X79 vs X99

[LukaP]

And this is another reason why I'm sure LGA2011 will die with Skylake. SATA needs to die altogether. We need to kill the SATA bus except for 2 3Gbps ports for high-capacity spinning rust, and the other ports can be replaced by a new micro-PCIe connector so we don't lose overall expansion.

we agree again. cant believe this (on the sata side)

 

the LGA2011 needs to die argument is absurd, there are always highend users who need > than mainstream, but arent ready to invest in an equivalent xeon

[Glenwing]

And this is another reason why I'm sure LGA2011 will die with Skylake. SATA needs to die altogether. We need to kill the SATA bus except for 2 3Gbps ports for high-capacity spinning rust, and the other ports can be replaced by a new micro-PCIe connector so we don't lose overall expansion.

 

Yeah, I just don't want the current SATA Express implementation to become too standard, the host connector design is just stupid, they could have easily fit it in the exact same form factor as a standard SATA port, but instead it's larger than 2 of them put together.  Super inefficient with space and looks ugly, lol.  We need some kind of Mini SATA Express connector to be introduced...

[LukaP]

 We need some kind of Mini SATA Express connector to be introduced...

we just need SATAe to become obsolete in favor of M.2 storage, one/two ports on the front, and 4 more at the back, like asrock does on mITX. we dont need that many cables anymore

[patrickjp93]

we agree again. cant believe this (on the sata side)

 

the LGA2011 needs to die argument is absurd, there are always highend users who need > than mainstream, but arent ready to invest in an equivalent xeon

No no no, I just mean the high-end socket needs a total ground-up redesign. Its inability to support USB 3.0 (should be 3.1 at this point >   ) natively is the biggest absurdity. Frankly I see LGA 2400 in the future to deal with the inevitable iGPU addition to the high end to increase throughput as programs go heterogeneous.

 

 

Yeah, I just don't want the current SATA Express implementation to become too standard, the host connector design is just stupid, they could have easily fit it in the exact same form factor as a standard SATA port, but instead it's larger than 2 of them put together.  Super inefficient with space and looks ugly, lol.  We need some kind of Mini SATA Express connector to be introduced...

SATAIO was prepared to do that by 2017. To introduce a totally new standard so soon would have cost too much money. I still want it gone though. SATA costs so much more than either USB or PCIe in royalties. It's 1/3 the cost of a board with 8 SATA ports or more until you hit high-end boards like the WS and overclock boards.

[LukaP]

No no no, I just mean the high-end socket needs a total ground-up redesign. Its inability to support USB 3.0 (should be 3.1 at this point >   ) natively is the biggest absurdity. Frankly I see LGA 2400 in the future to deal with the inevitable iGPU addition to the high end to increase throughput as programs go heterogeneous.

 

 

SATAIO was prepared to do that by 2017. To introduce a totally new standard so soon would have cost too much money. I still want it gone though. SATA costs so much more than either USB or PCIe in royalties. It's 1/3 the cost of a board with 8 SATA ports or more until you hit high-end boards like the WS and overclock boards.

well we get 3.0 for now (i was hoping for 3.1 too, guess its for the 100 series)... 

 

why 2400 though? and im not sure intel can fit so many pins without seriously increasing the already enormous size of the CPU and yeah, we will definately see more iGPU soon. 

[patrickjp93]

well we get 3.0 for now (i was hoping for 3.1 too, guess its for the 100 series)... 

 

why 2400 though? and im not sure intel can fit so many pins without seriously increasing the already enormous size of the CPU and yeah, we will definately see more iGPU soon. 

Or reduce the size of the underside capacitors and things by using graphene interconnects to make the space. It's also possible they could go BGA and mobo makers would be responsible for soldering. This would also heavily reduce TDP due to the direct connections.

 

As per 2400, look at how many pins we added from 775 to get 1156/5/0 to hold the iGPU connections. About 360...and the iGPUs are going to get a hell of a lot bigger.

[LukaP]

Or reduce the size of the underside capacitors and things by using graphene interconnects to make the space. It's also possible they could go BGA and mono makers would be responsible for soldering. This would also heavily reduce TDP due to the direct connections.

i dont believe BGA would work for an enthusiast platform. ever

[patrickjp93]

i dont believe BGA would work for an enthusiast platform. ever

We have enthusiast laptop CPUs don't we? LGA can't go on forever it just produces too much heat that can't be dissipated easily. BGA at least reduces resistance (heat generation) and gives the chip a good flow path to the board.

[LukaP]

We have enthusiast laptop CPUs don't we? LGA can't go on forever it just produces too much heat that can't be dissipated easily. BGA at least reduces resistance (heat generation) and gives the chip a good flow path to the board.

yeah, but enthusiast wont take it. i want to change the cpu maybe. or part of the mobo dies, i want to have the same cpu when i get the mobo back, so i know how to oc it, i could go on

[patrickjp93]

yeah, but enthusiast wont take it. i want to change the cpu maybe. or part of the mobo dies, i want to have the same cpu when i get the mobo back, so i know how to oc it, i could go on

I imagine the mono maker would ship back the same CPU.

[LukaP]

I imagine the mono maker would ship back the same CPU.

you know how much work it is to unsolder that and resolder it? not worth it really... 

[patrickjp93]

you know how much work it is to unsolder that and resolder it? not worth it really... 

Not much. I just soldered a 4980HQ to a reference board with a heat gun and it's working perfectly. The cooling mounts are the same as LGA 115x too. I did manage to tweak the base clock and get it up to 110. Yay 4790k equivalent with Iris Pro Graphics!

[LukaP]

Not much. I just soldered a 4980HQ to a reference board with a heat gun and it's working perfectly. The cooling mounts are the same as LGA 115x too. I did manage to tweak the base clock and get it up to 110. Yay 4790k equivalent with Iris Pro Graphics!

good job, but its still way more expensive for the end consumer, and not as simple as lga

[Thecrazyswede]

they probably figured a way to put the pins there more effectively, using less copper in the die, less different connections, so faster links (PCIe, QPI, etc)

 

Its still 2011 pins, same die size, so idk. seems to me they just shifted it around.

[LukaP]

Its still 2011 pins, same die size, so idk. seems to me they just shifted it around.

the locations of the different connections determine the efficiency of the transfers, which in turn determines the voltage and heat, as well as speed. if they managed to find a better way to utilise the space, we may see faster speeds than if it were the same layout as lga2011v2

[patrickjp93]

the locations of the different connections determine the efficiency of the transfers, which in turn determines the voltage and heat, as well as speed. if they managed to find a better way to utilise the space, we may see faster speeds than if it were the same layout as lga2011v2

What if all these 5% ilc increases are really just due to better sockets and not better architectures from Intel? 

[LukaP]

What if all these 5% ilc increases are really just due to better sockets and not better architectures from Intel? 

nope its not how it works. this may give better frequency for given voltage on the same chip (same as in, this exact chip), not a better IPC

[patrickjp93]

nope its not how it works. this may give better frequency for given voltage on the same chip (same as in, this exact chip), not a better IPC

But what if the IPC increase just has to do with keeping the pipeline more full due to better I/O? CONSPIRACY!

[LukaP]

But what if the IPC increase just has to do with keeping the pipeline more full due to better I/O? CONSPIRACY!

that could be, but then, HT before would not be working, if the pipeline is full anyway

[patrickjp93]

that could be, but then, HT before would not be working, if the pipeline is full anyway

It can still work if enough of multiple threads are available in the onboard cache. I'm joking of course as I'm sure you've guessed.

[LukaP]

It can still work if enough of multiple threads are available in the onboard cache. I'm joking of course as I'm sure you've guessed.

yes i have, but still cache is so small that if IO is lagging in a non HT, then HT will do nothing or very little (only the cache instructions basically)

[patrickjp93]

yes i have, but still cache is so small that if IO is lagging in a non HT, then HT will do nothing or very little (only the cache instructions basically)

Uh, 2MB of L3 cache is pretty big.

[TheProfosist]

So basically they moved the die slightly and called it a new socket ?.

looks skinnier and taller

[LukaP]

Uh, 2MB of L3 cache is pretty big.

for executing programs yes. for rendering, no

[patrickjp93]

for executing programs yes. for rendering, no

Which is why Intel came up with 128 honking MB of eDRAM lol. Someday when prices fall it will come desktop.