Originally Posted by Ansau
Been tweaking memory timings this week and I manage to get some improvements. Highest bandwidth so far: 777000000000000022CC1C0010626B4AD0570D17B50BD509004AE70014051420FA8900A00300000019123139AA2C3617
Around 220000MB/s allows me to reach 16k Firestrike Graphics, about 400 points more from generaleramon best mix. But I see a lot of texture corruption: http://www.3dmark.com/fs/11869882
Next week I will do further test and try 1625MHz strap.
Just a little nano comparison to my system. For encoding SMP and many/ most tasks SMP performs the fastest. Though this shows NUMA pulls way higher memory performance on this application and maybe some/ most possibly. I get even load on all 24 threads in SMP for encoding when NUMA only hits on some... so NUMA encodes slower by a lot but has way more memory bandwidth. Maybe I'll try NUMA for some games.
390X scores - PCIe 2.0 limitation here on my system, scores would be much higher on PCIe 3.0.
The very first test top left is 1.5Ghz per core locked down max power efficiency. Which still straight leaves my 4.8Ghz i7 in the dust on dual xeon's at 1.5Ghz. So funny. Paid literally 400 dollars for my T7500 and at 1.5Ghz the Xeon's only use 47.5 watts each/ total 95 watts. When the i7 would be using well over 200 watts. I'm trailing the 6950x on CPU-z multithread. Though the Ryzen R7 1700 straight smokes the dual xeon's on 65 flippin watts. haha The Ryzen R7 1700 is nothing short of Super-Computer performance without a shadow of a doubt. With the R7 1700 coupled with a Vega that doubles my 390x scores if possible? It will be insane.
Looks like possibly the 390x delimited at 1,250mhz surpasses Titan x maxwell. Though I'm PCIe 2.0 limited like no other, haha. Pascal puts the 390x to shame though most likely hahahaha. I bet if the Maxwell Titan X was clocked at 1333Mhz locked down static it would also put the delimited 390x to shame haha.Edited by chris89 - 3/13/17 at 1:32pm