Discussion Forums > Technology

Future Computer Parts / General Computer Discussions

<< < (284/341) > >>

kitamesume:
they have to make an overhaul in their chip's layout and the socket itself to counter-act the effects of the die-shrink.
one way to go about it is making the chips longer as this creates a larger contact area than a shorter rectangle.
same with the socket, that is to change the square socket in favor for a rectangular socket.

this as an example illustration:


the right-side is a full-die haswell sample.

as noted, the heat-spread area of the longer but narrow die is larger.
and then if you ran heat-pipes across it in this configuration you could fit more heatpipes.
while fewer heatpipes(less than 5) would benefit on contact thats parallel to the die.

as an added thing to this, is to implement a heatpump on the die-contact area when it is soldered
http://en.wikipedia.org/wiki/Thermal_copper_pillar_bump
in this example its the chip-to-PCB contact where the heatpump acts, if they make it a chip-to-IHS as well then the overall heat dissipation rate will increase.


haswell-Es might be a better choice this time around, considering the part that an 8-core haswell-E might get placed in the $700 price-point.
if you compare it to the lesser 6-core ivy-E, its priced around $600.
further more, if they push out the 4-cores out from the E series then we could expect a $500 6-core haswell-Es on the bottom tiers which is great.

that means to say the price-point per-core will drop quite considerably, and DDR4 should come out supported by these chips which is a lot of potential.
in all likeliness though you're looking at no less than $2000 for a ridiculously powerful rig.

Gh0st93:

--- Quote from: kitamesume on June 13, 2014, 02:30:32 AM --- (click to show/hide)they have to make an overhaul in their chip's layout and the socket itself to counter-act the effects of the die-shrink.
one way to go about it is making the chips longer as this creates a larger contact area than a shorter rectangle.
same with the socket, that is to change the square socket in favor for a rectangular socket.

this as an example illustration:


the right-side is a full-die haswell sample.

as noted, the heat-spread area of the longer but narrow die is larger.
and then if you ran heat-pipes across it in this configuration you could fit more heatpipes.
while fewer heatpipes(less than 5) would benefit on contact thats parallel to the die.

as an added thing to this, is to implement a heatpump on the die-contact area when it is soldered
http://en.wikipedia.org/wiki/Thermal_copper_pillar_bump
in this example its the chip-to-PCB contact where the heatpump acts, if they make it a chip-to-IHS as well then the overall heat dissipation rate will increase.

haswell-Es might be a better choice this time around, considering the part that an 8-core haswell-E might get placed in the $700 price-point.
if you compare it to the lesser 6-core ivy-E, its priced around $600.
further more, if they push out the 4-cores out from the E series then we could expect a $500 6-core haswell-Es on the bottom tiers which is great.

that means to say the price-point per-core will drop quite considerably, and DDR4 should come out supported by these chips which is a lot of potential.
in all likeliness though you're looking at no less than $2000 for a ridiculously powerful rig.

--- End quote ---
If I got an Ivy-E chip it would probably be on sale, But honestly all the new stuff is one of the reasons I have little interest in Haswell-E to start with yes there is a lot of potential with DDR4 ram but we won't see any of that in the first year and maybe even into the second year of its release into the wild. Honestly I expect the market DDR4 to be subpar to DDR3 in the beginning because iirc that tends to be the case with RAM. Besides and I know this sound backwards but I wouldn't want to jump in on a platform that is basically getting that massive of an overhaul because you are literally going in blind and can run into so many problems with everything.

So my thinking has been get the best I can from the current standards and in three or so years look at upgrading again when DDR4 should really be king of the hill in terms of both performance and cost, and storage and such have seen large jumps in performance, cost and etc. plus iirc should intel's roadmap hold true that should be around the time Skylake-E would come out and intel could really be on their game at such time. Also I have a feeling haswell's underwhelming performance increase over ivy bridge is going to also be carried over in Haswell-E Something else to think on is that might have something to do with the extra two cores in Haswell-E as to boost overall performance.

kitamesume:
subpar in performance, yes thats quite likely, but you're forgetting two points in DDR4's advantages.
one is the higher dimm density at lower cost, which could lead to 2x16GB modules being cheaper than say 4x8GB DDR3 of the same speed.
the other one is the lower vDDQ and lower overall power consumption, this could also point to a much cooler dimms which is quite a good thing.

aside from ram though the only thing to look forward on haswell-E is the price-per-core, whether they'll boot the 4core bottom tier off the list will affect the price points.
for example, if they add a 6-core no-HT on the bottom tier as a $350 option, that'd slaughter a lot of other things on the same price-point.

$350 ~ 6-core no-HT
$500 ~ 6-core w/-HT
$650 ~ 8-core no-HT
$800 ~ 8-core w/-HT
$1000 ~ 8-core EX w/-HT

Gh0st93:

--- Quote from: kitamesume on June 13, 2014, 03:39:54 AM ---subpar in performance, yes thats quite likely, but you're forgetting two points in DDR4's advantages.
one is the higher dimm density at lower cost, which could lead to 2x16GB modules being cheaper than say 4x8GB DDR3 of the same speed.
the other one is the lower vDDQ and lower overall power consumption, this could also point to a much cooler dimms which is quite a good thing.

aside from ram though the only thing to look forward on haswell-E is the price-per-core, whether they'll boot the 4core bottom tier off the list will affect the price points.
for example, if they add a 6-core no-HT on the bottom tier as a $350 option, that'd slaughter a lot of other things on the same price-point.

$350 ~ 6-core no-HT
$500 ~ 6-core w/-HT
$650 ~ 8-core no-HT
$800 ~ 8-core w/-HT
$1000 ~ 8-core EX w/-HT

--- End quote ---
Like I said DDR4 has a lot of benefits namely the increase capacity per stick, but as I said we probably won't see that right out of the gate and if it does happen then there will be a price premium and again it could suffer flaws. Also on the whole less power usage thing honestly that is truly great but I don't think that is some kind of headline big deal.... Also don't hold your breath on a $350 ~ 6 core unless it is some crazy 5GHz AMD rehash, As that would completely undermine the basic i7-k variants.

kitamesume:
not quite, the i7 LGA1150's HT will compensate for the lack of 2cores, even at a similar price for the processor, the difference in platform price will cause the gap to widen.
which means that even if  the 6core no-HT Haswell-Es will perform a bit better than i7-LGA1150, the difference in overall price will stave buyers from it.

hyperthreading is a weird thing tbh, i mean if they had released an i3-K instead of pentium-K it'd cannibalize the entire locked i5 line  :-|

edit: not to mention the difference isn't 100% more cores like the i3 vs i5, but rather its just 50% more cores since its 4cores w/ HT vs 6cores.
if we put it in thread count its 4C-8threads vs 6C-6threads, its quite likely that the i7-LGA1150 might even outperform said haswell-E.

Navigation

[0] Message Index

[#] Next page

[*] Previous page

Go to full version