DDR4 2133MHz CAS 10 Question

I am upgrading my system now and am considering a i5-6600 CPU on an ASrock Z170M Extreme4 motherboard. Since the CPU is non-K, I won't be overclocking the CPU or RAM. But since I might be using the onboard integrated graphics, I thought that the RAM with the best timings would improve performance a bit.

The Corsair CMD16GX4M4B2133C10 has a CAS latency of 10, is within Intel's spec of 2133MHz, but runs at 1.35v, which I know should work but is not 'officially' supported by Intel with DDR4 on Skylake.

Is it possible to run this memory at 1.2v (to be within Intel's spec officially), but with a CAS latency of no higher than 12?

Also, since the motherboard isn't designed for too much in the way of overclocking, will running all 4 modules at the tightest timings (@1.35v) cause any problems? (some RAM running at very high speeds with relatively tight timings will only work with 2 modules instead of 4 according the memory QVL.)

 

Thanks.

You are mixing things up a bit. Intel just references the current JEDEC DDR4 standard in their specifications. So unless the JEDEC decides to expand the DDR4 standard, anything beyond DDR4-2133 CL15 1.20V is out of spec, that includes higher voltages and frequencies as well as tighter timings. So if you absolutely want to stick to spec, forget about CL10. On top of that the actual performance gain of CL10 over CL15 at DDR4-2133 is negligible compared to what you get with higher frequency memory instead.

 

In terms of performance the sweet spot for my CMK16GX4M4A2800C16 with manual settings was at DDR4-3200 CL15 (1.36V) / CL16 (1.30V). In addition to that it also needed a lot less voltage than DDR4-3000 at CL12/CL13 or DDR4-2800 CL10/11 and was still faster despite the timings*.

 

As far as your other question is concerned, since the scaling of voltage with timings and frequency is different for the various IC types, i'd rather not guess if the a certain kit that does DDR4-2133 CL10 at 1.35V can also do DDR4-2133 CL12 at 1.20V, even though it would be reasonable to assume so.

 

*tested on ASUS ROG Maximus VIII Gene with a 6700K

Thanks for your help. You've cleared some things up for me. However, I'm not sure I understand why the performance gain of CL10 over CL15 at DDR4-2133 would be negligible for integrated graphics (though I can understand how that could be the case if you considered total system performance). Using Anandtech's Performance Index for DDR4 RAM, it looks to me like for my case such a change would improve RAM performance around 25%. Running at CL10 would be more like a 45% improvement. Am I missing something here? Thanks again.

I said the performance gain is negligible compared to what you get from using higher frequency memory. Just from tightening the primary timings you don't get anything near a 25% increase in raw memory performance, but with higher frequency memory that is another story.

 

Even with CL10 @ DDR4-2666 you wouldn't be able to match something like this:

 

http://abload.de/img/cmk16gx4m4a2800c160ope9.png

Sorry to labor the point, but I don't understand why DDR4 2133MHz @ CL10 would not have similar or even slightly better performance in some scenarios than DDR4 3333MHz @ CL15? I'm not questioning that you're probably correct, I'm just trying to understand why that should be the case. Thanks.

Don't confuse the CAS latency with the actual latency of the entire memory action. The CAS latency is the delay to access a particular column on the module. This is only one component in the overall latency.

 

On DDR4, the overall latency as measured by something like AIDA64 will usually be lower with higher frequencies. You would be unlikely to get anything close to the 41.9 ms score posted above with the current possible timings in the 2133/2400/2666 range. In addition, you would also have lower read/write/copy speeds at the lower frequencies. With the possible exception of the 2666/2800 frequencies, you usually get better results by going up to the next standard speed with normal or average timings, than trying to aggressively lower the timings on the speed below.

 

As to the issue at hand, I am not sure if you oppose overclocking the memory or simply would prefer 'safe' settings. As far as longevity, lower voltage is generally better. Most of the kits at 2666 or lower will run 1.20v at their XMP setting. Just about every board should be able to run 2666. One of the kits that can run 2666@1.20v XMP might serve you best while also giving you the ability to run lower speed and tighter timings, if you wanted it. Or perhaps one of the newer 2400 kits with a C10/11 could be relaxed to 1.20v and/or likely overclocked to 2666, but it would require manual tuning of the primary values.

It is just as c-attack described it, to understand why going from 15-15-15-31 to 14-14-14-31 nets you barely 1% of raw memory performance, you would have to understand how memory is accessed by modern processor architectures and that is quite complex. There are dozens of timings involved and manually calculating overall latency is near impossible, even if you have the proper documents from Intel and JEDEC at hand. That is why most people just compare settings against one another and that is what i do in my user reviews (i won't link them since they are written in another language).

 

Corsair Vengeance LPX CMK16GX4M4A2800C16 | Aida64 Read / Write / Copy / Latency (Intel Core i7-6700K @ 4500/4000, ASUS ROG Maximus VIII Gene)

 

DDR4-2133 15-15-15-31 | 31679 MB/s 33050 MB/s 31343 MB/s 55,5 ns

DDR4-2133 14-14-14-31 | 31823 MB/s 33029 MB/s 31467 MB/s 52,8 ns

DDR4-2133 13-13-13-31 | 31978 MB/s 33116 MB/s 31681 MB/s 51,7 ns

 

DDR4-2400 15-15-15-31 | 35475 MB/s 37186 MB/s 35230 MB/s 49,9 ns

DDR4-2666 15-15-15-31 | 39539 MB/s 41324 MB/s 39159 MB/s 46,6 ns

DDR4-2800 15-15-15-31 | 41416 MB/s 43408 MB/s 41005 MB/s 44,8 ns

DDR4-3000 15-15-15-31 | 43731 MB/s 46209 MB/s 43017 MB/s 44,0 ns

 

So while overall latency improves with tighter timings, the actual impact on bandwidth is negligible compared to the difference that even one step up in frequency already makes.

Thanks so much for your explanations. You guys are great. I think now I'll get the "K" CPU so I can overclock the memory a bit to improve the integrated graphics performance.