• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

What is the fastest DDR memory out there these days?

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.

magellan

Member
Joined
Jul 20, 2002
I'm just wondering what the fastest rated DDR4 (I'm guessing it must be DDR4) being sold out there these days?

Is the fastest DDR4 restricted to Skylake only usage? Or will it work on Haswell-E?
 
Problem is not with memory but with memory controllers and motherboards. 4266 is memory with the fastest XMP profile in mass production but you can run it only on maybe 2-3 motherboards on the market. Most Z170 motherboards are limited to ~3600 ( official 3200 ) and most X99 to 3200.
 
Here is my result @4200 17-17-17 1.35V on 2x8GB Trident Z 3600 16-16-16 1.35V. I think it's max you can expect from current batches of Samsung IC and my kit isn't the best. In mass production it will be something like 19-19-19 or 19-21-21 or 19-25-25. What is interesting, new 8GB modules are overclocking better than most of the highest 4GB modules available on the market. I mean you can get 2x8GB 3733 17-17-17 kit but you can get only 2x4GB 18-21-21 or something near.
 
Problem is not with memory but with memory controllers and motherboards. 4266 is memory with the fastest XMP profile in mass production but you can run it only on maybe 2-3 motherboards on the market. Most Z170 motherboards are limited to ~3600 ( official 3200 ) and most X99 to 3200.

My i7-5820 can only run it's cache at somewhere around 3600MHz., so I guess there wouldn't be much point to having 4266 Mhz. memory installed in it. Would it be possible to install say, 3600Mhz. or 4000 Mhz DDR4 memory in an x99 rig, run them at slower speeds and significantly drop the latency timings? If so, would it ever make a noticeable difference in anything?
 
Can get these new kits like 4x8GB TridentZ 3000 14-14-14 or 3200 14-14-14 and run them at 3200 12-12-12 or 13-13-13 but it won't be much faster in general work just because X99 has already high memory bandwidth.
Cache on boards with OC socket can work higher but it's still hard to stabilize it above ~3800-4000. Even if it's working in all benchmarks then it's not guaranteed it will be fully stable.
Here are results with 3000+3600 mentioned Trident Z on X99.
 
Problem is not with memory but with memory controllers and motherboards. 4266 is memory with the fastest XMP profile in mass production but you can run it only on maybe 2-3 motherboards on the market. Most Z170 motherboards are limited to ~3600 ( official 3200 ) and most X99 to 3200.

What does official 3200 mean?
 
What does official 3200 mean?

Most motherboards have guaranteed 3200 memory clock max and there is info that higher frequency is not guaranteed because of possible IMC issues. So nearly all memory manufacturers won't guarantee 3400+ clock even though they have memory on their tested memory list. I just said official as you can make RMA if board can't work at 3200 memory clock with any memory. Above that all depends from support.

Woomack does Skylake run the cache synchronously w/the CPU?

Depends from CPU. K series can run it independent. Locked series have always CPU=cache but in real both settings are affected by power saving technologies so you can see that CPU and cache clocks can be different if CPU isn't under full load or you pick some power saving plan.
On locked CPUs you can't set cache clock manually, it will run at the same clock as CPU.
 
Bart,

It was mentioned above something about the cache not being the same as the memory so there is no point to have memory that fast.
My i7-5820 can only run it's cache at somewhere around 3600MHz., so I guess there wouldn't be much point to having 4266 Mhz.
Now, that platform can't even reach those speeds in the first place, but outside of that, do you find any merit in that statement? Can't say that I do, but I wouldn't bet on it (hence why I reached out to ask).
 
There is always a point where faster memory is improving performance but it's not always bandwidth. In new Intel platforms most depends from cache clock but mainly in bandwidth department.
One example. Let's say that you have 4GHz CPU clock and 3GHz cache clock but memory 3200MHz. In this case raising memory clock won't help much in bandwidth but will help in latency. Latency is being affected by timings but also memory clock.
Now the same 4GHz CPU, 3.0GHz cache and 3.2GHz memory. If you tighten memory timings then you will see higher read/copy bandwidth ( but usually not higher than write bandwidth ) and lower latency.

On X99 no matter what you do, you will be limited to ~3200-3333 memory clock so you can only get higher frequency memory if this memory can run at lower clock and much tighter timings but will be way too overpriced. On this platform new IC will show the same results regardless if you get 3000 CL14 or 3733 CL17 kit. In both cases stable will be something like 3200-3333 13-13-13/14-14-14. If you want new and fast memory for X99 then check Trident Z 2x8/4x8GB 3000/3200 CL14.
This can be reply to post #7 :)

On Z170 memory is scalling up to 4000+ but the biggest gain can be seen up to about 3600-3733. Above that you need high cache clock to see differences in bandwidth. So also in this case not always high memory clock is good idea as you can use it almost only if you set CPU/cache above 4.5GHz.

Other thing is that you will barely see difference in general performance because of memory performance gain.
 
Back