board not junk!

wow nice efficiency :)

thanks everyone for your kind words,

your right jon.. this is what we were looking at a few days ago but our ram clocks was what was stopping us...i had myself a bit confused about this platform to start with, but its really no different to bloomfield for pi benching

Solid result!

Why would Gulftown be any different from Bloomfield? Nothing changed apart from size and amount of cores.

higher frequency benching... initially i thought it was scaling slightly different in terms of gains from increasing memory and increasing uncore.. but i was just confusing myself..

there are some minor differences, be them not to performance differences but temperatures used, but like having to manage a cold bugging IMC and pushing ram voltages...

im also not so sure you can get away with low uncore and tight RTL anymore unless you really can get those rams clocks up..

Huh?

I would assume that increasing the CPU frequency would increase the memory/uncore bottleneck, hence the need for better memory is bigger on Gulftown than on Bloomfield. The more performance coming from one component, the bigger the bottleneck from another component?

Something I am still puzzled about is the increased imc-to-dram bus. As far as I understand, it has been increased to a 144-bit to allow MEM/UNC ratio of 2/3. That would make me think GT can achieve same memory bandwidth at 2000/3000 as BF at 2000/4000. Someone care to test?

For benches, the above doesn't matter at all ... you still need UNC clocks for cache bandwidth :D

Noice! Sick results, congrats! Even Andre is pleased! :D :up: