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

DDR3+4930K vs DDR4+5820K comparison

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.
I think that memory ratios are all about BIOS. On 1.0 BIOS I saw memory settings up to 3200 on the list but 3000 couldn't boot. In 1.1 version ( which was improving memory compatibility ) is max 2666 and anything higher needs higher bclk. I could boot at max ~3060 but everything above couldn't boot. I hope it will be fixed as even my memory, which isn't high series can run @3000 16-16-16 1.35V but nothing higher no matter if I add voltage or set more relaxed timings.

I didn't check my CPU on ss but on water 4.5GHz needs ~1.35V to pass Prime95. I can run any benchmark at much lower voltage but it's not fully stable. Like I can run benchmarks @4.7GHz ~1.4V on water but when I run something like Prime95 then it's instantly crashing.

My 5820K is giving about the same performance as 4930K. At similar voltage 5820K is overclocking ~200MHz lower what gives about the same results in benchmarks. Hard to say I'm impressed especially with lower memory performance.
 
Thanks for testing Woomack! I'm new here, so I don't quite understand how can you use DDR4-2400 on Core i7-5820K which "officially" only supports DDR4-2133. Can someone explain it to me please?
 
Every memory controller has its standard clock like 2133 for Haswell-E and additional ratios for higher ( or lower ) clocks. On my motherboard are ratios up to x26.66. Memory ratio x bclk = memory clock. Base bclk is 100MHz so maximum effective memory clock without raising bclk = 2666MHz.
Best performance is at clocks between 2400-2800 and tight timings what my memory likes the most so I was mainly testing it at 2400+ clocks.
 
Last edited:
Thanks for explaining Woomack! So there shouldn't be any problem of running DDR4-2400 on a motherboard that supports that frequency, even though Haswell-E doesn't, right?
 
Thanks for explaining Woomack! So there shouldn't be any problem of running DDR4-2400 on a motherboard that supports that frequency, even though Haswell-E doesn't, right?

Standard memory clocks available on every board should include 2133, 2400, 2666 but also lower clocks. My MSI has option to set 1333, 1600, 1866, 2000, 2133, 2200, 2400, 2666.
 
If anyone is interested then so far I see no special difference in performance of single and double sided DDR4 modules. That's probably the biggest improvement comparing to DDR3 where we could see up to 20% lower scores in some tests on single sided modules.
Just as reminder, 4GB modules = single sided, 8GB modules = double sided.
 
Is that because of the quad channel topology as opposed to dual channel? Does DDR3 on X79 respond the same way?
 
Nah, on X79 single sided modules were also slower than double sided. The same as on dual channel platforms. For some reason Samsung IC was losing less in performance in single sided config comparing to Hynix.
In DDR4 I see almost the same performance on 4x4GB and 4x8GB at similar timings and clocks.
 
Hello, I'm a newbie in those cpus+ram. I need to run a bunch of math computation in memory (c++ and c# programs over several GB of data), so I'm planning to buy a new cpu, like a 4930k or 5820K cpu with 32GB memory (all the cores will be used during the computation)
What will you take the 4930K + 32GB DDR3 or the 5820K + 32 GB DDR4 or something else with machine price<$2K? Basically I'm looking for the fastest to do the task
thanks a lot.
 
Last edited:
Even if the 5820 came out a bit behind I would still take x99 because there is a higher than 0% chance that broadwell e has an 8 core midrange chip. Where as there is exactly a 0% chance of new chips being developed for x79.

If you don't need more that 28pcie lanes I can't see how the 4930k would be in any way a better value since the price gap for memory will be mitigated by the much cheaper 5820k.
 
If you buy new hardware then 5820k+DDR4 seem better idea. If you can get 4930K 2nd hand then it probably save a lot of money while will be about as fast after OC. Even if next gen CPUs will be compatible then I think we will also see new chipset or at least improved motherboards with higher memory support. Right now most motherboards support DDR4 up to 2666 and anything above requires to overclock bclk. Some highest series have memory ratios above x26.66 but price is much higher.
There is no clear answer what is better but new stuff has always better options for future.
 
Thanks a lot for your help. As you mention the 5820k has 28pcie but how do I know how many pcie I need? I might plug (later) a 4k screen not for gaming but for 2d graphics, developement etc...

Will it be better to take "A-Data XPG Z1 DDR4-2133 with CAS 13" or the one I put in the list but with a CAS of 15.
I can see also the Crucial one is DDR4-2133 cas 15 at $392 for 32gb....which is $200 less...I'm wondering how much faster it would go (%) if I use one memory vs another one.

Does the following config I put together so far looks reasonable for speed/$? if you have any advise do not hesitate
(the machine will work 24/7 with the 6 cores at 100%)
thanks
 
http://www.overclockers.com/forums/...cial-4x8GB-DDR4-2133-CL15-1-2V-CT4K8G4DFD8213
32GB Crucial 2133 C15 1.2V results.

Most memory kits will run at something like 2666 CL13-14 without too high voltage. My Crucials are working 100% stable at 2666 but it's never guaranteed when you buy lower clocked memory.

28 pcie lanes = 16+8 or 8+8+8 pcie config on board what means it's enough for 2-3 graphics cards and won't slow down much even with the fastest series. Difference between current graphics cards on PCIe x8 and x16 is between 0 to 3%. New drivers will probably give higher difference ;)
For 1 graphics card even with 2x GPU it's more than enough.
 
Some good reading here, and for the most part i can see and understand what is going on.

What i did think while looking threw those first test runs, was: I never felt anyone promised ram that was that much better (That said, i never read much about it, only what users have said on OC.com).
Clock for clock they seem pretty much that same with a variance that you find between identical SSD's, and the variance you find in identical CPU's.

What I am focused on here, is that we are getting the same clocks, but a lower voltage, which from my understanding also is an achievement in it self. I can only compare this to cars, same speed, use less fuel, is a still a gain.

Correct me if my understanding is way off?
 
Manufacturers are promoting DDR4 as much faster option than DDR3. Theoretical bandwidth is higher but in real we are getting much slower memory with much higher latency. Most of the work is doing CPU's cache which is twice as fast in Haswell-E than in Ivy-E. More fast cache = lower requirement to use tighter timings for memory so higher gain from higher memory clock.

But do we really get higher memory clock ? Right now max is 3333+ while most boards are prepared for x26.66 memory ratio and nothing else unless you want to play with bclk. Only some top series have x32 but you have to pay premium for them.

To make DDR4 really fast you have to drop timings a lot and raise voltage to about 1.5V+. I mean not something like 2666 CL14 but something like 2666 CL11-12 or 3000 CL12-13 and at the same time set much higher uncore clock which for regular cooling and motherboard will hit a wall at about 3500MHz. This clock will let you reach ~60GB/s.
60GB/s seem good bandwidth ... but think that X79 on 4x4GB DDR3-2400 CL10 memory is reaching 65GB/s and it's pretty standard memory nowadays. So where is that magical DDR4 bandwidth ? higher memory clock is not really raising results in benchmarks as timings have to be more relaxed.

I'm not saying that DDR4 is bad. It's great if you have good memory kit and know how to optimize all options. However regular users who will run auto or XMP profiles won't see any improvements. General performance will be boosted by fast CPU and cache so they won't really see any slowdowns caused by memory.

Lower voltage = useful only in servers
Higher bandwidth = useful mainly in servers as DDR3 designed for servers was max 1866 and most platforms couldn't run as high anyway
Higher capacity = useful mainly in servers as almost noone needed more than 32-64GB in desktops

DDR4 the same as 8 core CPUs are barely helping in anything in desktop computers. It's normal way of improvements to move to the new technology but so far only server environment can have any advantage of that.

It's not hard to find promises related to DDR4:
http://www.corsair.com/en/memory/popular-memory/ddr4-memory
"The DDR3 standard was announced in 2007, and while the specs have slowly improved over the last seven years, the standard’s being pushed to its limit. DDR4 breaks through that limit, with dramatically faster clock speeds and double the bandwidth… and it uses less power, too. If you play high-resolution games or you run data-intensive applications like video editing, you need the significantly better performance that DDR4 offers."

This is more about CPU/cache , not about DDR4 performance ;) Average marketing info which is trying to convince clients that new is better and it's worth to buy it. Most clients have no idea what are they buying but they believe it's good.
 
Last edited:
Hello,

Why the NB clock is 1200mhz in ddr4 system and 4200mhz in ddr3 ?

We know that the nb clock influences the performance of the memory.

thanks
 
Back