MemoryReviews

Ballistix Elite 64GB DDR4-3000 Quad Channel Memory Kit Review

Performance and Overclocking

Since this is our first test on this platform, then I will present test results on declared settings ( XMP ) but also overclocking results.

Overclocking is never guaranteed so the presented results may vary from results on other memory kits. I am not recommending overclocking if you do not know what are you doing. High voltages may damage hardware and it will not be covered by warranty.

I wasn’t really expecting so good results out of so high density memory kit. Without any issues I was able to set DDR4-3600. It was as easy as to set DDR4-3600 and 1.35V memory voltage in BIOS and restart PC. Memory was stable but its timings were not perfect so I started to play with additional settings. Quick switch to CL16-16-16 caused no problems. At CL15-15-15 I had to set 1.40V VDIMM, at CL14-14-14 required voltage was 1.45V while tests at CL13-13-13 passed at a bit above 1.5V.

Below you can see maximum settings at which most tests are passing. It’s amazing that so high density memory specified as DDR4-3000 CL15 runs without big problems at DDR4-3600 CL13 in quad channel mode!

It’s also our first memory kit which passed 100GB/s read bandwidth in AIDA64!

I’m not sure if it’s memory or maybe early motherboard BIOS but I wasn’t able to set any higher memory ratio. Next step would be DDR4-3733 but it couldn’t boot regardless of voltages or timings. I will test it again when new BIOS will be released and probably you will see additional overclocking results on our Twitter.

The screenshot above shows impressing performance but there is one thing I have to mention. Regardless if memory timings at DDR4-3600 are really tight like CL13 or more relaxed like CL16, maximum memory bandwidth isn’t much different. What mainly changes is memory latency so access time. It affects system response time but because of large CPU cache we can’t really see big difference while too tight timings may cause instability.

There is one more thing which is even more important on the X299 platform. Memory maximum bandwidth is clearly limited by CPU cache frequency. The i7 7900X default cache frequency is 2000MHz. Maximum after overclocking on our CPU is 3300MHz. Total memory bandwidth improvement because of CPU cache overclocking is up to 10GB/s in AIDA64 benchmark.

All results below were performed at default CPU cache frequency of 2000MHz.

Here is comparison of settings in AIDA64 Memory and Cache benchmark.

As you can see above, memory bandwidth is scalling good with memory frequency up to about 94GB/s. Above this point we can see platform limitations. We can pass it by overclocking CPU cache.

Even at XMP settings memory bandwidth is high and we don’t really need to overclock memory but this platform is for the most demanding users like computer enthusiasts and overclockers so why not to add couple of MHz more.

Memory latency is not scalling as we could expect. Both memory frequency and timings are affecting memory latency. On the graph are no lower settings than the DDR4-3000 but their latency would be higher than 85ns what is clearly not interesting.

XMP settings offer good balance between high bandwidth and low latency. Even though results could be better then it’s more related to whole platform behaviour than the memory itself.

Here are results in Cinebench R15 which are showing performance in rendering benchmark. Also some lower settings were added to this graph.

Results are not much different even though previous test is suggesting otherwise. Ballistix Elite XMP profile is not much slower than the results at DDR4-3600. On the other hand results look clearly better at DDR4-3000 than anything lower.

Memory controller on the Skylake-X is designed to work at DDR4-2666 so I guess that lower settings are not really interesting, especially that even high density memory can work at higher frequencies.

On other platforms I’m using HyperPi 32M as it reacts good to memory performance and used memory ranks. However, HyperPi is not supporting more than 16 threads and the i9-7900X has 20 threads so I decided to show results in SuperPi 32M which is about the same but base on single thread ( which is also easier to compare ).

Results are pretty interesting. Not always lower timings offer higher performance. Looks like balance of XMP profile is good and Ballistix team made some good work programming it this way. This wasn’t expected and I had to run tests couple of times to be sure of results.

The Ballistix Elite is available also in DDR4-3200 version which probably is based on the same memory IC as reviewed DDR4-3000 memory but its memory profiles are different and Ballistix guarantees fully stable work at higher frequency.

Most other benchmarks and games are not really showing performance improvements at higher memory frequency. Results in 3DMark, PCMark and VRMark at anything above DDR4-3000 are close to each other. Simply memory is already fast enough to provide optimal performance.

 

Related posts

Leave a Comment

* By using this form you agree with the storage and handling of your data by this website.

This site uses Akismet to reduce spam. Learn how your comment data is processed.

This website uses cookies to improve your experience. We'll assume you're ok with this, but you can opt-out if you wish. Accept Read More