AR Fatal1ty 970 Performance- RAM compatibility

dinin70 wrote: When I'm running furmark on CPU it gets pretty quickly over 60° (less than 10seconds) and I prefer stopping before harming too much the CPU.

The CPU will start throttling or will even shut down before it is damaged, assuming the safeguards work as they should. 60 C isn't even close to the danger zone of physically harming your CPU, but the CPU can become unstable well before thermal throttling kicks in.

In any case, if you didn't even have the radiator on, then something's not right. You should at the very least try installing the CPU cooler again from scratch. Remove the CPU cooler, remove all of the old TIM, apply new TIM (read some tutorials on applying the TIM beforehand), reinstall the CPU cooler and see if the temperatures remain the same.

By the way, I have only Fan on the CoolerMaster: must the fan be looking upwards or downwards?

I don't really know what upwards or downwards mean in this context. If you are using the TX3 Evo - or any other tower cooler - the fan should blow the air towards the back panel. This will increase airflow over the VRM section. You should also have a case fan on the back panel sucking hot air out of the case. If you get a top-down blowing CPU cooler, then, as the term suggests, the fan should be blowing air towards the motherboard.

Btw, the TX3 Evo you have doesn't seem to have an official recommended TDP, at least not on the manufacturer's website. Based on the weight and dimensions of the heatsink, as well as the maximum rpm of the 92 mm fan, I'd say it should be okay for 125 W TDP CPUs such as the one you have. However, under full load it may struggle with an octacore and you will likely see high socket temperatures unless you can direct airflow towards the socket.

PetrolHead wrote: Btw, the TX3 Evo you have doesn't seem to have an official recommended TDP, at least not on the manufacturer's website.

I noticed that too, odd, as most other brands do.

I based my remarks of it as it's recommended down to the foot warmer, heat the earth, arctic ice shelf be damned, Ozone? What ozone? ,  Socket 754 procs. And NewEgg reviews also.

Heck, if it can keep a Socket 754 processor cool ...... I thought to myself ... I still took at peek at the NewEgg reviews.

WKjun wrote: AMD's "stock" value is 1866 since the FX line of CPUs.

That's only true for FX CPUs if the motherboard has two available slots in total. If it has four, and even if you're only using two of them, then the stock value is 1600 MHz. If any of the used modules is dual rank, then the stock value is 1333 MHz. AMD's values are likely conservative, however.

Many people say, RAM frequency should be favoured to timings.

In some cases this may be true. There are a lot of variables, however, such as: Intel or AMD, IGP or dedicated GPU, what is the used application and what sort of frequency and latency differences are on the table. For example in this test some applications clearly benefitted from faster memory, some applications didn't really care and for some the sweet spot was one of the slower settings.

http://www.anandtech.com/show/7364/memory-scaling-on-haswell


Edited by PetrolHead - 13 Jan 2016 at 10:20am

It seems weird that a motherboard manufacturer would suggest using a setup that might perform worse. I would instead expect the lower RAM speeds having to do with how much load is put on the IMC, but even in this case the effect of the total amount of available slots seems a bit odd.

And yes, the anandtech article's conclusion does somewhat reflect that general conclusion (although it does add "unless you compare over large MHz ranges"), but if you look at the measurements - and especially if you don't use an IGP - you'll see that the results are far less clear cut. A lot of the results show an insignificant performance differences between the whole range of MHz and CL, especially if you don't include the measurements that can be considered statistical outliers. Even in cases like WinRAR and the h264 HD benchmark, where the trend seems to be pretty clear, you may find that 1866 MHz CL 11 is as fast as 2666 MHz CL10, or that at 2666 MHz the latency doesn't seem to matter, at 2400 MHz lower is better but at 2800 MHz higher was better.

It occured to me that maybe there should be a separate thread for this discussion. Could one of the admins transfer at least these latest messages over to the OC Technical Discussion sub-forum into a new thread and name the thread accordingly, maybe "RAM Clock Speed and Latency"?

PetrolHead wrote: It seems weird that a motherboard manufacturer would suggest using a setup that might perform worse."

It is weird! Maybe for the sake of compatibility?

For example, some excerpts of MB manuals:

970 Performance:
"Please install the memory module into DDR3_A2 and DDR3_B2 slots for the first priority."

Sabertooth 990FX R2.0:
"Recommended memory configurations"
Single Channel: DIMM_A2
Dual Channel: DIMM_A2 + DIMM_B2

At least at Auto, CMD Rate would most likely be 2T, forcing 1T to be unstable, depending on modules and speed set.