Skylake 6700k appears to be scaling significantly with RAMspeed in dedicated gaming

[Moejoe]

A German website has done Skylake tests with different ramspeeds and found major increases to dedicated gaming performance (they used a 980ti).

 

Seems like spending a bit more for memory could be important once again.

 

Translation by me:

The test results of the i7 6700k with significantly reduced RAM clock from 1.600 Mhz instead of 3000 Mhz are remarkable. While experience showed, that RAM clockspeed didn't have a large influence on FPS, the reduction in clockspeed lead to a 13% drop in FPS for the 6700k. 

This indicates that the Skylake architecture profits significantly better from higher data throughput than its predecessors. It should be noted, that 1600 MHz is a very low clockspeed für DDR4 - even the cheapest modules clock at around 2133 Mhz. We will probably take a closer look at the importance of RAM clockspeed in a seperate article. In the meantime we recommend to shop for DDR4 ram with higher clockspeeds, especially given that the price difference compared to the lower clocked ones is very small.

 

Source: http://www.gamestar.de/hardware/prozessoren/intel-core-i7-6700k/test/intel_core_i7_6700k,924,3234508,3.html#spielebenchmarks

[Bob Jim]

Wow... that is annoying, that means I cannot just go get cheap DDR4-2133 Mhz...

[JefferyD90]

with ddr4, the timings are more spaced out, meaning that when you reduce speeds it will have a larger impact. 1600mhz DDR4 is SIGNIFICANTLY slower than 1600mhz ddr3 ram.

[Lays]

Wow... that is annoying, that means I cannot just go get cheap DDR4-2133 Mhz...

 

3000 ddr4 really isn't all that much more

[GM_Peka]

Isn't 1600mhz extremely low for DDR4 though? You'd probably see big scaling on a 4790k going from 800 mhz to 1866.

[GM_Peka]

Wow... that is annoying, that means I cannot just go get cheap DDR4-2133 Mhz...

 Well getting high frequency RAM doesn't mean getting a better/more expensive bin, you'll just lose out in timings a bit, which matters a lot less than frequency, I've seen the corsair's 3200mhz kits hardly cost much more that there slower kits.

[Moejoe]

Anandtech had this to say on RAM clockspeeds:

 

Conclusions on Gaming, DDR3 vs DDR4

Focusing on purely the integrated graphics, DDR4 is the clear winner with an average 7.0% gain in average frame rates. We also took stock of the minimum frame rates too, and they went up 5.8% as well. So for integrated graphics, moving to even the base DDR4 memory kit wins out.

For discrete graphics card testing, only three differences stand out here. For GRID on the R7 240 the DDR4 set loses by 3.2%, but for the GTX 770 the DDR4 wins on Mordor by 6.4% and on GRID by 2.3%. All other differences are below 2%, mostly on the side of DDR4.

 

And as far as i can tell they were testing DDR3 1600 vs DDR4 2133. So i think this at least warrents further investigation.

 

Edit: Source: http://anandtech.com/show/9483/intel-skylake-review-6700k-6600k-ddr4-ddr3-ipc-6th-generation/8

[Lays]

 Well getting high frequency RAM doesn't mean getting a better/more expensive bin, you'll just lose out in timings a bit, which matters a lot less than frequency, I've seen the corsair's 3200mhz kits hardly cost much more that there slower kits.

 

Actually it is getting a higher bin, most of the DDR4 kits actually have extremely similar timings.  (There is a reason the value-oriented DDR4 looks low end and has C14/C15 2133 speeds, whilst other kits at same volts are doing Cl14 2800)

 

I bet a vast majority of 4x4GB 2133 kits are CL14-15, whilst you can easily get 3000 CL15 for only ~40-50$ more.  (In the US atleast)

 

This is why a lot of the overclocking nuts on X99 are getting the CL13 2666 kits, CL14 2800 or CL15 3000/3200 kits, they usually scale well and have Hynix IC's, which are very good for benching and seem to scale decently with volts & lower timings.

Seen a few guys that bought the 2666 c13 kits get C11 3000 with more volts added, quite impressive!

[DarkSmith2]

This isnt actually new. Happened to the 5820k scaling before. (10fps differences could be measured from 2133 to 3000)

 

But as dual channel ofc, the scaling kicks in harder.

[Moejoe]

http://anandtech.com/show/8959/ddr4-haswell-e-scaling-review-2133-to-3200-with-gskill-corsair-adata-and-crucial/7

 

Scaling was very situational there. (only one result had a 10% boost)

 

http://anandtech.com/show/7364/memory-scaling-on-haswell/7

 

And with the 4700k we had no scaling at all, even though it was dual channel. (at least once you got higher than 1600 MHz)

 

 

So i think this is a pretty significant change. Hey Linus you could do a video about this! *nudge nudge*

[boboman342]

If we could just have a single gpu that can keep 144fps at 4k instead of focusing on cpu horsepower... yeaaaaah that would be great.

[qwertywarrior]

at sub 60 fps

u will probably see a 1 fps difference

[WereCat]

We already knew that BF4 benefits from faster memory... this is not news. You wont notice a thing in 99% of other games.

[Ivo]

ahh did not read the post, nvm

[Moejoe]

We already knew that BF4 benefits from faster memory... this is not news. You wont notice a thing in 99% of other games.

 

Look at the sourcelink there are more benchmarks there and it appears to be quite consistent. (GTA V, LoL and Witcher 3)

[TheMissxu]

There are some kits that aren't crazily expensive

 

($50 vs $70) I suppose that is ~50% more, but at least they a reasonable price for a comfortably functioning PC

[Sauron]

Guys, ddr4-1600 is slower than ddr3-1333...

[TheMissxu]

Guys, ddr4-1600 is slower than ddr3-1333...

ddr4 2133 is the slowest you can go...

 

and you should review how CAS latency works (it's based on the frequency of the memory, so more frequency = lower nano-seconds/CAS)

[Kyuubixchidori]

190 to 198 is a tiny jump. looking at what? 4%? 

 

so at 60fps we are talking about 2fps. 

 

 

edit: looked at graph wrong

[SurvivorNVL]

Cl14, 2800mhz, here I come!

[Sauron]

ddr4 2133 is the slowest you can go...

 

and you should review how CAS latency works (it's based on the frequency of the memory, so more frequency = lower nano-seconds/CAS)

 

check the diagram, it says ddr4-1600 - presumably underclocked by them, meaning they didn't change latency or they would have specified it. In general, current ddr4 has pretty disappointing latencies

[othertomperson]

If we could just have a single gpu that can keep 144fps at 4k instead of focusing on cpu horsepower... yeaaaaah that would be great.

 

Until there's both a CPU that can consistently maintain 144fps across all games and a display standard that allows transmission of 4k above 60 hz that is a completely pointless thing to care about.

[EChondo]

Eh, not that huge of an increase, noticeable, but not huge.

 

Cool though.

[boboman342]

Until there's both a CPU that can consistently maintain 144fps across all games and a display standard that allows transmission of 4k above 60 hz that is a completely pointless thing to care about.

That's what I am saying, we need more development on the gpu/display side vs cpu!

[othertomperson]

That's what I am saying, we need more development on the gpu/display side vs cpu!

 

I wouldn't say that. As far as high refresh 1080p goes we definitely need faster CPUs than we have currently. I'm expecting Directx 12 to help give pre-existing CPUs like the 5960X a massive boost in performance, but that remains to be seen. As far as 1440p high refresh goes, we need both but the 980 Ti is still monstrous if you consider that its performance is on par with two 780 Tis in SLI.