Intel x'Lake Mobile Platform Memory Overclocking

Zucker2k

Senior member
Feb 15, 2006
751
59
136
#1
So, we all know, starting with Skylake in late 2015, Intel locked down its mobile platform to any form of overclocking, except for multiplier unlocked mobile chips with the HK suffix. The mainstream Haswell (4th gen) platform provided 2 extra turbo bins over the base turbo frequency but this was lost with Skylake (and Broadwell before it). Those of us who jumped on board the Mobile Skylake platform soon found out the 3.1GHz base turbo frequency of the upper tier, locked, 6700HQ was nothing to sneeze at and its 45w TDP pushed the thermal limits on many of the gaming platforms it was mostly paired with. With no way of overclocking the cpu, coupled with the thermal limitations of the mobile platform, most enthusiasts sought what little extra performance they could glean from the platform by undervolting the cpu in order to prevent thermal throttling under heavy gaming loads.
Now, overclocking typically involves cpu, gpu, and ram. On the 6th gen mobile platform, only gpu overclocking was widely attempted by enthusiasts with any amount of success. The cpu side was a dead end, since the locked chips had their multipliers fused off, and bclk multipliers were not exposed in bios.

That left RAM.

The main advantage RAM has going for it is that the platform allows RAM to be replaced. What this means is that unlike the cpu, the memory subsection of the platform must necessarily be flexible to accommodate any replacements down the line when the need arises. This flexibility works to the savvy overclocker's advantage because it means this area of the bios is not locked down on most platforms. Now, there are those who will argue that overclocking RAM is not worth it; the risk is too great and the reward too little. I agree 100%. However, because the Skylake Mobile platform officially supports a paltry DDR4 2133 MHz (and that's what mobile devices were shipped with), I find that overclocking my RAM results in a substantial boost in overall performance and responsiveness and especially in many benchmarks. Also, on a thermally constrained platform where cpu overclocking is nil, RAM speed helps to wring significant ipc overhead which would otherwise go to waste in bandwidth constrained scenarios. Finally, with hexacores and octocores being released on the mobile platforms, this post may offer an insight as to just how much bandwidth limited these platforms may be, with their DDR4 2400 MHz, and DDR4 2666 MHz RAMs respectively.

This post is not a guide, but rather to show how much the 2133 MHz RAM bottlenecks the 6th gen mobile quadcore processors, and implications for the hexacore and octocore platforms. Over the next few days I'll update this OP with stock and overclocked benchmark comparisons.

6700HQ

HWInfo showing memory overclocking is supported on this platform.
HWInfo_RAM.png

The RAM (2x8GB Hynix SoDimms)
Hynix.png

Stock AIDA64 Cachemem Benchmark (2133 MHz)
Stock.png

DDR4 2667 MHz Overclock
cachemem8.png

DDR4 2800 MHz Overclock
cachememz.png

To be continued.....

Edit: DDR4 2700 MHZ Overclock
cachemem_OC.png
 
Last edited:
Apr 27, 2000
11,488
820
126
#2
I guess you're going to show us something other than AIDA64 Cachemem benches at some point? Also, unless you adjust timings to keep latency uniform between the different clockspeed runs, you aren't really going to be showing the effects of bandwidth starvation (increased performance could come from lower latency).
 

Zucker2k

Senior member
Feb 15, 2006
751
59
136
#3
I guess you're going to show us something other than AIDA64 Cachemem benches at some point? Also, unless you adjust timings to keep latency uniform between the different clockspeed runs, you aren't really going to be showing the effects of bandwidth starvation (increased performance could come from lower latency).
Oh yes! I'll be running a few benchmarks. On the Intel platform, frequency trumps timings.
 
Apr 27, 2000
11,488
820
126
#4
On the Intel platform, frequency trumps timings.
That's not the point. If you're looking at bandwidth starvation as an issue, you have to make sure that your different runs are all at approximately the same memory latency (not timings, per se). Latency has a big impact on performance on Intel platforms, just as it does for AMD or CPUs from any other vendor. As you raise memory clockspeed, latency will go down unless you loosen timings somewhere.
 

Ed1

Senior member
Jan 8, 2001
424
1
81
#5
What is up on the 2800mhz L1, L2 and L3 results, much lower than the slower 2133 and 2667mhz.
 

Zucker2k

Senior member
Feb 15, 2006
751
59
136
#7
That's not the point. If you're looking at bandwidth starvation as an issue, you have to make sure that your different runs are all at approximately the same memory latency (not timings, per se). Latency has a big impact on performance on Intel platforms, just as it does for AMD or CPUs from any other vendor. As you raise memory clockspeed, latency will go down unless you loosen timings somewhere.
Bandwidth starvation is not the only thing overclocking could mitigate. In fact, increasing the frequency of memory is necessarily going to impact latency. This is desirable and expected, if not already implied by the very act of overclocking.
 

Zucker2k

Senior member
Feb 15, 2006
751
59
136
#8
UPDATE

I've been quite busy these past few days but I finally found time today to run some benches. I didn't run game benchmarks because I hardly play games anymore so I don't have any modern games. I did try a few synthetic gaming tests but there were hardly any significant changes in these tests.

DDR4 2800 MHZ on this first gen Hynix RAM is stable in everything except Linx (AVX2). I personally think it's a voltage thing since I couldn't tweak vccio and vddq voltages. I therefore settled for DDR4 2700 MHZ, which is my daily overclock with these settings: 13-14-14-35-312 timings @1.25v

Video Encoding: Handbrake 4K H265 Video Conversion (MKV -> MP4)
STOCK
encoded 1497 frames in 603.05s (2.48 fps), 8038.80 kb/s, Avg QP:27.29

OVERCLOCK
encoded 1497 frames in 575.30s (2.60 fps), 8038.80 kb/s, Avg QP:27.29

CB15 STOCK
CB15_Stock.png

CB15 OVERCLOCK
CB15_OC.png

CB20 STOCK
CB20_Stock.png

CB20 OVERCLOCK
CB20_OC.png

FRITZ CHESS STOCK
Fritz_Stock.png

FRITZ CHESS OVERCLOCK
Fritz_OC.png

LINX 0.65 STOCK (HT-Enabled)
LinX_Stock.png

LINX 0.65 OVERCLOCK (HT-Enabled)
LinX_OC_HT.png

LINX 0.65 STOCK (No HT)
LinXNoHT_Stock.png

LINX 0.65 OVERCLOCK (No HT)
LinX_OC.png
 

Zucker2k

Senior member
Feb 15, 2006
751
59
136
#9
Continued......

PERFORMANCE TEST 9.0 STOCK
PT_Stock.png

PERFORMANCE TEST 9.0 OVERCLOCK
PT_OC.png

INTEL XTU BENCHMARK STOCK
XTU_Stock.png

INTEL XTU BENCHMARK OVERCLOCK
XTU_OC.png

4x 32M Superpi STOCK
4xSPi32M_Stock.png

4x 32M Superpi OVERCLOCK
4xSPi32M_OC.png
 
Apr 27, 2000
11,488
820
126
#10
Bandwidth starvation is not the only thing overclocking could mitigate. In fact, increasing the frequency of memory is necessarily going to impact latency. This is desirable and expected, if not already implied by the very act of overclocking.
Correct! But you said, in your first post:

RAM speed helps to wring significant ipc overhead which would otherwise go to waste in bandwidth constrained scenarios. Finally, with hexacores and octocores being released on the mobile platforms, this post may offer an insight as to just how much bandwidth limited these platforms may be, with their DDR4 2400 MHz, and DDR4 2666 MHz RAMs respectively.
So I assumed that you were focusing only on how higher core counts could increase bandwidth requirements (leading to bandwidth starvation). Hence my commentary.

Thanks for posting some results, though.
 


ASK THE COMMUNITY

TRENDING THREADS