Memory for 8700k Maximus X Formula

[denali]

Wow, that's pretty sweet. 3600/13 is 3.611ns . voltage is high, but manageable. Almost a 25% improvement below the 4.656ns of the 3866/18 I've got inbound. Those were the highest I could find 2x16 ...with RGB, but non-RGB high was only 4000/19 4.750ns. 3600/13 Something to aim for! I saw you also had somewhere a 4000/12 on a 3200/14 set (or 15?) at 3ns(!!!). Thanks for your help so far. Maybe in a few weeks I can begin OCing.

 

[Woomack]

don't look at CL and ns, it doesn't mean much on current platforms as most is doing cache/CPU and some other things ... in typical work there is no difference between CL13 and CL16 or between 3600 and 4000, you simply won't see that

 

[denali]

don't look at CL and ns, it doesn't mean much on current platforms as most is doing cache/CPU and some other things ... in typical work there is no difference between CL13 and CL16 or between 3600 and 4000, you simply won't see that

This is very confusing to me. How does one decide which RAM to purchase? It seems as if those are the only significant parameters to evaluate. And prices seem to be dependent on them as well. Are you saying to just stick to frequency?

 

[Woomack]

All what you see in the web about counting memory performance is based on CL only ... CL as one timing out of really long list. What you see as memory performance in programs is CPU-Cache-RAM, not only RAM.
New platforms have large and fast cache so even if memory has higher latency then it doesn't change much as cache will cover delays. Higher memory frequency is also lower latency and higher bandwidth. Simply higher frequency gives more than lower latency but the best is always some balance. Too high frequency is also not helping because of memory controller limits but it's above 4000.
The easy/fast setting is 3600 16-16-16 as it works on pretty much every Samsung B, regardless if it's single or dual rank.

Here is what Crucial says about it:
http://www.crucial.com/usa/en/memory-performance-speed-latency

 

[denali]

All what you see in the web about counting memory performance is based on CL only ... CL as one timing out of really long list. What you see as memory performance in programs is CPU-Cache-RAM, not only RAM.
New platforms have large and fast cache so even if memory has higher latency then it doesn't change much as cache will cover delays. Higher memory frequency is also lower latency and higher bandwidth. Simply higher frequency gives more than lower latency but the best is always some balance. Too high frequency is also not helping because of memory controller limits but it's above 4000.
The easy/fast setting is 3600 16-16-16 as it works on pretty much every Samsung B, regardless if it's single or dual rank.

So all of the different prices for RAM, all of the marketing, is pretty much a circus act? There really isn't much of a difference? Did I spend to much for the same performance?

Aside from that. I went with 2x16, thinking it will be easier on my CPU overclock than 4x8. There are faster 4x8 kits, much more expensive, marriage killers. I really might want to expand to 64 later, I love RAM caches, PrimoCache.

I saw a chart that intrigues me, and it makes me wonder if this is what you are talking about. "As Freq goes up a single ns of timing becomes a much higher percent of speed/bandwidth."
http://www.overclock.net/forum/26225889-post6.html

http://content.quadninja.com/image/autodate/2017-07-15/ram_timings_in_ns.png

It makes me wonder what this would look like expanded out to 4800mhz. The graph dark cells do appear to be smoothing out at the 3200 area (around common timings 14-18), with maybe not much change after 3600 by extrapolation.

 

[Woomack]

Because of hardware limitations, latency can't go to 0 and can't even go close to some other lower values. Timings can't be too tight or memory won't be addressed without errors. Simply you can go up with frequency to lower latency but you can't go much down with timings to improve access time ( latency ). Hard to explain that and it's too big topic for one post.

RAM speed isn't only marketing. You pay for better chips and higher binning. There are 3 memory IC manufacturers: Micron, Hynix and Samsung. Each of them has couple of IC depends on density, voltages etc.
- Micron - used by most brands in lower series memory, not so expensive but also isn't overclocking well in most popular IC versions, can find it in kits between 2133-3000 at quite relaxed timings
- Hynix - used by most brands in lower and medium memory series, is overclocking better than Micron but doesn't like tight timings, can find it in kits between 2133-3466 at quite relaxed timings
- Samsung - the most expensive and used by some brands that use it in top memory series ( sometimes also lower series but not often ), can be found in everything between 2133 and 4800 at tight and relaxed timings, all new 3600+ kits in mass production are only on Samsung IC

Considering above, you pay premium for top memory kits which are based on the most expensive Samsung IC after additional binning by memory manufacturer who is also testing memory to guarantee its stability. Some brands like Corsair can make kits on barely overclocking IC and low base frequency and also ask for high price. Example can be Corsair Dominator Platinum which often is on worse IC than Vengeance while it costs more.

 

[wingman99]

Because of hardware limitations, latency can't go to 0 and can't even go close to some other lower values. Timings can't be too tight or memory won't be addressed without errors. Simply you can go up with frequency to lower latency but you can't go much down with timings to improve access time ( latency ). Hard to explain that and it's too big topic for one post.

RAM speed isn't only marketing. You pay for better chips and higher binning. There are 3 memory IC manufacturers: Micron, Hynix and Samsung. Each of them has couple of IC depends on density, voltages etc.
- Micron - used by most brands in lower series memory, not so expensive but also isn't overclocking well in most popular IC versions, can find it in kits between 2133-3000 at quite relaxed timings
- Hynix - used by most brands in lower and medium memory series, is overclocking better than Micron but doesn't like tight timings, can find it in kits between 2133-3466 at quite relaxed timings
- Samsung - the most expensive and used by some brands that use it in top memory series ( sometimes also lower series but not often ), can be found in everything between 2133 and 4800 at tight and relaxed timings, all new 3600+ kits in mass production are only on Samsung IC

Considering above, you pay premium for top memory kits which are based on the most expensive Samsung IC after additional binning by memory manufacturer who is also testing memory to guarantee its stability. Some brands like Corsair can make kits on barely overclocking IC and low base frequency and also ask for high price. Example can be Corsair Dominator Platinum which often is on worse IC than Vengeance while it costs more.

It does sound like a marketing circus act when trying to find good ICs for overclocking when not buying from G.SKILL?

 

[Woomack]

Regardless of brand, if you see 3600+ based on 8 or 16GB modules then it's 95% Samsung B. If it's 3000/3200 CL14-14-14 then it's also Samsung B.
What may cause issues are all memory kits up to 3466 at 16-18-18 or 17-18-18. Under these timings can put pretty much everything. In this way work most brands as it's easy to replace given IC with something else without changing whole product or adding new product number when is shortage of some IC. Brands like Crucial always separate different IC with PN so when you look for additional 2 memory sticks on the same IC then you can simply buy the same product number and be sure it's the same IC or at least really close one which is using exactly the same SPD/XMP profile so will work.

Even G.Skill is sometimes using cheaper IC. Like all early Trident Z were on Samsung. When Trident Z RGB were released then kits up to 3466 at more relaxed timings were almost all on Hynix.
Corsair is sometimes surprising. Even though most brands use Micron only in 2133-2666 kits then Corsair was using them up to 3200 when max OC for higher binned IC was 3200.

If you check what all brands are offering then you will see that 3600+ kits are barely available. Most brands are showing them in press releases but later you simply can't buy anything. The same is with even G.Skill but in this case it's 4266+ while in press releases we see memory up to 4800.
On the QVL lists are memory kits from brands like ADATA, Team Group, Avexir and some others at 4133+. In stores you can't buy more than 3200. Simply no matter what you do, if you want 3800+ then you have 2 brands that will offer that and because of ridiculous price of most higher Corsair kits, almost only G.Skill left.