FRONTPAGE Kingston DDR4 3000 MHz 4x4GB Review

[EarthDog]

Oh DDR3… it was real, it was fun, but is hasn’t been really fun. DDR3 has came, saw, kicked some rear, and is now starting to see its way out. While Intel’s mainstream platform and all of AMD still use it, the release of DDR4 spells the eventual end to our close friend. Its time to learn something new folks. DDR4 brings lower voltage requirements, higher densities, and higher speeds.

Return to Article...

 

[Woomack]

Memory is maybe good but motherboard has some serious issues with performance. Results are like on much lower clocked memory. I would check other bioses, even these older or even initial. MSI X99S SLI had the same problems on 3rd bios.

 

[EarthDog]

I have been through 3 bios on that board, and also 2 bios on the Asrock X99 Killer Pro with the same results.

Kingston is sending another set of sticks for testing and I am going to get a set from Lvcoyote so I have another to compare them with. That will figure out if its something on my end or the DIMMs themselves as it doesn't seem to be the motherboard... or if it is the motherboard, it did it across 2 different brand boards and as mentioned, a couple BIOS on each.

Its just weird.

 

[RGE]

Thanks for review. Im still debating whether to switch to E platform in future, but until DDR4 latency/speed matures to where it outperforms DDR3, just not excited about paying more for DDR4 for slower RAM, but that was just like when DDR3 first came out, it was slower than DDR2 because of latency.

 

[Woomack]

Did you try to set manually cache clock to 3000 or higher ? I see 1500MHz in CPU-Z/AIDA but on my board it was actually read bug and whatever I set, it was 1500 in system ( even though performance was fine ). When I changed to newer BIOS then software was reading it good but performance has dropped

In general I like design of this memory. Too large etc but on this platform it's not an issue and what's more important , it looks good The same as Predators on DDR3 but finally black PCB and black/gray color scheme.

This motherboard has profiles for Hynix and Micron memory ( I believe there are Hynix chips but I can be wrong ). I would try the one with tighter timings and compare performance. For some reason on my MSI all profiles with more relaxed timings didn't work at all.

Thanks for review. Im still debating whether to switch to E platform in future, but until DDR4 latency/speed matures to where it outperforms DDR3, just not excited about paying more for DDR4 for slower RAM, but that was just like when DDR3 first came out, it was slower than DDR2 because of latency.

DDR4 is not really slower the same as DDR3 wasn't slower than DDR2. The main issue in all cases are early memory controllers or other things that block max bandwidth. On X99 main issue seems limited cache clock. After raising this clock you may achieve much higher memory bandwidth without even touching any memory settings ( and I mean something like 5-10GB/s better results ).
When DDR3 was released then we had X48 and X58 platforms. X48 was slow as C2D were not really designed for DDR3 but X58 was great. My 1st DDR3 kit was 1600 7-7-7 1.65V and it was not long after DDR3 premiere.

 

[RGE]

Did you try to set manually cache clock to 3000 or higher ? I see 1500MHz in CPU-Z/AIDA but on my board it was actually read bug and whatever I set, it was 1500 in system ( even though performance was fine ). When I changed to newer BIOS then software was reading it good but performance has dropped

In general I like design of this memory. Too large etc but on this platform it's not an issue and what's more important , it looks good The same as Predators on DDR3 but finally black PCB and black/gray color scheme.

This motherboard has profiles for Hynix and Micron memory ( I believe there are Hynix chips but I can be wrong ). I would try the one with tighter timings and compare performance. For some reason on my MSI all profiles with more relaxed timings didn't work at all.




DDR4 is not really slower the same as DDR3 wasn't slower than DDR2. The main issue in all cases are early memory controllers or other things that block max bandwidth. On X99 main issue seems limited cache clock. After raising this clock you may achieve much higher memory bandwidth without even touching any memory settings ( and I mean something like 5-10GB/s better results ).
When DDR3 was released then we had X48 and X58 platforms. X48 was slow as C2D were not really designed for DDR3 but X58 was great. My 1st DDR3 kit was 1600 7-7-7 1.65V and it was not long after DDR3 premiere.

memory controller or ram, doesnt really matter, still paying more for slower speed.

And memory controller aside, I can buy 2x4gb DDR3 2600 CAS 10 for $100 that is 24/7 settings not just benching stable. When X99 first came it highest speed was 3000 CAS 15, and most was 2666 CAS 15, and havent improved that much yet.

Going by math performance estimates, and ignoring ddr3/ddr4 architectural differences since cant be quantified and not miracles, 2600 CAS 10 would have 7.69 ns nonsequential latency and 10.38 ns sequential latency. 3000 CAS 15 estimate is 10.0 ns nonsequential latency and 12.33 sequential latency.

So you are saying that 3000 CAS 15 is actually going to perform faster than 2600 CAS 10 if memory controller was equivalent?

 

[Woomack]

It's not only about latency. CPUs use fast cache to reduce delays and higher clocked memory to improve bandwidth. Also there is different rank/banks config. You can see that comparing single and double sided modules on DDR3 and DDR4 platforms. On DDR4 they're almost the same while on DDR3 single sided modules are up to 20% slower.
You can compare results in AIDA64 at 2400, even CL15 modules to DDR3 2400 CL10/11 ... but DDR4 needs faster than stock cache on the CPU side. Most work in this case make the cache.
I was comparing DDR3 to DDR4 at the same clock and timings but cache clock was 3000MHz -> http://www.overclockers.com/forums/showthread.php/750439-DDR3-4930K-vs-DDR4-5820K-comparison in this case DDR4 is slower in most tests while overall results are better because of CPU/cache.

Simply DDR4 will let you achieve higher bandwidth but not lower latency ( without overclocking ). For optimal results you still have to tweak more options.

Memory controller and additional factors like cache clock count a lot in this case. Compare 2600k, 3770k and 4770k at the same 2133 in any memory benchmark and you will see similar latency but much different memory bandwidth results.

I'm not saying that DDR4 is so great but it's just working in a different way and we can't really compare it to the previous generation.
I think that market is not prepared for DDR4 yet and memory manufacturers made almost no binning. What more , even JEDEC didn't make full specification for DDR4 yet.

I won't even mention motherboard manufacturers who knew about DDR4 for a long time and all still have problems with memory ratios ( including ASUS ).

 

[RGE]

I agree with the platform differences and errors in comparing and hence made the disclaimer in my statement, but I am still not expecting miracles from slightly higher bandwidth and very high latencies. So we agree on DDR4 not being ready for prime time yet.

When at least ~3600 CAS 15 is readily available and no higher priced than 3000 CAS15 now, and some of the other issues are sorted, I will look at it again. Just hope that occurs in not too distant future.

 

[Woomack]

I don't think we will see DDR4-3200+ on many motherboards. So far all of them have problems with higher memory ratios and not all are fully stable at higher bclk. I just think that when we will see something like DDR4-3600+ on the market then we will already have new platform or manufacturers make miracles with current boards/bioses.

 

[RGE]

That is sad to hear, but thanks for info..

 

[EarthDog]

I saw that cache speed and set it manually to be sure. But the performance did not change so it seems that was an rror in reading that value.