Skylake 2400MHz VS 3200MHz DDR4 Memory Performance Gains

[WYP]

A question I commonly see from new system builders is how fast should your memory be, to which the answer typically depends on exactly what you are using your PC for.

If you are all about value I would typically recommend buying your memory at where the best price to performance is, which is something that is currently in a massive state of flux ATM with the arrival of mainstream DDR4, with 2400 MHz DDR4 to even 3200MHz kits being available for fairly insignificant price differences.

Below here is an example for Corsairs Dominator DDR4 kits, which have only a £10 price increase when moving from 2400MHz to 3200MHz for their 2x4GB kit.

Corsair Vengeance LPX DDR4 2x8GB 2400MHz - £49.99
Corsair Vengeance LPX DDR4 2x8GB 3200MHz - £59.99

I have recently upgraded from a Corsair 2400MHz DDR4 memory kit to a 3200MHz DDR4 memory kit, which to be honest was an upgrade which was more about capacity than MHz, but nonetheless I have decided to do some testing with both kits to see what the performance gains are like.

Rig Specifications

OS: Windows 10
CPU: Intel i7 6700K @ 4.6GHz
CPU Cooler: Corsair H110
Motherboard: ASUS ROG Z170 Maximus VIII Hero
GPU: Sapphire R9 Fury
Memory: 2x4GB of Corsair 2400MHz DDR4 Memory, 4x4GB of G.Skill Ripjaws 3200MHz
SSD: 2x 256GB Crutial MX100 (not in Raid)
HDD: Seagate 3TB HDD
Case: NZXT H440
PSU: Corsair HX 1200i 80+ Platinum

Cinebench

Cinebench is a great benchmark for looking at comparative CPU performance, so it is a great place to begin my testing.

I have tested this with my CPU at both stock clocks and at my 4.6GHz overclock. while in previous generations of Intel CPUs this would be considered a nice overclock, it is not really that large of a learp from Intel's stock boost clock of 4.2GHz

Stock 2400MHz - 922
Stock 3200MHz - 933
4.6GHz 2400MHz - 1004
4.6GHz 3200MHz - 1011

Above I have also linked screenshots which validate these results.

As you can see here the performance gains from changing my memory have been minimal, though any extra performance here is good.

wPrime

In wPrime I expected to see a similar story to Cinebench, but this time we actually see different results when using the PC at stock and overclocked states.

At stock we can see a decrease in performance and while overclocked we can see a gain in performance when using faster memory.

Stock 2400MHz - 32M - 5.846s, 1024M - 178.656s
Stock 3200MHz - 32M - 5.903s, 1024M - 181.121s
4.6GHz 2400MHz - 32M - 5.354s, 1024M - 165.563s

4.6GHz 3200MHz - 32M - 5.333s, 1024M - 163.063s

3DMARK Sky Diver

Using 3DMARK Cloud gate I was actually very surprised at how much impact that memory speed had. The increase of memory speed from 2400MHz to 3200MHz almost had as much impact as my 4.6GHz overclock, which is simply fantastic to see.

This has made me wonder about some actual game performance, so I may do some more tests tomorrow.

Stock 2400MHz - 30370
Stock 3200MHz - 31100
4.6GHz 2400MHz - 31475
4.6GHz 3200MHz - 32091

3DMARK Fire Strike

Sadly our impressive performance increases in Sky Diver didn't translate to high score gains in 3DMARK Fire Strike, but this is due o the simple fact that 3DMARK Fire Strike is very GPU oriented, whereas Sky Diver is much more dependent on your whole system configuration.

Again we see increases in our scores from moving to 3200MHz memory, but sadly we see increases of around 20-30 points rather than the hundreds of points that we had in Sky Diver.

Stock 2400MHz - 12637
Stock 3200MHz - 12652
4.6GHz 2400MHz - 12758
4.6GHz 3200MHz - 12785

All in all I am very happy with my move to 3200MHz DDR4 from 2400MHz. While the performance gains are not the largest in the world we can definitely see that they have a positive impact on performance in most cases.

With the large score increases that I have seen in 3DMARK Sky Diver it has made me really want to do some more testing on this, so expect an update in the future where I look at a few games and possibly a few more CPU benchmarks.

 

[Kushiro]

Someones taking their writing to the next level

Thanks for the read.

 

[realneil]

It is a good read. Thanks

 

[NeverBackDown]

Great write up! I'm pretty interested to see gaming benches and if any differences occur. I think GPU limited scenarios will show less of a benefit however whereas I think in CPU limited games it could show a slightly bigger benefit as it is able to talk to the CPU faster.

 

[Excalabur50]

Good write up and interesting results!

 

[WillSK]

Really interesting to see dude. Nice one for putting this together

 

[Mysterae]

Below here is an example for Corsairs Dominator DDR4 kits, which have only a £10 price increase when moving from 2400MHz to 3200MHz for their 2x4GB kit.

Corsair Vengeance LPX DDR4 2x8GB 2400MHz - £49.99
Corsair Vengeance LPX DDR4 2x8GB 2400MHz - £59.99

You need a proof reader .

Interesting read, good to see there's benefits and not just marketing BS.

 

[WYP]

UPDATE,

I have conducted a few gaming tests to compare the two ram speeds, to see if RAM speed will have much of an impact in-game, but first of all we will look at a few benchmarks.

This time I have decided to use my CPU when overclocked at 4.6GHz for all tests in order to fit in more games.


3DMARK API Overhead Test

I will start of with 3DMARK's API overhead test, which is a test of how many draw calls can be achieved by a particular system.

What we can see here is that a greater number of draw calls are achieved by using faster memory, especially in the DX11 Multi-threaded test.

DX11 ST

2400MHz - 1503819
3200MHz - 1681045

DX11 MT

2400MHz - 1389768
3200MHz - 1681045

DX12

2400MHz - 17752774
3200MHz - 18537037

Mantle

2400MHz - 20056157
3200MHz - 21052666


3DMARK Vantage

We can see in 3DMARK Vantage that we have a boost of almost 1000 point in our P-score and scores which are almost identical for the X score.

Performance Score

2400MHz - 46391
3200MHz - 47304

Extreme Score

2400MHz - 32403
3200MHz - 32397

Game Testing

Total War Attila

Memory Type - 1080p, 1440p, 4K
2400Mhz - 36.2, 25.0, 20.4
3200MHz - 36.5, 25.1, 20.4


Sleeping Dogs

Memory Type - 1080p, 1440p, 4K
2400Mhz - 89.5, 55.6, 26.1
3200MHz - 89.6, 55.6, 26.2

Hitman Absolution

Memory Type - 1080p, 1440p, 4K
2400Mhz - 64.41, 4418, 22.51
3200MHz - 64.59, 44,27, 22.53

Shadow Of Mordor

Memory Type - 1080p, 1440p, 4K
2400Mhz - 101.29, 75.94, 44.08
3200MHz - 101.06, 76.25, 44.21

Sniper Elite

Memory Type - 1080p, 1440p, 4K
2400Mhz - 167.2, 121.4, 66.8
3200MHz - 167.3, 121.5, 66.4

Tomb Raider

Memory Type - 1080p, 1440p, 4K
2400Mhz - 105.4, 75.0 , 38.4
3200MHz - 105.5, 74.1, 38.6


Conclusion


Looking at all these games we can see that in the majority of cases that the increase in RAM speed has increased performance, though only marginally.

No game titles had any benefit from the faster RAM that would be noticeable in game as it all was less than a frame difference throughout all 6 titles, meaning that for gaming that you would be better off focusing you cash on components like your GPU rather than your system memory.

 

[NeverBackDown]

It does better in Synthetic benches than gaming, makes sense, as those benches are designed to push a system as far as it will go. I like those Mantle results in the API tests along with DX12. Nice to see they get a decent bump. Would be interested in the future to see games with DX11 and DX12 APIs and if any differences are any apparent.

Good job WYP, nice read

 

[WYP]

It does better in Synthetic benches than gaming, makes sense, as those benches are designed to push a system as far as it will go. I like those Mantle results in the API tests along with DX12. Nice to see they get a decent bump. Would be interested in the future to see games with DX11 and DX12 APIs and if any differences are any apparent.

Good job WYP, nice read

Cheers mate.

I did find the API overhead testing very interesting, as the ram speed really improved things.

It is a shame that these gains do not translate to much/any real in-game performance, but I'm more than happy with my purchase.

 

[NeverBackDown]

I would pay more for that extra speed increase simply because it's not that much more money. so I would be just as happy spending that little extra too.

I wonder why though that games don't see any performance increases worth noting. It's probably because they aren't memory bandwidth limited but more CPU limited and increasing the memory speed doesn't really help the CPU process faster.

 

[WYP]

I would pay more for that extra speed increase simply because it's not that much more money. so I would be just as happy spending that little extra too.

I wonder why though that games don't see any performance increases worth noting. It's probably because they aren't memory bandwidth limited but more CPU limited and increasing the memory speed doesn't really help the CPU process faster.

I don't think an i7 6700K at 4.6GHz would be CPU limited by much, after all it is the highest IPC CPU on the market.

The places we see the most gains are Sky Diver, Vantage P-score and in Draw calls/ API overhead tests, which are all tests which are CPU limited.

It is unsurprising that gaming has no real performance differences, but it would have been nice to have at least one exception.

 

[NeverBackDown]

I don't think an i7 6700K at 4.6GHz would be CPU limited by much, after all it is the highest IPC CPU on the market.

The places we see the most gains are Sky Diver, Vantage P-score and in Draw calls/ API overhead tests, which are all tests which are CPU limited.

It is unsurprising that gaming has no real performance differences, but it would have been nice to have at least one exception.

It is the fastest IPC CPU on the market, however games are CPU limited still as they still overload one core. That's probably why they don't really benefit from faster memory. The Synthetic are better designed for getting ever last drop out of the CPU because that is there focus, whereas games have a lot more going on and don't get that benefit. I wish it was that way, but it's not
Otherwise I would OC my memory

 

[WYP]

It is the fastest IPC CPU on the market, however games are CPU limited still as they still overload one core. That's probably why they don't really benefit from faster memory. The Synthetic are better designed for getting ever last drop out of the CPU because that is there focus, whereas games have a lot more going on and don't get that benefit. I wish it was that way, but it's not
Otherwise I would OC my memory

What we can really hope for with DX12 is that developers can try and make better use of CPU resources and their draw calls.

 

[NeverBackDown]

What we can really hope for with DX12 is that developers can try and make better use of CPU resources and their draw calls.

Yeah DX12 enables them to use more draw calls and better resource management so in theory it should mean they do make better use of it. Either way I don't think memory speed will change anything even with DX12.

 

[Cigarjohn]

From my experiences, latency tends to be important when doing heavy video editing. This is where lower latency matters. But for gaming and say the difference between 13 latency and 18 latency memory, you won't be able to tell the difference.