Archive Home arrow Reviews: arrow Memory arrow G.Skill RipJaws DDR3-1600 CL7 Memory Kit
G.Skill RipJaws DDR3-1600 CL7 Memory Kit E-mail
Reviews - Featured Reviews: Memory
Written by Bruce Normann   
Friday, 18 December 2009
Table of Contents: Page Index
G.Skill RipJaws DDR3-1600 CL7 Memory Kit
Closer Look: G.Skill RIPJAWS DDR3-1600
RAM Testing Methodology
Performance Test Results
Final Thoughts and Conclusion

Performance Test Results

Four benchmark applications for memory performance have been in rotation here at Benchmark Reviews for some time now, and there are no new contenders that offer any more or better information: Passmark Performance Test, Lavalys EVEREST, SiSoftware Sandra, and Crysis. The first three are synthetic benchmark suites specifically targeted at several aspects of memory performance. Each one has a unique approach, which provides a diverse set of measurements so that performance trends are brought to light. The last benchmark, Crysis, offers insight into how memory performance affects a gaming application that stresses the CPU and memory almost as much as it does the graphics subsystem. CPU speed is always a factor in memory tests, and we did our best to eliminate it as a variable. During overclocking, we had to adjust the Northbridge clock frequency, which has a halo effect on the overall system, but we were able to keep the CPU clock the same.

G.Skill_RIPJAWS_Dual-Channel_DDR3-1600_CL7_Memory_Kit_Passmark_Performance_Test.jpg

In Passmark Performance Test, there were very minimal gains, either from higher clock frequencies or tighter timings. The cached memory read test saw literally no difference between the five tested configurations; any differences are buried by experimental error. The uncached read test scored less than 1% improvement between the 1066 MHz and 1744 MHz settings. One of the nice aspects of this benchmark is the consistency of the results, I feel confident that even the small improvement measured here is real and repeatable.

The write performance was the bright spot of this test, clocking in a 3% gain as clock speed increased. Once again, the results were very consistent for this test, and while 3% may not seem like a lot, at least it is real, measureable and repeatable. I am hoping for more differentiation in the remaining tests, though.

EVEREST Ultimate Edition offers three simple memory bandwidth tests that focus on the basics; Read, Write, and Copy. In order to avoid concurrent threads competing over system memory bandwidth, the Memory benchmarks utilize only one processor core and one thread.

G.Skill_RIPJAWS_Dual-Channel_DDR3-1600_CL7_Memory_Kit_Everest_Mem.jpg

The Everest Read benchmark measures the maximum achievable memory read bandwidth. The code behind this benchmark method is written in Assembly and it is extremely optimized for every popular AMD and Intel processor core variants by utilizing the appropriate x86, MMX, 3DNow!, SSE, SSE2 or SSE4.1 instruction set extension. The benchmark reads a 16 MB sized, 1 MB aligned data buffer from system memory into the CPU. Memory is read in forward direction, continuously without breaks.

In Lavasys Everest we see more dramatic performance differences between the speed settings, and we can also see the effect of timings. From best to worst, there is a 25% improvement in read performance. We can also see how the tighter timings that were achieved at 1333 MHz almost made up the difference in speed between 1333 and 1600 MHz. It's also interesting to note that the timing changes at 1066 MHz made very little difference.

The Everest Write benchmark measures the maximum achievable memory write bandwidth. The code behind this benchmark method is written in Assembly and it is extremely optimized for every popular AMD and Intel processor core variants by utilizing the appropriate x86, MMX, 3DNow!, SSE or SSE2 instruction set extension. The benchmark writes a 16 MB sized, 1 MB aligned data buffer from the CPU into the system memory. Memory is written in forward direction, continuously without breaks.

The write performance is relatively flat, as speed settings increase, until we get to the overclocked configuration, where we were able to bump up the memory clock by increasing the Front Side Bus (FSB) by 9%, from 200 to 218 MHz. We reduced the CPU multiplier to keep the CPU clock the same, but as most people know, increasing the FSB clock makes almost everything faster. In fact, the best performance is usually achieved by pushing the FSB even higher and using a lower FSB:DRAM strap. But that's not a fair way to test memory products...

The Everest Copy benchmark measures the maximum achievable memory copy speed. The code behind this benchmark method is written in Assembly and it is extremely optimized for every popular AMD and Intel processor core variants by utilizing the appropriate x86, MMX, 3DNow!, SSE, SSE2 or SSE4.1 instruction set extension. The benchmark copies an 8 MB sized, 1 MB aligned data buffer into another 8 MB sized, 1 MB aligned data buffer through the CPU. Memory is copied in forward direction, continuously without breaks.

Copy performance was influenced the most by cranking up the memory clocks. We achieved a 38% increase in performance on this benchmark, which seemed to depend mostly on clock speed and less on clock timings. Overall, there were some significant performance gains to be had in the Everest set of benchmark tests. Not bad for a product that is supposedly optimized for a completely different operating environment. So far I see no reason that these new, low voltage RAM sets can't be used to good effect on the "old" AMD platform.

G.Skill_RIPJAWS_Dual-Channel_DDR3-1600_CL7_Memory_Kit_SiSoft_Sandra.jpg

Sandra is based on STREAM, a popular memory bandwidth benchmark that has been used on personal computers to super computers. It measures sustained memory bandwidth not burst or peak. Therefore, the results may be lower than those of other benchmarks. STREAM 2.0 uses static data (about 12M) - Sandra uses dynamic data (around 40-60% of physical system RAM). This means that on computers with fast memory Sandra may yield lower results than STREAM. It's not feasible to make Sandra use static RAM - since Sandra is much more than a benchmark, thus it would needlessly use memory.

A major difference is that Sandra's algorithm is multi-threaded on SMP/SMT systems. This works by splitting the arrays and letting each thread work on its own bit. Sandra creates a thread for each CPU in the system and assigns each thread to an individual CPU. Another difference is the aggressive use of scheduling/overlapping of instructions in order to maximize memory throughput even on "slower" processors. The loops should always be memory bound rather than CPU bound on all modern processors.

The other major difference is the use of alignment. Sandra dynamically changes the alignment of streams until it finds the best combination, then it repeatedly tests it to estimate the maximum throughput of the system. You can change the alignment in STREAM and recompile - but generally it is set to 0.

The results from SiSoft Sandra look a lot like the Read performance results in Lavasys Everest. They scale more as a result from increasing clock speeds than clock timings. Interestingly, the Integer and Floating Point results are almost identical, and the individual results were also very consistent from run-to-run. The overclocked pair, running 1744 MHz at CL8 bested the 1066 MHz CL7 set by 50% in both tests. That's a pretty significant gain, and a testament to the strength of the memory controller built into the AMD Phenom II architecture.

G.Skill_RIPJAWS_Dual-Channel_DDR3-1600_CL7_Memory_Kit_Crysis_L-M-H_Resolution.jpg

Crysis needs no introduction on this website. It is well known as one of the most demanding benchmarks, and our move to DirectX 10 has only increased the overall difficulty of achieving reasonable frame rates at high resolutions. In this scenario, where we want to reduce the influence of the video card in the results, we are primarily interested in the low resolution tests, and minimizing the video processing that is handled by the graphics subsystem.

Starting on the right and moving to the left, we can see that at 1680x1050 and 1280x1024 resolutions, there are minimal differences in gaming performance with changes in memory. Concentrating on the lowest resolution we tested, 1024x768, there is a noticeable, 12 FPS difference in average frame rate between the lowest and highest performing memory configurations. I say noticeable, meaning that it is easily measured; I doubt that you or I could visually tell the difference between an average of 109 and 121 frames per second in Crysis.

Overall, the synthetic tests mostly showed measureable performance improvements from increased memory speeds and tighter timings. Our toughest gaming benchmark, in terms of CPU and memory usage only showed measureable changes at low resolution. But, as GPU power increases in the system, this influence will be felt at higher resolutions. Similarly, if you are still using DirectX 9, where the GPU has an easier task, the impact will be greater.

We're left with the question of value, then. How much difference does premium, high speed memory make, especially compared to investing money in other system components. Continue on to Final Thoughts for the answer to that question, and a discussion of how I really feel about XMP and other memory standards.



 

Comments 

 
# Mrphil ede 2010-07-10 08:37
I am awaiting delivery of an Asus P7P55D-E Premium m/b + 1156 i7-860 CPU, and memory chosen from the QVL, but is non compliant with the requirement to keep below 1.65 V on overclocking. Also it does not have XMP. I ahave three options, return the motherboard and prooc for an AMD kit (which I have not researched, or for an Intel 1366 based kit. A third option is to keep the m/b and proc and change for the F3-12800CL9D-8GBRL Ripjaws recommended by G.Skill for the job.
1) It is unclear whether or not the G.Skill F3-12800CL9D-8GBRL has XMP, can you shed any light on this please.This is further complicated that Intel limit the number of memory modules to one per channel for 16000 and/or XMP memory modules. The implications of this are not obvious.
2) Can I run more than one 16000 memory module in each channel if I underclock to 1333 ?
3) The Asus 1366 board QVL for the KHX1600C8D3K3_6GX memory states it can run with 2x 2x3 modules, contravening abovel X58 slot limits.
Report Comment
 
 
# I would swap the memory....BruceBruce 2010-07-10 12:44
The CPU and motherboard you have chosen are great, IMHO, there are very few reasons why anyone would move (up?) from an 1156-based mobo to an 1136 unit. The F3-12800CL9D-8GBRL Ripjaws will indeed work, and that part number referes to an 8GB kit of two DIMMS. Are you planning on 8GB of memory? If so, this is a good choice. Don't worry about the XMP profile, although I suspect they have one, based on the fact that theGSKILL syas it is for an Intel system. Even without it, you can still go into BIOS and get the same performance. Take a look at my recent review of the ASUS P7P55D-E Pro motherboard, where I used the memory from this review in a P55 system.

As for the number of memory sticks per channel, the two DIMM kit you mentioned will give you 8GB and run in Dual-Channel mode on your chosen mobo, so everything is good. If you need more than 8GB of RAM, you are doing something very speciallized, and I need to know what that is before making any recommendations.
Report Comment
 
 
# RE: I would swap the memory....Phil Ede 2010-07-10 15:27
Hi Bruce

I have read your m/b review, thank you for putting so much valuable data in one place, written in clear technical English.

The main reason for needing the memory and performance is for computing the performance of very large broadcast shortwave antennas, including the effects of powerlines. The conducting elements are cut into thousands of segments, and the amount of memory needed goes up as the square of the number of segments. Also, I like to put the "engine" (a legacy FORTRAN program) into an optimisation loop where it may run the model thousands of times making small changes to find the optimum design, less memory needed and more speed, with a heavy load on the processing. Using the Ripjaws I would hope to be able to duble the memory to 16GB later if necessary. In both cases paging to disc takes time. I have run other prgrams (at work) running on Unix machines (the first to operate 64 bit) which can take days to complete.

Thanks for the comments, I get the feeling that with the advent of nanotechnology and scouring the memory specs I am aware of the effect of industrial inertia, with many companies finding it difficult to keep up. The Asus QVL is full of anomalies and I am not sure how far to trust the Taiwanese based G.Skill, high on technology, but what about quality control and reliability?

The alternative board would have been ne of the PX58D boards. One oddity is that every page I read introduces the product as new or latest, which it might have been when it was written, but without a timeline it is not clear what really is the "latest" and whether or not the second latest may not have benefits such as fitting in better with industrial inertia, a point in question being DirectX11 which is supposed to "come with Windows 7"

Have you ever written any reviews with view to servicing the mathematical and scientific community ? I also run other legacy FORTRAN programs for ionospheric and mathematical modelling. I know a Senior University lecturer whose students run TC3D (Systat software) for their PhD projects, that gives equations to fit data representing the state of the ionsphere at any place and time. The result may take the form of 100 order or so polynomial (with 100 sines and cosines in it). This then has to be inserted into the program code, compiled, and run. This is fine for some academic case study (perhaps even using Martian data) but useless for me as I need to model the ionosphere (here on earth) for any place, time, month, and year in an 11 year solar cycle. The sheer thought of putting TC3D plus a command line FORTRAN compiler inside an optimisation loop, within a C++ program, even with the required source data, would be enough to scare the most hardened gaming overclocker out into the garden shed to pot up petunias. Then, as a colleague once put it "memory is cheap and life is short" Put that way, 24GB of ram running with a CPU that clocks bits faster than the carrier frequency of a state of the art defence radar pulse is worth every penny if it gets results required by humble penniless research students struggling under the fundopause (the height above the earth at which space funding stops).

Many thanks

Phil Ede
Report Comment
 
 
# I see now.....BruceBruce 2010-07-10 18:42
Thanks for the explanation, it really helps. If your appplication is using the page file on the HDD, and more memory will eliminate that, then more memory is the biggest bang for the buck. The X58 motherboards certainly offer the simplest way of bulking up on memory. As for AMD vs Intel, Olin did some interetsing benchmarks with CADD applications in his AMD Phenom-II X6-1090T Black Edition Processor review on this site. Look on the "Processor" page, links on the left... Also check put the Everest benchmarks in the same review. They may be the most relevant benchmarks from the ones most review sites commonly use.

Don't worry one bit about GSKILL quality. They are one of the best and have a vey strong and loyal following in the enthusiast community. They are not the cheapest product, in general.
Report Comment
 
 
# RE: G.Skill RipJaws DDR3-1600 CL7 Memory KitRealNeil 2010-09-29 04:24
Hello Bruce,
I have a Gigabyte MA790GPT-UD3H AM3 Mainboard that immediately saw the Intel XMP profile on my 8GB (4-2GB sticks) of G.Skill 1600 RipJaws RAM in BIOS, and set all of the timings properly when I selected it. The X3-720 CPU runs fine setup that way.
Would this compatibility be because of a little extra effort by Gigabyte to use the XMP profiles?
Report Comment
 
 
# RE: RE: G.Skill RipJaws DDR3-1600 CL7 Memory KitOlin Coles 2010-09-29 05:12
XMP is a function added to the BIOS, and firmware updates also update the profile list. I'm a fan of using the latest motherboard firmware, especially on Gigabyte and ASUS motherboards.
Report Comment
 
 
# COMPATIBILITY (G.Skill RipJaws DDR3-1600 CL7 Memory Kit )ash 2010-10-01 09:05
Hi Bruce,

I have an ASUS P7H55-M and a Core I5-760, I want to buy this memory kit (G.Skill RipJaws DDR3-1600 CL7 Memory Kit ), however this is not included in the QVL of my motherboard... does this mean, i cannot use this kit? isn't this compatible with my system?
Report Comment
 
 
# COMPATIBILITY (G.Skill RipJaws DDR3-1600 CL7 Memory Kit )ash 2010-10-01 09:15
also, I had a look at Microsoft's page for the i5 760 and it says-

Memory Specifications
Max Memory Size
(dependent on memory type) 16 GB
Memory Types DDR3-1066/1333
# of Memory Channels 2

It says it only supports 1066 and 1333 RAM. Does that mean I have to get different RAM? or can I still use the 1600 RAM but just at 1333 speeds? or can i still use it for 1600mhz with just a few tweaks on the settings of my mobo?
Report Comment
 
 
# No Guarantees, but...BruceBruce 2010-10-01 11:01
The F3-12800CL7D-4GBECO memory that IS listed on the QVL list means that ASUS qualified that memory at the exact same speeds and timings. The only difference is the voltage that it took to achieve it. The "ECO" series runs at lower voltages (~1.35V) than the older, Ripjaws models, which need ~1.65V.

Yes, it will take some work in the BIOS settings to get the full performance from these DIMMs, but the board will definitely support those speeds and timings. XMP is supposed to make it "easy", but I've only been able to get high performance memory to run at max settings, with Standard BIOS settings about 50% of the time. The other 50% of the time, I have to go in and tweak something to make it run reliably at those speeds and timings.

Bottom line: don't be afraid to buy them, but be prepared for the possibility of having to do some work.
Report Comment
 
 
# guide.ash 2010-10-04 02:52
thanks for the help. one more thing... do you have tutorials or guide on how to work on this? i really need help on this, i'm kinda new at this.. thanks..
Report Comment
 
 
# RE: guide.Olin Coles 2010-10-04 06:46
We have two guides that can help:

benchmarkreviews.com/index.php?option=com_content&task=view&id=162&Itemid=38

benchmarkreviews.com/index.php?option=com_content&task=view&id=583&Itemid=38
Report Comment
 
 
# Wrong Test Timingspswfps 2010-12-10 08:47
Hi, I think you've set the tRC too low. Try setting it to 40 clocks instead of 32. I have found these sticks perfectly stable on my AMD rig @ 1600MHz 7-8-7-24-40 1T using 1.615V.
Great sticks for the price.
Report Comment
 
 
# They're in a better place now....BruceBruce 2010-12-10 09:56
I'm now using these sticks on an Intel P55 Express system, and they are running fine at 7-8-7-24. My long-term opinion is the same, they're very nice, and a good price.
Report Comment
 
 
# Correct Test Timings...pswfps 2010-12-10 11:59
Cool. Your P55 board has probably just picked up the correct tRC of 40 from the SPD settings on the DIMMS, as per your CPU-Z screenie. I had to set it manually on my 890GX board.
Report Comment
 

Comments have been disabled by the administrator.

Search Benchmark Reviews
QNAP Network Storage Servers

Follow Benchmark Reviews on FacebookReceive Tweets from Benchmark Reviews on Twitter