Socket 3647 goodness

lutjens

Gawd
Joined
Feb 18, 2013
Messages
792
Found this on Supermicro's site...

Supermicro | Products | Motherboards | Xeon® Phi x200 Serverboard | K1SPE

What a beast of a Socket...and it supports 260W TDPs processors. Hopefully, there'll also be Skylake-EP processors with such a TDP...such a part would substantially decrease the single threaded performance deficit on the HCC chips (and supercharge their multi-threaded performance as well).

Something to look forward to, in any event...;)
 

morningreis

[H]ard|Gawd
Joined
Aug 10, 2005
Messages
1,574
I'd like to know how a setup with this platform differs from the Xeon Phi co-processor aside from it not residing in a PCIE slot.
 

lolfail9001

[H]ard|Gawd
Joined
May 31, 2016
Messages
1,496
I'd like to know how a setup with this platform differs from the Xeon Phi co-processor aside from it not residing in a PCIE slot.
Well, to start with, you actually have a conventional x86 system on your hands, instead of wacky setup Xeon Phi cards have.

Found this on Supermicro's site...

Supermicro | Products | Motherboards | Xeon® Phi x200 Serverboard | K1SPE

What a beast of a Socket...and it supports 260W TDPs processors. Hopefully, there'll also be Skylake-EP processors with such a TDP...such a part would substantially decrease the single threaded performance deficit on the HCC chips (and supercharge their multi-threaded performance as well).

Something to look forward to, in any event...;)
Mate, keep up with leaks, Intel ARE separating enthusiast and server platforms with Skylake/Kaby Lake, Purley chips will be as hard locked as they come unless Intel screws up: leaves the BCLK door open and lets mobo makes make a worthy successor to EVGA SR-2. Also, looks like unlocked quad core Kaby Lakes will be moved to Socket R4 as well.
 

lutjens

Gawd
Joined
Feb 18, 2013
Messages
792
Mate, keep up with leaks, Intel ARE separating enthusiast and server platforms with Skylake/Kaby Lake, Purley chips will be as hard locked as they come unless Intel screws up: leaves the BCLK door open and lets mobo makes make a worthy successor to EVGA SR-2. Also, looks like unlocked quad core Kaby Lakes will be moved to Socket R4 as well.

I'm well aware of the plans to separate the dual Xeon parts from the single socket enthusiast ones. What Socket 3647 represents is the possibility to have Xeons that are high core count, but also have decent clock speeds as well and be MUCH better single threaded performers. A 260W TDP Xeon would effectively be factory overclocked compared to the detuned chips of today that are forced to make do with very low clocks, just to stay in a low, outdated thermal envelope. They'd be running at much closer to their full potential. The E5-2679 V4 is a 20-core 200W chip running at 3.2GHz across all cores. Increasing the TDP to 260W (and getting rid of the FIVR) should allow a very healthy further boost in clock speeds.
 

lolfail9001

[H]ard|Gawd
Joined
May 31, 2016
Messages
1,496
I'm well aware of the plans to separate the dual Xeon parts from the single socket enthusiast ones. What Socket 3647 represents is the possibility to have Xeons that are high core count, but also have decent clock speeds as well and be MUCH better single threaded performers. A 260W TDP Xeon would effectively be factory overclocked compared to the detuned chips of today that are forced to make do with very low clocks, just to stay in a low, outdated thermal envelope. They'd be running at much closer to their full potential. The E5-2679 V4 is a 20-core 200W chip running at 3.2GHz across all cores. Increasing the TDP to 260W (and getting rid of the FIVR) should allow a very healthy further boost in clock speeds.
That's correct, though i am of opinion that even 260W TDP won't make these Skylake-EX chips go beyond 3.4Ghz all-core clock just because of sheer amount of transistors. And even then, they will enjoy the 2679 status at best (though that's still infinitely more obtainable than 1691v3, sigh).
 

lutjens

Gawd
Joined
Feb 18, 2013
Messages
792
That's correct, though i am of opinion that even 260W TDP won't make these Skylake-EX chips go beyond 3.4Ghz all-core clock just because of sheer amount of transistors. And even then, they will enjoy the 2679 status at best (though that's still infinitely more obtainable than 1691v3, sigh).

Probably not, but hopefully we'll see some 16-20-core chips with a 260W TDP at or above 3.6GHz, all core. The fully-fledged 28-core Skylake-EX probably hits 260W very easily, probably right about the 3.4GHz you predicted. Hopefully, a higher TDP rating like 260W will permit a more aggressive Turbo Boost Max 3.0 implementation that will help the higher-core count chips put at least 4-cores up to an aggressive clock without the thermal monster rearing it's ugly head. This would help narrow the single-threaded performance deficit even further.;)
 
Top