Slow write speed on 1866Mhz?

gijoe50000

Distinguished
May 27, 2013
176
5
18,715
Got new RAM today 2x4GB GSkill Ares 1866Mhz, 9-10-9-28. Installed ok and its running at 1866Mhz now in the BIOS and CPUZ etc...
I ran a memory benchmark using MAXXMEM2 and the results suggest slow speeds ~5000MB/s
read speeds are around 13MB/s and latencys are about 55ns. Anybody know if these speeds sound normal?
OS is on an SSD with 1/4 free space. and have an almost empty 2TB Seagate for storage..
 
The rest of your system?

If it's the system in your sig it's probably just the IMC on the FX processor. AMD's memory controller is not know for being able to use higher speed RAM. The exception being the graphics side of their APUs.
 
oh right, thought I had them in my sig: FX-8350, Gigabyte 990FXA UD5, GTX 670 PE, 4GB 1866Mhz Apotop, XFX Pro Series
750W, Storm Stryker, Windows 8, Antec Kuhler 620 cooler, 64GB SSD, 2TB Seagate
7200RPM HDD.
 
If you look at the fourth post down in the link I posted you will see he got better scores with DDR3 1600. He also has his processor overclocked to 5.178Ghz. Looking again the RAM is also overclocked to 1926.

I'm not really too familiar with overclocking the newer AMD CPUs but that has to help his score.I do not believe it would make that much of a difference though. Still sounds like you are having problems somewhere.
 
Thanks for the reply.
Yea everything pretty much up to date I'd say.
Its funny alright, just reading somewhere that the AMD IMC's only support up to 1333Mhz RAM.. ?
Still I'm thinking the read and writes should be roughly the same speed though, yea?
 
I have the similar problem with EVO Corsa 1866,9-10-9-28 (2*4Gb,dual) and FX -8350 (non OC). I found on net that results for FX-8350 should be, at least, cca 15000 MB/s for read and 10000 for write in MAXXMEM2. Still slow but that's AMD memory controller problem.
I get 12000 for read and 7300 for write (Asrock 970 Extreme4).
Where I can find the latest drivers for AMD 9xx chipset series?
Thanks!
BR, Igor
 
yea that's about the most I've been able to wrangle from mine too, hit 7000-8000 once but every other time its about 5000 maybe 6000....
From what I gather the memory controller is integrated into the FX-8350 and there aren't any such thing as drivers for the CPU.. but correct me if I am wrong?
Get the chipset drivers from the Motherboard manufacturer website..
 
That's true.FX8350 has built in memory controller but IO functions are performed through chipset connected via hypertransport link. From processor standpoint IO is nothing more that memory regions mapped (thanks to MMU) to some range so for example by writing/ into this range you are writing to some PCIE device. Generally chipset performance affects memory controller performance indirectly. That's i would like to try with another chipset drivers. Also slight overclock of CPU to 4.3 Ghz while DOWNCLOCKING the memory from 1866 to cca 1GHZ with CAS 7 show memory performance increase so writing speed get's stabile around 8500MB/s. This tells me that the bottleneck is not 'bad' AMD controller.
 
I found the root cause. Asrock 970 extreme sets CPU/NB frequency to 2200 MHz and that's to low for 1866 even 1600 Mhz RAM's. IT should be 2400Mhz to not limit memory bandwith. Note that HT link freq. should be set to 2400 as well. If this one drops bellow CPU/NB you may expect problems with IO, like keyboard freezing or not responsive , sound stuttering etc (if APM is disabled). SO On my board both has to be set to 2400 (highest for HT link) With this mem read goes high up to 18 Gb/s per thread and write 12Gb/s per thread (MaxMEM2). But this needs more testing as Sandra measures memory bandwith 360Gb/s for four threads or 90Gb/s per thread what is nonsense. However if I enable AMD APM in BIOS what's disabled by default if CPU/NB is manually increased. Sandra sees (small) improvements from 19Gb/s to 21 Gb/s and MAXMem 15Gb/s read 10 Gb/s write. There is no nonsence any more but CPU throttles at full load (prime 95,temp 55 degrees C). Also I wan't to keeo APM enabled and turbo core active but (I don't want to overclock CPU, just to get memory bandwith as it should be) if so with 'overclocked CPU/NB CPU throttles on full load. Note that this is due to load not temperature.
 
I had been thinking along the same lines. I have NB and HTlink at 2400Mhz for the last few days but the max I'm getting is still ~12800 and ~8000. I've been trying lots of different OC settings from FSB=280 to CPU 5.2Ghz. And lots in between. 5.2Ghz was stable-ish but scary with temps.. 🙂

I also found out that catalyst control didn't update my PCI bus to 3GIO when updating my motherboard drivers, so I had to update that manually in device manager a few days ago. Wonder if there's any other drivers it didn't install?
My memory is currently running at 1.55V 2133mHz 11-12-11-34 but maxxmen shows it as 2-16-2-22 and NB as 800Mhz? Its incorrect 90% of the time, except sometimes after a clear cmos its correct. I'll mess around with the APM and see if it changes anything..
I might drop the frequency and timings to 8 or 9-1600 and see if there's any change.
 
My NB frequency went once to 800 MHz and test's showed that was real (it was slow as DDR1). That happened while HT was set to auto and board has set HT link value to 2000 Mhz (when nb is overclocked). Normally HT auto setting is 2400Mhz.

Enabling APM , for me, keep things 'normal' (but slower) with two major side affects: CPU throttles on high load and prime 95 sometimes reports HW error especially worker on core zero. Strange enough with APM disabled and overclocked NB memory become REALLY fast. Fast enough that synthetic Sandra test get's crazy but no crashing. Seems that memory was (much) faster with non overclocked CPU then with 4.3 GHZ overclocking keeping the same memory and CPU/NB frequency? Somehow I thinking that strange results are due to faster L3 cache. Furthermore, try set in local security policy 'lock memory page' rule to enable(http://msdn.microsoft.com/en-us/library/ms190730.aspx). This may boost performance in case when greater blocks of memory are being copied what's usually part of some benchmarking tests. (Although memory tests should be designed to test only memory excluding as much as possible all other factors as IO speed... ) I didn't try yet to disable that setting,and check whether I still see strange results in Sandra.
Question: Can turbo core stay enabled if APM is disabled? BIOS allows that. I thought that these two are related e.g. that Turbo core decision is made upon APM calculation.