hyper-nova
n00b
- Joined
- Aug 3, 2016
- Messages
- 10
I have two x58 (Gigabyte x58-UD3R) based systems, one of which I recently bought a new kit of DDR3 for. I have had both of these systems for about 5 years but only recently started using them again. I don't ever recall seeing the memory frequency (shown at boot) being above 1333 MHz, despite one of the kits (Kingston) claiming to support 1866 MHz, and the other (new Corsair kit) supporting 1600 MHz.
A caveat to this is that these systems are fitted with Intel Xeons, x5670 and x5675 respectively. Intels website states that these chips support memory up to 1333 MHz, but I am assuming there is no reason why I cannot overclock the memory controller (uncore) part of the CPU?
I have Linux systems running on both, but on one I also have access to a Windows system.
As far as I can tell, this is what seems to be happening. The SPD (memory) multiplier, regardless of what I set it to, always defaults back to x10. I have no idea why this is the case.
I have tried locking the uncore multiplier to 2x the memory multiplier, and reducing BCLK if required to ensure the system should be able to post with the uncore freqency specified. (For example, if I change the memory multiplier to x14, and therefore set the uncore to x28, I reduce BCLK to ensure uncore is not above 3 GHz, as I know it can post with uncore of 3 GHz without any voltage changes.)
I have also tried leaving the uncore multiplier at default. I read conflicting information about uncore and memory multipliers. Some posts suggest uncore has to be at least twice the frequency of the memory frequency, some other posts suggest that since memory is DDR (data is driven on the rising and falling clock edge, therefore the real clock freqency of the memory is only half that of what the BIOS / number on the module states) that uncore has to be twice half that of the memory frequency (aka the same). Others say it can be anything you like, and there is no restriction.
Still, regardless of this tangent, I am not convinced this is related to the problem.
Obviously x58 is pretty old these days but I'm hoping there are some around with experience of these systems.
The reason I want to boost the memory up to higher speeds is that I am currently running workloads which are memory bandwidth limited. My workloads generate a lot of cache misses, so being able to run these higher memory speeds is probably advantageous.
A caveat to this is that these systems are fitted with Intel Xeons, x5670 and x5675 respectively. Intels website states that these chips support memory up to 1333 MHz, but I am assuming there is no reason why I cannot overclock the memory controller (uncore) part of the CPU?
I have Linux systems running on both, but on one I also have access to a Windows system.
As far as I can tell, this is what seems to be happening. The SPD (memory) multiplier, regardless of what I set it to, always defaults back to x10. I have no idea why this is the case.
I have tried locking the uncore multiplier to 2x the memory multiplier, and reducing BCLK if required to ensure the system should be able to post with the uncore freqency specified. (For example, if I change the memory multiplier to x14, and therefore set the uncore to x28, I reduce BCLK to ensure uncore is not above 3 GHz, as I know it can post with uncore of 3 GHz without any voltage changes.)
I have also tried leaving the uncore multiplier at default. I read conflicting information about uncore and memory multipliers. Some posts suggest uncore has to be at least twice the frequency of the memory frequency, some other posts suggest that since memory is DDR (data is driven on the rising and falling clock edge, therefore the real clock freqency of the memory is only half that of what the BIOS / number on the module states) that uncore has to be twice half that of the memory frequency (aka the same). Others say it can be anything you like, and there is no restriction.
Still, regardless of this tangent, I am not convinced this is related to the problem.
Obviously x58 is pretty old these days but I'm hoping there are some around with experience of these systems.
The reason I want to boost the memory up to higher speeds is that I am currently running workloads which are memory bandwidth limited. My workloads generate a lot of cache misses, so being able to run these higher memory speeds is probably advantageous.