Triple Channel DDR3-1600 memory roundup
Apart from liquid cooling, which is my main hardware obsession, I also like to dabble with RAM from time to time.
Thanks to gnerous sponsoring from A-DATA, GeIL, G.Skill, Mushkin (EU) and OCZ I am now able to present a small memory roundup of triple-channel DDR3-1600 kits.
Here are the names and specs of the different kits in this roundup:
http://www.abload.de/img/01dataersb.jpg
Next, I would like to let images do the talking and present the different memory kits in a series of pictures:
A-DATA XPG 1600+
http://www.abload.de/thumb/03q6w4.jpg http://www.abload.de/thumb/04d7hg.jpg http://www.abload.de/thumb/05c6ab.jpg http://www.abload.de/thumb/08p5mx.jpg http://www.abload.de/thumb/10d56r.jpg http://www.abload.de/thumb/z03w8pf.jpg
GeIL PC3-12800 Ultra
http://www.abload.de/thumb/20y2up.jpg http://www.abload.de/thumb/23m3qq.jpg http://www.abload.de/thumb/2568md.jpg http://www.abload.de/thumb/26copyw4hp.jpg http://www.abload.de/thumb/28o86h.jpg http://www.abload.de/thumb/29y4t3.jpg
G.Skill PI PC3-12800U
http://www.abload.de/thumb/6187cz.jpg http://www.abload.de/thumb/62a2ok.jpg http://www.abload.de/thumb/63e0gb.jpg http://www.abload.de/thumb/64j1ec.jpg http://www.abload.de/thumb/65n717.jpg
Mushkin XP3-12800
http://www.abload.de/thumb/7034er.jpg http://www.abload.de/thumb/71a0e3.jpg http://www.abload.de/thumb/72v18j.jpg http://www.abload.de/thumb/7385iy.jpg
OCZ Blade
http://www.abload.de/thumb/42d416.jpg http://www.abload.de/thumb/43b2yp.jpg http://www.abload.de/thumb/44t7fw.jpg http://www.abload.de/thumb/45g1n3.jpg http://www.abload.de/thumb/47c2gp.jpg http://www.abload.de/thumb/48b38e7.jpg http://www.abload.de/thumb/49b85om.jpg http://www.abload.de/thumb/50c4q5.jpg http://www.abload.de/thumb/z02e6wi.jpg
Test setup
Hardware:
- Asus P6T
- Core i7 920 (standard clock)
- Western Digital 320GB HDD
Software:
- OS: Vista 64Bit
- Super Pi Mod 1.5 XS (1M)
- Everest Ultimate Edition Version 5.00
- SiSoft Sandra (15.72)
- Driverheaven Photoshop Benchmark (Link)
All of the benchmarks were re-run three times to average out random variances. All of the tests, including the overclocking, were done with CR1.
Results at stock settings
Everest Bandwidth:
http://www.abload.de/img/02everestrwbfp33.jpg
Everest Latency:
http://www.abload.de/img/03everestlat3wn5.jpg
Sandra Bandwidth:
http://www.abload.de/img/04sandrabwzo2x.jpg
Sandra Latency
http://www.abload.de/img/05sandralat9wvw.jpg
This first set of Benchmarks shows a clear tendency among the memory-kits: OCZ is ahead of the pack with the Blade kit, followed by Mushkin. The three other kits get similar results, usually with the A-DATA kit lagging slightly behind.
Super Pi Mod 1.5
http://www.abload.de/img/superpi0s8g.jpg
With the Super Pi calculation, as usual the results are extremely close together. If we look at the averaged out results from five 1M-runs per kit, there's basically no significant difference left. Interestingly though, the A-DATA kit managed to get below 15.300 s in two out of five runs while none of the other kits got below 15.303.
Driverheaven PS Benchmark
http://www.abload.de/img/06ps7w4z.jpg
The Driverheaven Photoshop Benchmark simply runs a series of complex filters over a very huge (~190MB) image file. Photoshop can time how long this takes, so you get fairly precise results from this. The benchmark is mainly interesting because it's completely un-synthetic. It's unlike other benchmarks in that it simply measures a actual use of hardware rather than using some script specifically designed to test memory.
Here, we see something of a role-reversal with the A-DATA kit finishing first and the OCZ Blade trailing behind.
Overclocking
I do two different overclocking tests.
1. Timings
Sticking to stock frequency of 800MHz, I lower the timings to the minimum value that is still benchmark-stable. Here are the timings achieved:
http://www.abload.de/img/07timingw1zx.jpg
Here's an idea of what performance gains this leads to:
Everest Read
http://www.abload.de/img/07everestocbsrp.jpg
While the OCZ's performance doesn't improve, this is simply because the timings were really good to begin with. It can defend it's pole position. The GeIL kit gains a whopping 741 MB/s in this benchmark, thanks to it's lowered timings, and can secure a second plase behind the OCZ Blade. The G.Skill kit gains 540 MB/s and the A-DATA kit gains 534 MB/s.
Everest Latecy
http://www.abload.de/img/08everestoclatd721.jpg
In this benchmark, the GeIL kit really shines thanks to it's low timings.
Sandra Latency
http://www.abload.de/img/09sandralatoct3ra.jpg
In Sandra's latency test, the GeIL kit is once again the best performer, followed by the OCZ blade kit.
2. Maximum frequency
For this test, I set timings to 8-8-8-26, VRAM to 1.7V and QPI to 1.45V and see how far I can push the frequency while remaining benchmark stable.
Results:
http://www.abload.de/img/10ocfreqjvu3.jpg
The kits by GeIL, OCZ and A-DATA really shine in this test. I was surprised at how far the frequency could be pushed for each one. That one MHz difference between the top two is simply because the clocks were achieved with a different combination of BCLK and multiplier. Obviously, that one MHz doesn't mean more performance for the GeIL kit.
The only disapoinment here ist the G.Skill kit which didn't really want to leave it's stock frequency.
Benchmark results for overclocked memory aren't shown because these results are skewed due to different CPU clocks resulting from the BCLK and multiplier combinations.
IMPORTANT NOTE ON OVERCLOCKING RESULTS:
I'm assuming you know that overclocking voids your warranty and can damage your system etc. etc. so this note is about something else. Note that I only go for benchmark-stability in these tests and also that I limit voltages to relatively safe levels.
I'm sure experienced overclockers using crazy voltages can push these kits to higher clocks and I also assume that most of these kits won't be 24/7 stable at the settings I ran them at.
Conclusion
Most of the kits tested have strengths as well as weaknesses. Take the OCZ Blade kit for example: While it dominates almost everywhere, it does fall behind in the Photoshop-benchmark and it comes at a very high price. So this memory is great for anyone looking for stylish, high-performance memory but not an option for limited budgets.
The A-DATA kit, on the other hand, finishes last or next to last in most benchmarks, but delivers great performance in the Photoshop benchmark and Super Pi. Also, it allows for very low timings and great overclocks and gains lots of performance that way.
The GeIL kit is similar, as it really starts to shine once you adjust the timings and it's overclocking capability is simply amazing.
The Mushkin memory is somewhere in between. It comes with more performance than average out of the box but doesn't do as well as the above mentioned kits when it comes to overclocking.
The G.Skill kit is the only one that doesn't really stand out in any way. It's never particularly good nor particularly bad, though it doesn't overclock well at all. Ultimately, it only has a reasonable price and very nice heatspreader going for it.
Other than for overclocking and benchmarking, memory performance will not be very noticable in everyday use, so for many, picking memory based on how pretty the heatspreader is and what the price tag says, is probably a reasonable choice.
Hope you enjoyed this report.
If you like, check out the original, German article on DeXgo.com.
Cheers,
Shane
