaniums. That's long, long, long overdue. The next iteration of the E7 chipset is suppose to be compatible with the next iteration of Itanium. Intel is working toward a common upper end Xeon and Itanium socket. That way if the end the Itanium product updates folks with big iron can go forward with just a software adjustment and transition to an E7 alternative.

Right, which seems some time out yet and would mean these EX processors aren't for it and Intel are bringing the EX moniker to the E5 series. I'd forgotten they were converging these two, so of course this is the future of 8-way systems.

 

[deconstruct60]

The package number assigned to these parts (5579-1.2) only lists LGA 1155, LGA 2011, LGA 1356.

that package number has 3 sockets associated with it so .....

All EX parts listed are 15-core and all EP are 6 or 10. To me this says all these parts are for the same socket,

So it has to be a single socket. That is a pretty big leap.

It is likely there will be 4 core -EP products ( in the E5 1600 series ). So no I don't think this is even an exhaustive list. In the previous E7 line up there was a 8000 , 4000 , and 2000 set of offerings.

http://ark.intel.com/products/family/59139/Intel-Xeon-Processor-E7-Family/server

[ the 1000 series definitely are a waste of time though at this point. Likely the 2000 series also. ]


It also is a huge leap to conclude that E7 would exist entire of just 15 core configurations.

I suppose they could be using EP for 6 and 10 core parts whether they are for LGA 2011 or a new socket, with 15-core EX parts only for a new socket. That still doesn't explain the lacking socket number and no mention of it yet in any form when new sockets are usually revealed well in advance.

Frankly the E7 probably doesn't need a new socket. Either moving to a merge with the Itanium or staying still with Westmere-EX limitations. Remember -EX appeared after the more mainstream Westmere models did. There are 512GB per socket and 32GB per DIMMs support present already:
http://www.anandtech.com/show/4259/westmereex-intels-flagship-improves

Lack of "news" would be more indicative that it was staying the same more so than merging with the new E5 socket. Unless they are gimping the QPI link count across the entire line up it doesn't make sense. At best perhaps the E7 4000 slotting into something a E5 4000 would slot into.

 

[PowerPCMacMan]

I doubt

We will see these in the 2013 Mac Pro.. Perhaps in the iMac or MacBook Pro since more tender loving care is given to those. No, nothing to see here.. moving right along!

 

[Umbongo]

that package number has 3 sockets associated with it so .....

So it has to be a single socket. That is a pretty big leap.

It is likely there will be 4 core -EP products ( in the E5 1600 series ). So no I don't think this is even an exhaustive list. In the previous E7 line up there was a 8000 , 4000 , and 2000 set of offerings.

http://ark.intel.com/products/family/59139/Intel-Xeon-Processor-E7-Family/server

It also is a huge leap to conclude that E7 would exist entire of just 15 core configurations.

edit: I must have missed this article. These are E7-4800V2 for socket LGA 2011. Confusion cleared up.

http://news.softpedia.com/news/Hasw...with-14-Cores-and-4-Channel-DDR4-292122.shtml

 

[deconstruct60]

edit: I must have missed this article. These are E7-4800V2 for socket LGA 2011. Confusion cleared up.

http://news.softpedia.com/news/Hasw...with-14-Cores-and-4-Channel-DDR4-292122.shtml

One wonders what the difference is between an E5 4600 v2 and a E7 4800 v2 that fit in the same socket. I kind of wonder if Intel isn't playing product renaming bingo with the E5 4600 product line up and just promoting to E7. If the E5 4600 series disappears that is probably closer to the truth.


But the article does say the socket is the same but the pin out is different.
"... Again, a different pin-out will be used, but the same dimensions as socket 2011. ... "
For example, if they flip a few PCI-e pins to QPI pins they could boost the QPI linkages with the E7 variant, but loose some PCI-e lanes . It is doubtful that even a small handful of folks are going to find use for 160 PCI-e lanes anyway. Shrinking to 32 PCI-e lanes per package would allow for 4 x8 PCI-e v3 sockets per package would could be used for 16 x8 slots on a single very large motherboard.

If that is the case, personally I'd label one of those variants something different. Like 'socket 2011' and 'socket 2011Q' . I don't think the socket is technically the same when the connectivity is different.

 

[deconstruct60]

We will see these in the 2013 Mac Pro.. Perhaps in the iMac or MacBook Pro since more tender loving care is given to those.

It is unlikely Apple will use any -E , -EN , or -EX variant in any future Mac. An -EP variant? Yes, but all of these other ones usually are only interesting for the very tangential insights into -EP.

Lots of care is given to the iMac and MBP because the majority of Mac buyers by those two subsets of the Mac line up. Without those two the whole Mac market would implode. It be highly misguided for Apple not to spend alot of time with and resources on those two.

 

[GermanyChris]

I think they should put them in the mini

 

[JavaTheHut]

I think they should put them in the mini

How many WATTS does the cpu consume - You could put a kettle on the top to dissipate heat and get a pot of tea at the same time

 

[GermanyChris]

How many WATTS does the cpu consume - You could put a kettle on the top to dissipate heat and get a pot of tea at the same time

Then the "I dunno why you don't sell your old Mac Pro and buy a mini" people would be right! Just think more appeased Mac users.

 

[deconstruct60]

Then the "I dunno why you don't sell your old Mac Pro and buy a mini" people would be right! Just think more appeased Mac users.

Frankly, that kind of core count is already on track for future Minis. It just isn't purely x86 cores. The CPU + GPU core count is technically at that level now. The Ivy Bridge GPU has 15 cores

http://www.realworldtech.com/ivy-bridge-gpu/2/

They are not quite as mathematically robust as the x86 ones but by the follow on to Haswell (Broadwell 2014 or so) quantity will negate many of the shortcomings. (effectively the same "attack of the killer micros" that has swept the supercomputer space ).

For single user , single instance , embarrassingly parallel math workload applications it will increasing work just as well.

E7 are primarily targeted multiple user , multiple instance , not so embarrasingly parallel workloads that are extremely rare on mainstream Mac so Power/Watts has very little to do with it. The transistor resource allocation is far more misaligned on a fundamental level.

 

[GermanyChris]

Frankly, that kind of core count is already on track for future Minis. It just isn't purely x86 cores. The CPU + GPU core count is technically at that level now. The Ivy Bridge GPU has 15 cores

Image
http://www.realworldtech.com/ivy-bridge-gpu/2/

They are not quite as mathematically robust as the x86 ones but by the follow on to Haswell (Broadwell 2014 or so) quantity will negate many of the shortcomings. (effectively the same "attack of the killer micros" that has swept the supercomputer space ).

For single user , single instance , embarrassingly parallel math workload applications it will increasing work just as well.





E7 are primarily targeted multiple user , multiple instance , not so embarrasingly parallel workloads that are extremely rare on mainstream Mac so Power/Watts has very little to do with it. The transistor resource allocation is far more misaligned on a fundamental level.

This reminds me of this Ars article..http://arstechnica.com/security/201...very-standard-windows-password-in-6-hours/GPU GPU processing will only get better and used for more things. A 680 has 1536 cores you start being able to leverage those the 15 core 30 thread CPU seem sort of meek. I'd love to be able to manually off load tasks to the GPU real processing could be done in the background while you save you "real" processors for your main focus work.

 

[deconstruct60]

edit: I must have missed this article. These are E7-4800V2 for socket LGA 2011. Confusion cleared up.

It looks like the confusion is just starting....

" ... this time again with yet another different pinout within the same physical Socket 2011 dimensions. The Haswell EP Xeon E5 2600 / 4600 v3, ..."
http://vr-zone.com/articles/idf-sf-...-now-with-moar-cores/17188.html#ixzz2JZ9DLd5t

Namely Haswell EP won't change physical socket but will change the pin out at the E5 level also. Changing the pin outs pragmatically still demands a motherboard change. However, the processors aren't compatible. (neither necessarily is the RAM in the Haswell case).

Additionally, the article says that E7 4800 v2 will have 3 QPI links. Since the pin count is the same and need at least 16 pins that highly likely means will loose something else ( unless there is a relatively large number of unused pins in reserve). 3 QPI links is still enough to do gluessless 8. The 4th QPI link in previous E7's was used to link to the Northbridge ( for high speed PCI-e lanes). That's embedded now. Just like old 3600/5600 series needed a QPI to Northbridge that is now gone. So

http://www.qdpma.com/systemarchitecture/systemarchitecture_qpi.html

There are only 3 blue lines to connect. There may or may not be C600 support chip variants that can connect to two E7's with two DMI links so the red links to 'IOHx' aspect may change a bit in the above diagram. 3 QPI also allows for 1 hop max on 4 socket configurations also (every processor package is directly connected to the other 3). For E5 4600 that isn't true (have a 2 hop max. ). So that will be part of the differentiation. E5 is a tad slower I/O, but likely cheaper (like 1600 or 2400 versus 2600 options ).

It wouldn't be surprising if the E7 v2 had the same 24 PCI-e lane limitation as the E5 2400/1400 series.

It looks like Intel has picked physical socket 2011 as a sort of pin superset. They intend to reuse the physical package infrastructure, but electrically move things around overtime. That likely will lead to slightly more folks plugging in stuff that is electrically incompatible just because it physically fits. (e.g., Thunderbolt into mini display port).

I still think that putting a suffix on the socket indicator would probably clear up much of the confusion, because pragmatically they are different.