Z170, DDR4 2666, low memory bandwidth "scores"...

Message Topic Search Topic Options Post Reply Create New Topic Printable Version Translate Topic

   Hi,On my Z170M Extreme4 with an Intel Core i3-6100 (3.7GHz) I get terrible DDR4 bandwidth.

The memory is Kingston HX426C15FBK2/16 : http://www.kingston.com/dataSheets/HX426C15FBK2_16.pdf

The memory is working in dual channel mode at its XMP settings so
2666Mhz 15-17-17-35-347-2T

The measured bandwidth sounds very low to me (but I can be wrong) :
Single core Read 13.7 GB/s
Single core Write 10.3 GB/s
Single core Mixed 13.3 GB/s
Multicore Read 17 GB/s
Multicore  Write 10.2 GB/s
Multicore  Mixed 12.4 GB/s

Does it looks like expected DDR4 2666 bandwidth? or is there any evidence of bad perfs?

Thanks

EDIT : added some screens of one of the bench 

With some OCing

Edited by propergol - 07 Feb 2016 at 2:25am

Author	Message Topic Search Topic Options Post Reply Create New Topic Printable Version Translate Topic
propergol Members Profile Send Private Message Find Members Posts Add to Buddy List Newbie Joined: 03 Feb 2016 Status: Offline Points: 22	Post Options Post Reply Quote propergol Report Post Thanks(0) Quote Reply Topic: Z170, DDR4 2666, low memory bandwidth "scores"... Posted: 07 Feb 2016 at 12:12am
	Hi, On my Z170M Extreme4 with an Intel Core i3-6100 (3.7GHz) I get terrible DDR4 bandwidth. The memory is Kingston HX426C15FBK2/16 : http://www.kingston.com/dataSheets/HX426C15FBK2_16.pdf The memory is working in dual channel mode at its XMP settings so 2666Mhz 15-17-17-35-347-2T The measured bandwidth sounds very low to me (but I can be wrong) : Single core Read 13.7 GB/s Single core Write 10.3 GB/s Single core Mixed 13.3 GB/s Multicore Read 17 GB/s Multicore Write 10.2 GB/s Multicore Mixed 12.4 GB/s Does it looks like expected DDR4 2666 bandwidth? or is there any evidence of bad perfs? Thanks EDIT : added some screens of one of the bench With some OCing Edited by propergol - 07 Feb 2016 at 2:25am

parsec Members Profile Send Private Message Find Members Posts Add to Buddy List Moderator Group Joined: 04 May 2015 Location: USA Status: Offline Points: 4996	Post Options Post Reply Quote parsec Report Post Thanks(0) Quote Reply Posted: 07 Feb 2016 at 1:31am
	What are you using to test memory bandwidth, so we can use it ourselves for testing and comparison purposes? Depending upon the test being run, and if you have CPU power savings options enabled, C States in particular (which I normally have enabled), can cause relatively high latency. The processor cores can be literally powered down, and the time it takes them to wake up can definitely affect benchmark and real world performance. That is even noticeable when running storage benchmarks on SSDs, whose latency is still much greater than a processor's and DRAM memory under any circumstance. The maximum memory bandwidth of all Skylake processors, including yours, is 34.1 GB/s: http://ark.intel.com/products/90729/Intel-Core-i3-6100-Processor-3M-Cache-3_70-GHz The memory timings of DDR4, being almost if not exactly twice that of common DDR3 memory (DDR4 of 17, DDR3 of 9) is a factor not in favor of DDR4 memory.


propergol Members Profile Send Private Message Find Members Posts Add to Buddy List Newbie Joined: 03 Feb 2016 Status: Offline Points: 22	Post Options Post Reply Quote propergol Report Post Thanks(0) Quote Reply Posted: 07 Feb 2016 at 1:57am
	Thanks for the infos. I have tried everything including reseting bios serval times, turning off all CPU C states stuff, using XMP profil, using manual profil, even used slot A1-B1 rather than advised A2-B2. I get consistant same results using different benchmarks. The results I did post are coming from a very small and short bench from http://www.userbenchmark.com/ My results are here at the bottom of the page you have the memory bench : http://www.userbenchmark.com/UserRun/737637 SiSoftware Sandra gives me nearly the same results. Edited by propergol - 07 Feb 2016 at 2:00am

Xaltar Members Profile Send Private Message Find Members Posts Add to Buddy List Moderator Group Joined: 16 May 2015 Location: Europe Status: Offline Points: 25073	Post Options Post Reply Quote Xaltar Report Post Thanks(0) Quote Reply Posted: 07 Feb 2016 at 2:00am
	As Parsec said, without knowing the benchmark you used we can't know for sure. Just an educated guess though I would be inclined to believe they are low scores based on my own memory benchmark results across several different benchmarks. I use Maxxmem2 for my reviews as a memory benchmark, maybe you can try run that 3-4 times and average the results and post back. You should easily find it via google and there is a free version. Here is a benchmark of my system using DDR4 @2133 but I have tightened my timings quite a bit as you can see. [edit] you posted while I was posting Edited by Xaltar - 07 Feb 2016 at 2:00am


propergol Members Profile Send Private Message Find Members Posts Add to Buddy List Newbie Joined: 03 Feb 2016 Status: Offline Points: 22	Post Options Post Reply Quote propergol Report Post Thanks(0) Quote Reply Posted: 07 Feb 2016 at 2:18am
	I did use the same benchmark as you and did one test with "stock" settings @1.2V and an other with some OC and lower timings @1.35V. Aside from my latency all other results are....low :

parsec Members Profile Send Private Message Find Members Posts Add to Buddy List Moderator Group Joined: 04 May 2015 Location: USA Status: Offline Points: 4996	Post Options Post Reply Quote parsec Report Post Thanks(0) Quote Reply Posted: 07 Feb 2016 at 2:25am
	Here are a few runs I did using AIDA64 Extreme. Sorry for all the pics, the free version's main memory benchmark blocks out to many results on one page: READ WRITE LATENCY My memory is at 3200, so above yours in speed. All benchmarks are different and can't be compared between each other. Please try what Xaltar used, MaxxMEM², and I used AIDA64 Extreme free edition, here: http://www.aida64.com/downloads Run them and post your results, and then we can decide if your memory is lacking in speed. I would be surprised if it is honestly, IMO it's the difference in benchmarks you are seeing. I'll try running the one you used later.


parsec Members Profile Send Private Message Find Members Posts Add to Buddy List Moderator Group Joined: 04 May 2015 Location: USA Status: Offline Points: 4996	Post Options Post Reply Quote parsec Report Post Thanks(0) Quote Reply Posted: 07 Feb 2016 at 2:35am
	propergol, I see you beat me with my suggestion, but that is great! No need to run AIDA64, we can use MaxxMEM. So to be on the same page... uhm, benchmark, here ya go: Your results do seem low. Is that caused by your CPU being different? Xaltar and I have the same processor, mine OC'd to 4.2GHz. More from me later.


Xaltar Members Profile Send Private Message Find Members Posts Add to Buddy List Moderator Group Joined: 16 May 2015 Location: Europe Status: Offline Points: 25073	Post Options Post Reply Quote Xaltar Report Post Thanks(0) Quote Reply Posted: 07 Feb 2016 at 2:37am
	Maxxmem2 is pretty accurate from my testing and even my DDR3 kits have provided better results than yours so something is definitely up there. I would try clearing CMOS, I usually use this method to be absolutely sure it is cleared correctly. Once you have cleared CMOS do not alter anything in the BIOS, leave everything at factory defaults and run the benchmark again to get a basal score. Once you have that you can then go back and set your XMP profile and bench again. If the score is still low try setting your RAM to DDR4 2600 (using a 100mhz base rather than the 133mhz base of 2666mhz) and see if that makes a difference. Post back your results.


propergol Members Profile Send Private Message Find Members Posts Add to Buddy List Newbie Joined: 03 Feb 2016 Status: Offline Points: 22	Post Options Post Reply Quote propergol Report Post Thanks(0) Quote Reply Posted: 07 Feb 2016 at 3:40am
	I already have clear CMOS jumper+battery at least 3 time. Booted with stock settings = same results. My 2 Kingston sticks are on the compatibility list for the Asrock Z170M Extreme4 so I realy don't catch what's happening.

Xaltar Members Profile Send Private Message Find Members Posts Add to Buddy List Moderator Group Joined: 16 May 2015 Location: Europe Status: Offline Points: 25073	Post Options Post Reply Quote Xaltar Report Post Thanks(1) Quote Reply Posted: 07 Feb 2016 at 4:04am
	That is very odd. I assume you have no stability issues with the system and that it has passed stress testing with either Intel Burn in Test (IBT) or Prime95? Those scores are sub DDR3 1333 and that should not be the case at all. It may be worth trying to reseat the CPU and checking for bent pins in the socket. Did you try setting the RAM speed to 2600 and testing again? Also when you say same results do you mean DDR4 2133 (stock settings) and DDR4 2666 are yielding the same (or very close) scores? Another quick question, are you on the latest BIOS for your board? I would avoid the one that disables Sky OC if you plan to overclock the system though.