DDR4-4000: does it make a difference?

Steve

Posts: 3,035   +3,142
Staff member

Normally I don’t pay much attention to e-mails that hit my inbox from manufacturers claiming that a certain motherboard can overclock a Skylake Core i7 processor the highest, or that they have claimed the 3DMark record. They are in my mind boring marketing tactics that mean little to nothing to the consumer.

Last month I received one such e-mail from Asrock that nonetheless caught my attention. It claimed their Z170M OC Formula was the only motherboard to support G.Skill’s Trident Z DDR4 4333 MHz modules. Initially I thought, how useful is that? Are there even any benefits from running DDR4 memory on the LGA1151 platform that high?

For the most part we test using DDR4-3000, as it occasionally shows some benefits over the more typical 2400 and 2666 MHz speeds. Going to 4000 MHz and beyond is a massive increase in frequency (and cost) and I struggled to imagine where this would be useful, particularly when gaming. Then again, curiosity had gotten the better of me...

Read the complete article.

 
This test was made with very powerfull CPU and GPUs, but only with 8GB of RAM. I think there might be a chance, that in benchmarks, the sysmtem runs out of memory and than faster memory can give some more FPS. Maybe if the test was made with at least 16GB of RAM, than DDR-2133 can give same FPS like DDR-4000.
 
"t's great to see how much of a step forward the Skylake memory controller is, even if the IPC performance was underwhelming when compared to Haswell. "

Except that skylake was getting the same bandwidth on 3000 MHz ddr4( 27.4GB/s) that haswell was getting with 2133 MHz ddr3 (27.2 GB/s) according to this very site. Other sites show haswell hitting 28GB/s with the same speed. How is that an improvement in any way? That's a major regression, not an improvement.

Either A). Skylake's memory controller is garbage, or

B). DDR4's latency really does make that extra speed useless, by hampering bandwidth.

So why is skylake praised for having a better controller, when it is demonstrably inferior to haswell?
 
This test was made with very powerfull CPU and GPUs, but only with 8GB of RAM. I think there might be a chance, that in benchmarks, the sysmtem runs out of memory and than faster memory can give some more FPS. Maybe if the test was made with at least 16GB of RAM, than DDR-2133 can give same FPS like DDR-4000.
Its very rare for Photoshop to use more than 8GB of RAM, Handbreak never will, and I've yet to see any game use more than 8GB. I personally have 16GB in my system as one day we'll breach the 8GB barrier, but today is not that day.
 
This test was made with very powerfull CPU and GPUs, but only with 8GB of RAM. I think there might be a chance, that in benchmarks, the sysmtem runs out of memory and than faster memory can give some more FPS. Maybe if the test was made with at least 16GB of RAM, than DDR-2133 can give same FPS like DDR-4000.
Its very rare for Photoshop to use more than 8GB of RAM, Handbreak never will, and I've yet to see any game use more than 8GB. I personally have 16GB in my system as one day we'll breach the 8GB barrier, but today is not that day.

IDK wouldn't the benefit of a fast ram be that you can use less?? and at this time the lower end DDR4 are already cheaper then the equivalent DDR3, while applications should also be able to better utilize the faster memory now that it has become the standard
 
This test was made with very powerfull CPU and GPUs, but only with 8GB of RAM. I think there might be a chance, that in benchmarks, the sysmtem runs out of memory and than faster memory can give some more FPS. Maybe if the test was made with at least 16GB of RAM, than DDR-2133 can give same FPS like DDR-4000.

I see where you're coming from. I also wouldn't mind seeing if swapping the i7 for an i5 would make a noticeable difference.

Does anyone think the results would be any different? Thanks.
 
"t's great to see how much of a step forward the Skylake memory controller is, even if the IPC performance was underwhelming when compared to Haswell. "

Except that skylake was getting the same bandwidth on 3000 MHz ddr4( 27.4GB/s) that haswell was getting with 2133 MHz ddr3 (27.2 GB/s) according to this very site. Other sites show haswell hitting 28GB/s with the same speed. How is that an improvement in any way? That's a major regression, not an improvement.

Either A). Skylake's memory controller is garbage, or

B). DDR4's latency really does make that extra speed useless, by hampering bandwidth.

So why is skylake praised for having a better controller, when it is demonstrably inferior to haswell?

It is well established that DDR4’s latency makes is slower than DDR3 at the same frequency. However, the sky appears to be limit with Skylakes memory controller which we have just shown can handle DDR4-4000 speeds without an issue. It can very likely operate at even greater speeds, G.Skill say they have it working flawlessly at 4333MHz now.
 
Very few reviews show any real gain in handbreak I cant help but think something in the test system is exaggerating the results is it just that you run a different handbrake test to most? or is something else wrong maybe ram sub timings?

for example here it was found that g.skill loosened sub timings on a 3200 c14 kit making it no quicker than 3200c16
g.skill provided some sub timings to tweak in a attempt to improve results and it did
http://forums.anandtech.com/showthread.php?t=2472220

It would be good to know the exact timings\sub timings used at each frequency as these can have a huge impact on performance but you are overlooking them

Thanks
 
DDR4 at 3000 MHz (15-17-17-35) bought in December 2015 down over 25% in price when I looked yesterday.
higher and higher speeds available, gotta love the trickle down effect
 
I'm surprised by these results. I would have bet that once the game is loaded into RAM, any speed for that matter, that it was all GPU all day to determine if you're going to get any more FPS. I'm anticipating a new system build this year, and these results have given me something to consider.
 
I would have bet that once the game is loaded into RAM, any speed for that matter, that it was all GPU all day to determine if you're going to get any more FPS.
in many games the gpu is all that maters
this review shows that the difference in the witcher drops significantly once they go from 980ti sli to a single 980ti

some parts of games are heavier on the cpu than others
so while much of a game may be gpu limited another part like novigrad city in witcher 3 may drop below 60fps due to the cpu

ram reviews should really compare different speeds of cpu against the ram so you can see just how much your getting
but as a rough guide a 3000mhz ram kit combined with a 4.5ghz 6700k could perform similar to a 5ghz 6700k with 2133 ram in games

cpu have a very small cache which is significantly faster than ram so whenever they need need data not in there cache they have to wait on the ram which is why there can be large performance gains in cpu performance from faster ram
 
This test was made with very powerfull CPU and GPUs, but only with 8GB of RAM. I think there might be a chance, that in benchmarks, the sysmtem runs out of memory and than faster memory can give some more FPS. Maybe if the test was made with at least 16GB of RAM, than DDR-2133 can give same FPS like DDR-4000.
Its very rare for Photoshop to use more than 8GB of RAM, Handbreak never will, and I've yet to see any game use more than 8GB. I personally have 16GB in my system as one day we'll breach the 8GB barrier, but today is not that day.
There are games that use all of the 8gb of memory, even if you don't multitask in the background. Might be an inefficient engine/game at fault or a memory leak, but it does happen.
 
It would be interesting to know if this kind of scaling happens with Zen or if it's unique to Intel's arch.
 
Hi all !

I got a 5820K@4,4Ghz and a RAM kit ready for 3600Mhz but because of my proc OC my RAM is stuck at 2666Mhz....

How did you guy succed in having a 4,5Ghzproc OC AND a RAM kit set to 4000Mhz ?

Thanks for futur answers :)
 
Hi all !

I got a 5820K@4,4Ghz and a RAM kit ready for 3600Mhz but because of my proc OC my RAM is stuck at 2666Mhz....

How did you guy succed in having a 4,5Ghzproc OC AND a RAM kit set to 4000Mhz ?

Thanks for futur answers :)

The LGA2011-3 (Haswell-E) processors to my knowledge can't run at those memory speeds.
 
Kinda pointless comparison when you aren't changing the latencies. you said it didn't affect it much but still.
 
Back