i saw a 1fps increase from overclocking my ram from 1333 to 1600mhz.
so OP... Ram overclocking really doesnt give any gains overall.. its so minimal it can be shruged off.
Not all games do the same workloads in the same proportions. The same goes for settings. Some cpu architectures are also more bandwidth heavy than others. You cannot broadly proclaim that memory overclocking is always useless just because it didn't result in a significant increase in performance in your one test which you haven't given us any details on. Users of nehalem cpus have seen significant gains, especially in games that do extensive post-processing effects via efb ram copies.
I also couldn't notice any lost of performance at 900Mhz ..
But thanks for all the good answers
I think this is the first time I disagree with NV. Technically nothing he said is incorrect but Like Venomx1 and admin89 I have never seen any significant performance boost from OCing ram. I probably just don't run any programs that are very bandwidth demanding.
Xalphenos Wrote:I think this is the first time I disagree with NV.
Xalphenos Wrote:Technically nothing he said is incorrect
.......
So you disagree with me but you don't dispute that everything I said was true?
Xalphenos Wrote:I have never seen any significant performance boost from OCing ram
I fail to see how the fact that you're never personally observed it has any significance.
Xalphenos Wrote:I probably just don't run any programs that are very bandwidth demanding.
Bingo!
Saying applications don't need high memory bandwidth is like saying PC games don't need high cpu performance. In most cases it's true, but not always.
Video encoding, cpu based image filters, and software based 3D rendering are good examples of applications that consumer desktops might run that benefit from lots of memory bandwidth. A lot of server apps are memory bandwidth heavy which is why server cpus tend to emphasize it more than consumer desktop cpus.
@slax65
Yeah that's fine.
(04-21-2013, 04:07 AM)NaturalViolence Wrote: [ -> ]So you disagree with me but you don't dispute that everything I said was true?
I have a habit of not being clear, sorry about that. Probably doesn't help that every time I sit down at the computer I have a 2 year old run over and want to play mario kart.
OK so I don't disagree with you, but based on personal experience and the few benchmarks I've seen on the matter I think most users would not see a significant performance gain from higher memory clocks. Not in real world use anyway.
(04-21-2013, 04:07 AM)NaturalViolence Wrote: [ -> ]Video encoding, cpu based image filters, and software based 3D rendering are good examples of applications that consumer desktops might run that benefit from lots of memory bandwidth. A lot of server apps are memory bandwidth heavy which is why server cpus tend to emphasize it more than consumer desktop cpus.
Yeah the only one of those things I do is video encoding. And that I always do on my Phenom II x6 so it probably doesn't benefit from the higher memory bandwidth as much as a Nehalem would.
Out of curiosity do know of any good benchmarks I could read on the subject? I've not seen any in a few years.
Xalphenos Wrote:Probably doesn't help that every time I sit down at the computer I have a 2 year old run over and want to play mario kart.
Sell it and get a cat
They're much easier to maintain.
Xalphenos Wrote:I think most users would not see a significant performance gain from higher memory clocks. Not in real world use anyway.
True. Modern x86 cpus are very good at dealing with slow memory via advanced caching techniques. On top of that we've shifting a lot of the things that do still benefit from it onto the GPU.
One thing that is still done on the cpu yet benefits from increased memory bandwidth is file compression/decompression.
Xalphenos Wrote:And that I always do on my Phenom II x6 so it probably doesn't benefit from the higher memory bandwidth as much as a Nehalem would.
Actually it would benefit more. Bloomfield/gulftown could get away with using slower memory mainly because they had a triple channel memory bus. Especially in the case of bloomfield where you have 3 memory channels and 4 cores.
Compare that to your phenom II X6 with 2 memory channels and 6 cores. 0.33 vs. 0.75 ratio in bandwidth per Hz per core. At the same memory frequency bloomfield will deliver more than twice the memory bandwidth per core so your phenom II is a lot more likely to run into memory bottlenecks.
Keep in mind memory bottlenecks usually only occur in heavily multithreaded software that can actually tax the memory bus.
OP can probably get away with using slower memory but unless the OC benefit is significant I wouldn't bother.
Xalphenos Wrote:Out of curiosity do know of any good benchmarks I could read on the subject? I've not seen any in a few years.
Not really. I did a quick google search on the topic and dug up these:
http://www.tomshardware.com/reviews/core-memory-scaling,2342-8.html
http://www.tomshardware.com/reviews/memory-scaling-i7,2325-10.html
Take a look at the WinRAR results in particular.
(04-21-2013, 08:13 AM)NaturalViolence Wrote: [ -> ]Sell it and get a cat
They're much easier to maintain.
Now to just convince my wife.:p JK I love the little guy.
Thanks for the links I'll give them a read when I get a chance.
Xalphenos Wrote:Now to just convince my wife.
Do it when she's not home. You can always find another wife
.
I would agree with you that for a general consumer desktop high memory bandwidth isn't necessary and provides extremely little if any benefit for the majority of applications that you would be running. However there are some exceptions. And if you're going to use the system as a server it is very important to have plenty of memory bandwidth since server apps tend to me I/O heavy instead of compute heavy.