G.Skill was kind enough to send this kit for evaluation prior to the Sandy Bridge-E release so it could be used in the launch review. We saw how it performed at DDR3-1866 / 9-11-9-28 on Intel’s Siler board and at its rated speed & timings on the Rampage IV Extreme. Now we’ll have a look at the kit itself and run a few comparisons while we’re at it.
Product Tour & Specifications
If you saw the preview of this kit, you can skip the following non-installed photos, they were published in that piece. If not, read on!
The first thing you’ll notice is that the box is massive for a kit of memory.
It’s huge because it contains two matched sets of matched sets of memory (yes, that was intentional) with fans included.
The memory contained within is a quad-channel, 4 x 4 GB set of quite fast RAM.
- Speed: DDR3-2133 / PC3-17000
- Timings: 9-11-9-28
- Voltage: 1.65 V
- Model #: F1-17000CL9Q-16GBZHD
Aesthetics on the Ripjaws Z kits are solid and make for a good looking, giant set of RAM. Like other Ripjaws series, they have low(ish) profile heatsinks, so if you don’t use the fans you have better than even odds that your air cooler and fans will fit even on boards with DIMMs close to the socket.
Speaking of fans, because of the configuration of memory on X79 motherboards – a pair or a quartet of DIMMs on each side of the socket – if you want fans on your memory, you’ll have to have two.
Of course, like most memory that runs 1.65 V, you don’t really need fans with anything that qualifies as decent airflow. These could help with VRM air flow though, so if you do run them it certainly wouldn’t hurt. Installed, they can end up looking a little silly though. On boards with PCIe slots close to the DIMM slots, you could run into a potential issue with your GPU.
So instead I went with the more understated look sans fans. To reiterate, make sure you have adequate airflow if you chose not to use them.
They don’t exactly match the Rampage IV Extreme, but I like red and blue together. There are red RipjawsX kits, but the faster kits with tighter latencies are all blue.
Our test setup was the same as that from the Rampage IV Extreme review.
|Motherboard||ASUS Rampage IV Extreme|
|RAM||G.Skill RipjawsZ Quad-ChannelDDR3-2133 / 9-11-9-28|
|OS||Windows 7 Professional x64|
The CPU is water cooled, the GPU is on stock cooling (no 3D tests were run for this review because RAM makes very little difference in FPS) and the RAM is cooled with a fan elegantly affixed to the test bench with masking tape. The benchmark comparisons were all run at the stock CPU speed (on this board that is 3.9 GHz on all cores).
Stability at Rated Speed
Before performance testing, it’s always good to look at stability. Ultra-X has generously supplied their RAM stress tester, the R.S.T. Pro 3 PCI Express. There are only a couple of review sites that have one of these and we are proud to be one of them. It’s not your standard Memtest++ bootable CD. This is a standalone, bootable piece of hardware that plugs into a PCIe x1 slot and, man, does it ever beat on some RAM. By the time the tests finish, the sticks can get quite toasty.
Several manufacturers use this to test their memory before it goes out the door to reduce RMAs. If it passes this test, it is stable and then some. Manufacturers generally run the full memory test for three passes. We’re overclockers though and tend to abuse sticks a little more than other people, so for our purposes, five passes sounds good. Running this test on 16 GB of RAM ended up absolutely hammering the RAM for over eight hours with a load more that it will ever experience in every day use, so that should be sufficient to say whether or not they’re stable.
Indeed, they were very stable, passing with flying colors at their rated DDR3-2133 with timings of 9-11-9-28 (and a 2T command rate).
To run performance numbers, we have a bit of a varied comparison for you. This should satisfy not only those that want to know how this kit performs, but also how a) a lower-end quad channel kit should perform and b) how dual- and triple- channel RAM compare to the new quad-channel beasts.
Synthetic testing with SiSoft Sandra is going to show the biggest difference between the various testing conditions. I went with this instead of AIDA because of AIDA’s inability to saturate the quad-channel lanes with its current version. Hopefully they’ll address that in a future revision.
As you can see, there is a massive increase in bandwidth over dual channel. Interestingly (and coincidentally) we can see the bandwidth penalty when dropping a channel. Triple-channel DDR3-2133 is roughly equal to quad-channel DDR3-1600, with the faster triple-channel having a slight advantage. The same thing occurs with cache bandwidth.
Only in latency does the DDR3-1600 speed show a deficiency in favor of the faster RAM with fewer channels.
Rendering, Compression and Encoding
In everything other than the SiSoft tests, you’ll see the quad channel advantage is much less, if there is one. In the real-world tests, we’ll have a look at rendering, encoding and compression.
It seems to be a mixed bag here. Quad channel definitely helps in Cinebench R10. Testing numbers in this bench vary quite a bit, so it might take more than three runs to get a more consistent average between kits. The conclusion from that bench is that quad-channel performs better than dual- or triple-channel, but not by a ton.
Cinebench R11.5 is a lot more consistent as far as scores, and it definitely shows favoritism toward speed rather than number of channels.
Both PoV Ray and x264 Benchmark (rendering and encoding, respectively) show this kit coming in above the competing configurations, but not by very much. PoV Ray favors speed first, then channels – but all results were within a mere 1% of each other. The 2nd Pass of x264 is the actual encoding (Pass 1 is just a read through prior to encoding) and it shows a similar curve. This kit comes out on top, but again with less than 1% variation.
7zip definitely prefers speed to channels. The quad-channel kit even came in just a smidge lower than the equally fast dual- and triple-channel runs.
WPrime historically doesn’t give a rip about your RAM. I know people that claim single-channel doesn’t even make a difference; as long as your speed is in the ‘normal’ benchmarker range, that’s what matters.
In this case, WPrime 32M does seem to actually care a little bit. This kit came in first by greater than a percentage point over the slower quad-channel results and over two percent faster than the same speed in dual-channel. Interestingly, it seems to prefer channels to speed, unlike the real-world results.
WPrime 1024M shows the same results, but with less of a difference overall. These results were all within a percentage point, but if you’re counting (and we are), this kit did better than the other configurations.
SuperPi (specifically, 32M) is where I expected to see the biggest difference outside the synthetics and it didn’t disappoint. Channels beat speed and speed + channels beats all else, as it should be, with this kit coming out ahead all around.
Overclocking this kit was a fun experience. Being four massive 4GB sticks for 16GB total, I didn’t expect much headroom, but it did give up some MHz, which was a pleasant surprise. First the kit was over volted a hair to 1.675V and I saw how far it would go with rated timings, then slightly looser at 9-11-10-28, then a little looser than that at 10-11-10-28. The stopping point was that which it would no longer run the AIDA64 Cache & Memory benchmark.
Overclocking at rated timings didn’t take us very far, with a stopping point only 22 MHz over stock. Loosening up the tRP (third timing) showed a much stronger gain, reaching DDR3-2300; a gain of 83 MHz total. Not a bad overclock at all. Raising CL one got a few extra MHz, but surprisingly not as much as the tRP change.
A (very legitimate) criticism from past reviews was that I failed to run SuperPi 32M to see the fastest it would run with relatively stable conditions. The maximum, bleeding edge of stability point at which it would pass SuperPi 32M ended up being DDR3-2294 / 10-11-10-28. This is with 1.69 V applied, which is still pretty conservative.
I couldn’t stop there though and had to see what the max the system would run was. That ended up bieing DDR3-2340 / 10-11-10-28.
Now, you should note that this CPU doesn’t have the best memory controller. I have a kit of stellar G.Skill Pi DDR3-2400 from a couple years ago that I couldn’t get to go any farther than this kit did in the SuperPi 1M run above. Thus, that is literally the max the system would go, not this kit. If I had a better memory controller, I could probably have loosened up tRCD by one and shot to DDR3-2400 or more. It’s already very close at these timings.
Of course, raising timings comes with its own trade-off in slower bandwidth/latencies. So when clocking memory with higher latencies, you have to test to find out if the trade off is worth it.
Final Thoughts & Conclusion
G.Skill has a strong kit here for sure. It is perfectly stable at stock as should be expected. It even overclocks pretty well; better than expected for such a large kit. Good overclocking is to be expected from G.Skill though. You’ll need a stellar IMC, but they have pushed 64 GB of their Ripjaws Z RAM all the way to DDR3-2600 SuperPi 32M stable, which is phenomenal by anyone’s standards.
This Ripjaws Z kit won’t make it to that kind of speed, but it meets my criteria for a solid set of RAM: 1. It is absolutely rock solid stable with only needing to change voltage, speed and the first four timings, and 2. It can be tweaked and overclocked past its rated speed with little effort.
Of course, price is always a concern and quad channel memory in general isn’t going to come cheap. This kit still isn’t available retail that I can find. There is a similar kit, but without fans and slightly looser timings (9-11-10-28) that goes for $179.99. If I were to venture a guess, this kit will probably end up around the $199-209 price point. That timing difference, while it looks insignificant, is actually quite acute and indicates this kit has stronger ICs and according to G.Skill, it has more overclocking headroom.
Then you need to consider how much quad channel is worth it to you. As you can see, in everything other than synthetic testing and SuperPi 32M, most everything else doesn’t yet take advantage of all of this bandwidth. If you are, for instance, migrating to this platform from a socket 1366 system and already have strong triple-channel RAM, you might want to try them out in this platform and see what you think (pay attention to your manual on how to run dual- or triple-channel RAM on your X79 board). Likewise, if you have a stellar dual channel kit that runs very high speeds with tight latencies, try them out.
What your older dual- and triple-channel kits won’t get you is capacity. Recently dual channel kits have started migrating to 2 x 4 GB configurations, but in their heyday most performance triple channel kits were 3 x 2 GB, less than half the capacity of this kit.
If you find yourself migrating to the X79 platform, this G.Skill Ripjaws Z offering is a good way to get copious amounts of fast RAM with solid timings in a matched set, with some overclocking ability thrown in for good measure – all of the things required to make it Overclockers Approved.
- Jeremy Vaughan (hokiealumnus)