Only in other forums (search for "AVX3") and both are wild speculation threads, just like this one... viz.: beyond3d.com and evga.com
Last edited:
Where did you see that info? That would be exciting if true.NostaSeronx said:AVX3 for Skylake and Skymont might as they will start using 512-bit FPUs.
Where did you see that info? That would be exciting if true.
I wouldnt put my faith in that articles content.
Singlechannel memory controller, thats a funny one too. Talk about prank.
And even IB Us support 32GB memory, why would Haswell go backwards.
Singlechannel memory controller, thats a funny one too. Talk about prank.
And even IB Us support 32GB memory, why would Haswell go backwards.
I'm just speculating at this point but in cases like the Windows 8 Surface, if it saves power, would you still demand 32GB of memory and >dual channel memory?
It could make sense for ULV Celerons to only use 1 channel of memory.
Supporting 16GB or 2000GB memory doesnt conserve anything.
What about the extra memory controller?
Its not gonna save much, since the main power draw is in the DIMMs density. And the iGPU will die due to bandwidth starvation. Even CPU will be affected. So it would be better to keep dualchannel and simply axe it to a GT1 or lower frequency 100Mhz.
Its not gonna save much, since the main power draw is in the DIMMs density. And the iGPU will die due to bandwidth starvation. Even CPU will be affected. So it would be better to keep dualchannel and simply axe it to a GT1 or lower frequency 100Mhz.
Its not gonna save much, since the main power draw is in the DIMMs density. And the iGPU will die due to bandwidth starvation. Even CPU will be affected. So it would be better to keep dualchannel and simply axe it to a GT1 or lower frequency 100Mhz.
You saw that is a 15w CPU?
And today we got 17W. (With dualchannel.)
So not for me...not paying for an IGP...neither to Intel or AMD...
Not...They've play this trick on Sandy Bridge, where no extra space is required for the AVX compared to Westmere for the new register file system and a new front-end. Although I have absolutely no idea where and how much other small improvements can be made this time.Haswell is very nice on paper but let's see those FMA units in practice before we make "double throughput" statements. Also let's see if each Haswell core can sustain 2x256bit load and 1x256bit store operations per second and by which magic will intel extend the datapaths without blowing up the die size compared to IB(note that Haswell will also have more die area dedicated to new GPU compared to IB;the leaked die shots of SB/IB/Haswell chips from chiphell only show that shown version of haswell with the new GPU is just slightly larger than QC IB chip-by which magic I might ask?)
So you paid a $70-100 motherboard X79 chipset premium that could have been used towards and SSD or a faster GPU, then put up with higher power consumption of i7 3820 for no performance benefit, while missing out the Haswell launch and will be waiting X months before being asked to pay another $70-100 premium for a Haswell-E (or w/e it is) motherboard? Makes sense.
Even in this ES0 version the die area is 15% larger than QC IB.
Singlechannel memory controller, thats a funny one too. Talk about prank.
Performance is my metric..the rest is just static to me.
Well, when heterogeneous computing comes in full swing (quick sync etc)... I'm sure buying a chip with IGP will start counting as performance for you.