TiN
02-23-2010, 10:04 PM
Here I will post interesting moments, pics and issues which engineers deal with when design high-speed and reliable hardware.
I guess this is interesting to some overclockers, to know how things are created.
How complex can be 1366 bi-socket design?
Here is some data about Kontron - 2009 PCB TLA computer category winner
http://blogs.mentor.com/davidwiens/files/2009/10/kontron3.jpg
http://blogs.mentor.com/davidwiens/files/2009/10/kontron1.jpg
http://blogs.mentor.com/davidwiens/files/2009/10/kontron2.jpg
Most of current top motherboards are 8 layer PCB's, with components mounted on both sides. Of course complexity of computer boards are much lower that enterprise/high-speed hardware, but still it's advanced tech.
Stats of this 2 socket board next:
http://blogs.mentor.com/davidwiens/files/2009/10/kontron-stats.jpg
308m of trace distance, and 10K of vias!
High current requirements (total 300A for the dual CPU) required close decoupling and made it a hot spot.
This is also interesting ;)
So when we overclock, we see only the most-most top smallest part of whole hardware stuff present. Ramp clock, few voltages, dram timing, and all that easy. No way :). Real things go much more deep, all these crosstalks, skews, serial transiever modes, noise, clock jitter, and decoupling is covered from eye. Everybody tell you that overclocking is dumb easy today? Don't believe, real things are much more harder today.
That's btw, confirmed, by "coldbug mods" and such things. Almost every LN2 clocker who tried Clarks know only "ah, we need to short teh pin to ground, and rock'n'rool". But who precisely knows, what is affected by this exactly?
Not counting shammy ofc. :D
I guess this is interesting to some overclockers, to know how things are created.
How complex can be 1366 bi-socket design?
Here is some data about Kontron - 2009 PCB TLA computer category winner
http://blogs.mentor.com/davidwiens/files/2009/10/kontron3.jpg
http://blogs.mentor.com/davidwiens/files/2009/10/kontron1.jpg
http://blogs.mentor.com/davidwiens/files/2009/10/kontron2.jpg
Most of current top motherboards are 8 layer PCB's, with components mounted on both sides. Of course complexity of computer boards are much lower that enterprise/high-speed hardware, but still it's advanced tech.
Stats of this 2 socket board next:
http://blogs.mentor.com/davidwiens/files/2009/10/kontron-stats.jpg
308m of trace distance, and 10K of vias!
High current requirements (total 300A for the dual CPU) required close decoupling and made it a hot spot.
This is also interesting ;)
So when we overclock, we see only the most-most top smallest part of whole hardware stuff present. Ramp clock, few voltages, dram timing, and all that easy. No way :). Real things go much more deep, all these crosstalks, skews, serial transiever modes, noise, clock jitter, and decoupling is covered from eye. Everybody tell you that overclocking is dumb easy today? Don't believe, real things are much more harder today.
That's btw, confirmed, by "coldbug mods" and such things. Almost every LN2 clocker who tried Clarks know only "ah, we need to short teh pin to ground, and rock'n'rool". But who precisely knows, what is affected by this exactly?
Not counting shammy ofc. :D