TR Forums

For anybody operating under the theory that LGA-2011 is a big socket, Intel has some bad news for you:

https://www.servethehome.com/big-socket ... -lga-3647/

Knight's Landing chips are indeed huge. Think credit card sized packages.

Look at the size of that...

...package. 

edit: BRUNO! Why aren't pasted links clickable anymore?

derFunkenstein wrote:

Look at the size of that...

...package. 

edit: BRUNO! Why aren't pasted links clickable anymore?

I think phpBB was confused by the colon preceding the link. Putting it on its own line fixed the problem.

Good catch. I PM'd Bruno and had the same result, but my PM also had a colon in front of the link.

test viewtopic.php?f=2&t=118621

edit: that seems to be it. Good catch, Ned! We now return you to your regularly-scheduled forum thread about enormous sockets.

Reminds me of the Pentium Pro...

GOOD GOD! That is a big ass Socket and CPU. How heavy are the HSFs for this socket?

whm1974 wrote:

GOOD GOD! That is a big ass Socket and CPU. How heavy are the HSFs for this socket?

And have we reached the point where the HSF has to be bolted to the mobo tray to avoid broken mobos?

Though, most of these beasties will sit on horizontal mobos.

To all that have pulled a CPU when removing their heatsink-
This is the method that solves that problem.........................

LGA 3647 is awesome! Thank goodness the mainstream shmucks were able to fund the development of this monster. Are there any consumer mATX boards on the market yet that come equipped with this socket? I'm looking to upgrade my HTPC.

End User wrote:

LGA 3647 is awesome! Thank goodness the mainstream shmucks were able to fund the development of this monster. Are there any consumer mATX boards on the market yet that come equipped with this socket? I'm looking to upgrade my HTPC.

What in hell are you going to do with 72 Atom Cores? I'm quite sure the standard CPUs will be much faster for the vast majority of your cases.

LGA-3647 is going to be only for upcoming Skylake-EP/Knights Landing and their future counterparts. The socket is basically a throwback to sockets used for the old Xeon MP line-up.

It is going to be "server/HPC" only. It is part of Intel's answer to Nvidia/AMD's GPGPU solutions. I doubt we will see desktop/workstation flavors anytime soon.

The article itself is old news. It is has been known for a while that Intel was switching out sockets for their next-generation EP line-up.

whm1974 wrote:

End User wrote:

LGA 3647 is awesome! Thank goodness the mainstream shmucks were able to fund the development of this monster. Are there any consumer mATX boards on the market yet that come equipped with this socket? I'm looking to upgrade my HTPC.

What in hell are you going to do with 72 Atom Cores? I'm quite sure the standard CPUs will be much faster for the vast majority of your cases.

72 quad-thread Atom cores with beefed up AVX2 execution resources...

They're fun chips.

whm1974 wrote:

End User wrote:

LGA 3647 is awesome! Thank goodness the mainstream shmucks were able to fund the development of this monster. Are there any consumer mATX boards on the market yet that come equipped with this socket? I'm looking to upgrade my HTPC.

What in hell are you going to do with 72 Atom Cores? I'm quite sure the standard CPUs will be much faster for the vast majority of your cases.

I think you need to re-calibrate your sarcasm detector.

just brew it! wrote:

whm1974 wrote:

End User wrote:

LGA 3647 is awesome! Thank goodness the mainstream shmucks were able to fund the development of this monster. Are there any consumer mATX boards on the market yet that come equipped with this socket? I'm looking to upgrade my HTPC.

What in hell are you going to do with 72 Atom Cores? I'm quite sure the standard CPUs will be much faster for the vast majority of your cases.

I think you need to re-calibrate your sarcasm detector.

Well OK maybe I need to.

There aren't any more pins on the KNL-F sockets, but they do have a neat little diving board / tail where the integrate Omnipath exits.

I have seen some very, very interesting systems with these on-board. One of the cooler items is the very large integrated memory pool on the new chips. 16GB of "MCDIMM" goodness...

This stuff looks like EX, rather than EP-series stuff.

If true that Skylake-EP is gonna be LGA-3467, I wonder what that means for Skylake-E and the consumer/prosumer HEDT boards. Maybe we'll see an LGA-2011-4? Or will they just phase out that socket and start adding more cores to LGA-1151?

This is why we don't see GPUs in sockets. What LGA-3647 effectively represents is a huge "graphics motherboard" with a GPU-like processor that has a vast number of memory interconnects.

Topinio wrote:

Reminds me of the Pentium Pro...

Exactly my first thought.

Side note: Had a Pentium Pro server years ago. It was a quad-capable server, but we'd ordered it as a dual. So on the order it was listed as "base single CPU system with dual CPU upgrade option." They screwed up in production and the system ended up with three Pentium Pro CPU's instead of two. Obviously we didn't complain about it. That thing ran NT 4 Server like a champ.

LOL! I can see how it happened though:

Base single CPU system (1 CPU) with dual CPU upgrade option (+2 CPUs) = 3 CPUs!

curtisb wrote:

Exactly my first thought.

Side note: Had a Pentium Pro server years ago. It was a quad-capable server, but we'd ordered it as a dual. So on the order it was listed as "base single CPU system with dual CPU upgrade option." They screwed up in production and the system ended up with three Pentium Pro CPU's instead of two. Obviously we didn't complain about it. That thing ran NT 4 Server like a champ.

Niiiice. Ours were 2S workstations only and ran Slackware 3, IIRC, certainly the shiny new 1.2 kernel and ELF Got them on RH before retirement though, a big move and a big improvement at the time.

I just realised this new chip's pin count is up an order of magnitude too...

Topinio wrote:

I just realised this new chip's pin count is up an order of magnitude too...

Moving the memory controller (and other functions) on-die, having multiple memory channels, etc. will tend to do that.

My first CPU had 40 pins.

You can buy CPUs in 8-pin DIP packages.

whm1974 wrote:

What in hell are you going to do with 72 Atom Cores? I'm quite sure the standard CPUs will be much faster for the vast majority of your cases.

Physicalization where each core cluster gets its own VM. As long as the cores are able to meet minimum latency metrics (i.e. response times), performance just scales with core counts on select work loads (think web servers etc.).

jihadjoe wrote:

This stuff looks like EX, rather than EP-series stuff.

If true that Skylake-EP is gonna be LGA-3467, I wonder what that means for Skylake-E and the consumer/prosumer HEDT boards. Maybe we'll see an LGA-2011-4? Or will they just phase out that socket and start adding more cores to LGA-1151?

It appears that Intel will use the same socket for EP and EX. From all indications, the EX line will continue to use serial-to-parallel memory buffers to boost memory capacity.

While not exceedingly popular, Intel did offer Sandy Bridge-EN, Ivy Bridge-EN and Haswell-EN in socket 1356. These were cut down version of the EP line in the same generation and only had three memory channels instead of four. Kinda unique is that the Haswell-EN chip could be used with DDR3 memory. Not sure if Intel will continue to offer a 'small form factor' chip for blades but a consumer Sky Lake-E could utilize suck a socket if Intel didn't want to go all-in on LGA 3647. Then again, the Xeon D line up has reduced the roll of the EN lineup significantly.

just brew it! wrote:

Moving the memory controller (and other functions) on-die, having multiple memory channels, etc. will tend to do that.

Sure, just found it amusing that we were now up at that point.

just brew it! wrote:

My first CPU had 40 pins.

Yeah, tech was a bit different ... my first two machines were, 8048 then 6502 (not that I knew it at the time).