Personal computing discussed
Moderators: renee, Starfalcon
Crayon Shin Chan wrote:Just found out today: LinX with 10000 problem size, at 10 runs isn't enough at all! I was wondering why my system wasn't rock solid even though it passed 10 minutes of Prime95, Memtest86+ and LinX - but when I upped the problem size to 15000, the CPU quickly showed its true colours. It really doesn't like being anywhere near 4GHz, I think.So now I'm at 270x14=3780MHz, down from 275x14=3850MHz. I ran 54 runs of LinX with problem size 17000 on it - now I'll just use it for a week or two without OCing my GTX470 and see if any problems crop up. Then I'll keep the cores at 3.78GHz or less and try to up the bus speed to 285MHz to get the uncore working faster.
Crayon Shin Chan wrote:Turbo is pushing your CPU beyond 4.1GHz, and your CPU can't go faster than 4.1GHz with the amount of cooling/voltage you're throwing at it. Nobody overclocks a CPU with Turbo on anyway - why artificially limit the number of cores that can run at a specified frequency if you can run all of them at a number close to or equal to that specified frequency?
Mr Bill wrote:I'm watching this thread with interest. I've been tempted to upgrade to a Phenom II X6 1090T.
Check out
pages 7-8 of the Lost Circuits Sandy Bridges review
versus page 5 of the Tech Report Sandy Bridges review.
Look at the two memory bandwidth graphs from the two reviews. Take the Phenom X6 1100T. That 15.6 versus 13.4 is the difference caused by using black edition memory timings DDR3 at 1600MHz and NB at 2400MHz in the Lost Circuits review. I get the same boost with my X4 955 BE.
Even running no core overclocking, it gives a nice boost to most things. Once you get that working then mess around with core frequencies. The AOD benchmark will fail every time if you don't get the timings and voltages right when pushing the memory. But for my X4 955 BE, once you have the memory working and code that into the bios, you can fiddle the cores without messing up the memory. I fear long term strain so I just run the memory timings and don't bother overclocking the cores.
mmmmmdonuts21 wrote:Mr Bill, I ran the Sandra benchmark for memory on my system you listed above and got 13.67 GB/s. I forgot what my latency bench was. I can probably look when I get home. Any ways my memory is G. Skill RipJaws 1600 (4x4gb) @ 9-9-9-24-2T and NB@2400 HT@2100. I don't know if that sways you in anyway but I figure I post it anyways.
Mr Bill wrote:mmmmmdonuts21 wrote:Mr Bill, I ran the Sandra benchmark for memory on my system you listed above and got 13.67 GB/s. I forgot what my latency bench was. I can probably look when I get home. Any ways my memory is G. Skill RipJaws 1600 (4x4gb) @ 9-9-9-24-2T and NB@2400 HT@2100. I don't know if that sways you in anyway but I figure I post it anyways.
This system? So what version of Sandra did you use?
MB: Gigabyte 870a-ud3
RAM: 4x4GB GSkill RipJaws 1600
CPU: 275*14 (3850) @ 1.5V
NB: 2750 @1.35V
RAM: 1466 (9-9-9-24-2T)
HT: 2000