Skylake gaming performance scaling with DDR4 RAM frequency.

[Frosty11]

Guys at Digital Foundry tested the i3 6100 with two frequencies of DDR4 RAM. One test was done with a DDR4 2133 MHz RAM and the other one with DDR4 2666 Mhz RAM. Their testing methodology seems very sound to me considering that they used a Titan X to remove any GPU bottlenecks and used Ultra settings to increase CPU workload. These were the results:

 

As you can see using higher frequency RAM gives a FPS boost of around 7-10%. And in Ryse the performance almost doubled with faster RAM.

 

But, they did not test the i5 6600k and i7 6700k in the same way to see if they show a similar performance boost when paired with higher frequency RAM. Do you guys think that the Skylake i5 and i7 processors will show similar scaling in gaming performance when paired with higher frequency RAM?

 

If the performance does scale for Skylake i5 and i7 processors then is this not a very good reason to buy an i7 6700K over the i7 4790K when building a new system?

 

 

Skylake vs Devil's Canyon:

 

          Many Skylake vs Devil's Canyon benchmarks done by reputable websites were done with a 2133 MHz DDR4 RAM for the Skylake CPU. Was this the reason that there was no increase in gaming performance seen with Skylake? Here are the gaming benchmark results for i7 6700K with DDR4 2666 Mhz RAM vs i7 4790K with 2133 MHz RAM.

 

System configuration

 

Benchmarks

 

 

 

 

Here we can see how the i7 6700K performs better than the i7 4790K across all tests.

 

 

Fallout 4 performance scaling with RAM frequency (Tested on Haswell Refresh i5 4690K):

 

          In another test done by Digital Foundry on Fallout 4 they found out that there is a significant increase in performance with higer frequency RAM. Benchmarks:

http://cdn.wccftech.com/wp-content/uploads/2015/11/fo4_ram_test.jpg

 

 

Final thoughts:

 

          RAM frequency hasn't mattered much up till now. But, it seems like games have started to benefit from higher frequency RAM. Considering that someone buying a i7 4790K would want to use the system for quite some time before an upgrade. And they wouldn't want their gaming performance to be held back by memory frequency. So, would you guys recommend spending a little extra and getting a Skylake platform with high frequency DDR4 RAM?

 

Sources:

i3 6100 benchmarks: http://www.eurogamer.net/articles/digitalfoundry-2015-intel-core-i3-6100-review

Skylake vs Devil's Canyon: https://www.youtube.com/watch?v=GVb3joMhtsI

Fallout 4 benchmarks: http://wccftech.com/fallout-4-performance-heavily-influenced-by-ram-speed-according-to-report/

[Gofspar]

looks like im gucci with 3200 cl15

[Virtual01]

My devil's canyon is now filled with salt

[App4that]

This has been covered many times before. You don't need to have DDR4 to benefit from faster RAM speeds. I have a 4690k and DDR3 RAM that was advertised as being 1600MHz, I'm running it at 2200MHz at 1.64v. The jump from 1600MHz to 2200MHz was noticeable playing Fallout 4 where it mostly impacted the minimum fps, another words the increase in RAM speed helps with fps dips.

 

Skylake has it's benefits but to the average gamer using a 1080p monitor that difference is meaningless outside bragging rights with benchmarks. My 4690k and Nitro 390 max out every game I have ever played.

[thorpj]

How does RAM frequency affect the framerate? Unless there's a bottleneck where the RAM wasn't fast enough with the lower speed RAM (which cannot be the case, it's high speed DDR4), i don't see how that could happen.

 

Perhaps shaders are stored on the RAM?

[App4that]

How does RAM frequency affect the framerate? Unless there's a bottleneck where the RAM wasn't fast enough with the lower speed RAM (which cannot be the case, it's high speed DDR4), i don't see how that could happen.

 

Perhaps shaders are stored on the RAM?

Draw calls, and yes a RAM bottleneck. I could stand in a spot in Lexington while playing Fallout 4 and have 60% GPU and 50%CPU usage while getting under 40fps. The switch to 2200MHz RAM speed raised the usage on the GPU and my fps to the mid 40s, usually 45fps. That's the difference between noticing the dip and not noticing it.

[Darel321]

I can sense something fishy here

[thorpj]

Draw calls, and yes a RAM bottleneck. I could stand in a spot in Lexington while playing Fallout 4 and have 60% GPU and 50%CPU usage while getting under 40fps. The switch to 2200MHz RAM speed raised the usage on the GPU and my fps to the mid 40s, usually 45fps. That's the difference between noticing the dip and not noticing it.

Wow. I never thought RAM could ever be a bottleneck unless you didn't have enough of it, or it was *really* slow.

 

The common wisdom was usually that 1600mhz (DDR3) is perfectly fine. As such i thought this rush for super fast RAM was a waste of time (especially considering that the cas latency goes up as you increase the frequency).

 

Thanks for straightening things out.

[BrightCandle]

We had the same sort of scaling appear on previous generations but only in certain games, Arma 3 for example after its released showed considerable gains from increases in RAM frequency.

 

What this says about the game and its data don't fit well into the cache of the CPU and that large amounts of additional memory bandwidth help the games to perform. Considering how well the 5820 does compared to the 4790k that doesn't surprise me, I have suspected for a while that the performance benefits of the 6 cores (including the older ones like the 3930k) is based quite a lot on the increased memory bandwidth the extra cache provides. Extra memory bandwidth and extra cache are becoming more and more important to CPU performance as the gap between the components widens.

[App4that]

Wow. I never thought RAM could ever be a bottleneck unless you didn't have enough of it, or it was *really* slow.

 

The common wisdom was usually that 1600mhz (DDR3) is perfectly fine. As such i thought this rush for super fast RAM was a waste of time (especially considering that the cas latency goes up as you increase the frequency).

 

Thanks for straightening things out.

The cool thing is if you have a Z97 MOBO, the chances are you can overclock your RAM and have the benefit of it for free. Only side effect I have noticed is it runs a little hotter, but I have a fan on it now and the temps are well under control.

[riklaunim]

Some benchmarks use frequency/latency index to grade performance. 1600MHz/CL9 would have similar index to 2133MHz/CL12 - and CL12 DDR4 isn't cheap. Second thing is memory throughput, which can be higher on better DDR4 memory.

[thorpj]

The cool thing is if you have a Z97 MOBO, the chances are you can overclock your RAM and have the benefit of it for free. Only side effect I have noticed is it runs a little hotter, but I have a fan on it now and the temps are well under control.

This is true for any motherboard that can overclock a cpu, isn't it.

 

I wouldn't imagine that it's exclusive to Z97.

[airdeano]

didn't i see corsair do a similar comparison and mostly all called BS and marketing?

[App4that]

This is true for any motherboard that can overclock a cpu, isn't it.

I wouldn't imagine that it's exclusive to Z97.

I can only speak for the Z97, as that's what I have. I've heard mixed reports on other board types. But if you have a multiplyer in the bios you're golden. And ready to spin the silicon lottery wheel lol.

[thorpj]

I can only speak for the Z97, as that's what I have. I've heard mixed reports on other board types. But if you have a multiplyer in the bios you're golden. And ready to spin the silicon lottery wheel lol.

Ah okay, i thought so.

 

I will OC at some point. Right after i get a decent case and CPU cooler (since i will OC that as well)

[Frosty11]

Draw calls, and yes a RAM bottleneck. I could stand in a spot in Lexington while playing Fallout 4 and have 60% GPU and 50%CPU usage while getting under 40fps. The switch to 2200MHz RAM speed raised the usage on the GPU and my fps to the mid 40s, usually 45fps. That's the difference between noticing the dip and not noticing it.

 

 

We had the same sort of scaling appear on previous generations but only in certain games, Arma 3 for example after its released showed considerable gains from increases in RAM frequency.

 

What this says about the game and its data don't fit well into the cache of the CPU and that large amounts of additional memory bandwidth help the games to perform. Considering how well the 5820 does compared to the 4790k that doesn't surprise me, I have suspected for a while that the performance benefits of the 6 cores (including the older ones like the 3930k) is based quite a lot on the increased memory bandwidth the extra cache provides. Extra memory bandwidth and extra cache are becoming more and more important to CPU performance as the gap between the components widens.

 

 

Some benchmarks use frequency/latency index to grade performance. 1600MHz/CL9 would have similar index to 2133MHz/CL12 - and CL12 DDR4 isn't cheap. Second thing is memory throughput, which can be higher on better DDR4 memory.

Increased frequency means increased bandwidth right? Is this what helps gain the FPS boost? If this is the case then DDR4 can provide more bandwidth than DDR3. So, there could be situations where Skylake with DDR4 could come on top with more bandwidth.

[App4that]

Ah okay, i thought so.

I will OC at some point. Right after i get a decent case and CPU cooler (since i will OC that as well)

Overclock allz the thingz!!! Last thing on my list is the monitor. Then all the things are Overclocked.

[BrightCandle]

Increased frequency means increased bandwidth right? Is this what helps gain the FPS boost? If this is the case then DDR4 can provide more bandwidth than DDR3. So, there could be situations where Skylake with DDR4 could come on top with more bandwidth.

 

Yes. Yes.

[riklaunim]

Increased frequency means increased bandwidth right? Is this what helps gain the FPS boost? If this is the case then DDR4 can provide more bandwidth than DDR3. So, there could be situations where Skylake with DDR4 could come on top with more bandwidth.

 

Benchmarks are needed. Linus DDR3vsDDR4 RAM on Broadwell show little to no difference. Over the weekend I'll try to compare 1333 MHz DDR3 CL9 vs 2133 MHZ DD4 CL14. freq/cl ratio is nearly identical for both of them and I'm curious what will be the difference (on the same mobo/CPU )

[Frosty11]

Benchmarks are needed. Linus DDR3vsDDR4 RAM on Broadwell show little to no difference. Over the weekend I'll try to compare 1333 MHz DDR3 CL9 vs 2133 MHZ DD4 CL14. freq/cl ratio is nearly identical for both of them and I'm curious what will be the difference (on the same mobo/CPU )

If you are on Skylake then make sure that your RAM is DDR3L and not DDR3 because Intel has stated that there is no official support for DDR3. Prolonged use of DDR3 with Skylake can damage the IMC in the Skylake CPU. Skylake officially supports only DDR3L and DDR4.

Source: http://wccftech.com/skylake-does-not-support-ddr3-damage-ddr3l-only/

[riklaunim]

If you are on Skylake then make sure that your RAM is DDR3L and not DDR3 because Intel has stated that there is no official support for DDR3. Prolonged use of DDR3 with Skylake can damage the IMC in the Skylake CPU. Skylake officially supports only DDR3L and DDR4.

Source: http://wccftech.com/skylake-does-not-support-ddr3-damage-ddr3l-only/

 

I know. My primary DDR3 stick can do L or no-L mode, but I'm bit mixed about that support as motherboard makers support it officially and if so they should make it actually safe.

[Frosty11]

I know. My primary DDR3 stick can do L or no-L mode, but I'm bit mixed about that support as motherboard makers support it officially and if so they should make it actually safe.

So your mobo should ideally choose the DDR3L mode, right? Maybe you should shoot a mail to your mobo customer support to confirm that it uses the DDR3L mode of your RAM and not the DDR3 mode.

BTW does you mobo support RAM OC?

[riklaunim]

it's this one: http://www.asrock.com/mb/Intel/B150M%20Combo-G/

 

- 2 x DDR3/DDR3L DIMM Slots
- Supports DDR3/DDR3L 1866(OC)/1600/1333/1066 non-ECC, un-buffered memory