DDR3L Memory Speed on Haswell Mobile

We've recently found that Haswell responds very well to high speed memory, at least in the desktop space. Our own testing has been independently corroborated by work at Xbit Labs and AnandTech, and we continue to be confident that on high end Haswell processors, faster memory is the way to go. DDR3-1600 just doesn't cut it on that front anymore.

But what about notebooks? High speed notebook memory is still a fickle beast; desktops have the luxury of being able to enable XMP timings, but notebook memory has to make do with some SPD trickery to get past DDR3-1600. Notebook CPUs also don’t have the raw performance headroom their desktop counterparts do; an i7-4770K, even at stock speeds, will turbo to 3.7GHz on all four cores and more than likely stay there as long as needed. The mobile i7-4900MQ, on the other hand, only guarantees 2.8GHz. While it can hit 3.6GHz on all four cores, the reality is that it spends much of its life closer to its nominal clock speed.

With that in mind, we decided to test our new MSI whitebook (complete with the i7-4900MQ) at DDR3-1333 CAS 9, DDR3-1600 CAS 9, and DDR3-1866 CAS 11. The whitebook sports the aforementioned Intel Core i7-4900MQ along with an NVIDIA GeForce GTX 880M and all solid state storage, making it a best case scenario for high speed memory. Memory was tested in 2x8GB kits.

I’m keen on producing real world situations for these benchmarks as well as isolating situations where faster memory will and won’t help, and with that in mind I’ve run the following tests:

• 3DMark Fire Strike
• Adobe Lightroom 5.3 64-bit
• Adobe Premiere CC (HDV encode)
• Adobe Premiere CC (H.264 encode)
• BioShock Infinite
• Company of Heroes 2 (maximum settings, no AA)
• Final Fantasy XIV: A Realm Reborn (tested on the HD 4600 IGP at 768p High and the GeForce GTX 880M at 1080p Maximum)
• Handbrake 0.9.9 64-bit
• SPECapc Lightwave 9.6
• StarCraft II

It’s a pretty exhaustive list. Pay attention to Final Fantasy XIV in particular; MMOs have a different performance profile than games like BioShock Infinite, stressing the CPU and memory much more than is typical. I also went ahead and tested it on the IGP separately, as MMOs are also more likely to get played on integrated graphics than monsters like BioShock Infinite.

BENCHMARK	1333 C9	1600 C9	% faster than 1333	1866 C11	% faster than 1333
3DMark Fire Strike	4498	4519	0.5%	4510	0.3%
Adobe Lightroom 5.3 (in sec)	100	97	3%	95	5%
Adobe Premiere CC (HDV) (in sec)	798	775	2.9%	762	4.5%
Adobe Premiere CC (H.264) (in sec)	702	681	3%	672	4.3%
BioShock Infinite (Avg)	55.9	56.2	0.6%	56.7	1.4%
BioShock Infinite (Min)	10.2	10.3	1.3%	12.6	24.4%
Company of Heroes 2	38.5	38.3	-0.5%	40.5	5%
Final Fantasy XIV (HD 4600)	23.2	26.8	15.5%	28.7	23.5%
Final Fantasy XIV (GTX 880M)	66.8	68.5	2.6%	71.1	6.5%
Handbrake 0.9.9 x64 (High Profile) (in sec)	3090	3062	0.9%	3017	2.4%
SPECapc Lightwave 9.6 (Render) (in sec)	630	622	1.3%	620	1.6%
SPECapc Lightwave 9.6 (Interactive) (in sec)	600	600	0%	588	2%
SPECapc Lightwave 9.6 (Multitask) (in sec)	1024	1015	0.9%	1012	1.2%
StarCraft II (Avg)	71.7	75.2	4.9%	74.7	4.1%
StarCraft II (Min)	42	44	4.8%	45	7.1%

Predictably, 3DMark shows essentially no performance difference. GPU limited situations like BioShock Infinite and, surprisingly, Company of Heroes 2 don’t really benefit from the higher memory speed. The primary benefactors of the faster memory speed continue to be Adobe’s suite of software, but Final Fantasy XIV and StarCraft II get a nudge upward in performance as well.

It wouldn’t be completely unreasonable to blame the modest gains going from 1600 to 1866 on the substantially higher CAS latency, but really there are likely a few different things going on here. In addition to the latency, it’s harder to get high memory speeds stable in a notebook platform (preventing us from using DDR3-2133 on this particular sample), and the performance envelope is much smaller in general. On the desktop, increasing the clock speed on the CPU would increase the impact memory speed had on performance. Therefore, it stands to reason that at the reduced clocks of a notebook CPU, memory speed would become much more negligible.

With all that in mind, running high speed memory in your notebook isn’t a total wash. If you play more CPU-centric games like MOBAs and MMOs, or if you’re pinned to the integrated graphics, faster memory is a way to extract more performance from the system. In many notebooks this might even be the only means of increasing performance, since you’re unlikely to be able to upgrade the CPU or graphics hardware.

While a powerful desktop system can benefit substantially from an upgrade to its memory speed, notebooks are going to be less exciting candidates. The flip-side is that, as I said, upgrading memory might also be the only way to get more performance out of your notebook. If you need it, this is a way to go. But if you’re already running DDR3-1600 and happy with your system, you can reasonably pass on faster kit for the time being.