Kingston KHX13000D3LLK2/2GX DDR3-1625 Evaluation

The world of RAM is an interesting one. A world that seems to recreate itself over and over again with each new version. Companies fight and struggle to keep the performance edge. Who has the best speed? Who has the flashiest heat spreaders? Who can take more voltage? But you know, in the end the question most often asked (by the consumer)”which brand will work in my board?” As mainboards are tweaked by the manufacturers to give them a better performance edge we end up with compatibility issues with RAM. One company that seems to consistently dodge the compatibility bullet is Kingston. For some reason Kingston has been able to regularly provide RAM that performs well and just plain works. We have had many of their products her on Planetx64 in the past and now are taking a look at another one. A pair of PC3-13000 2GB kits (KHX13000D3LLK2/2GX), will they just work and provide us with the power and performance we want? Follow along and find out.

Product: Kingston KHX13000D3LLK2/2GX      
Author: Sean Kalinich
Reviewed on: May 1st 2008
Product cost: $335.00 (per kit)
Manufacturer: Kingston
Spelling and Grammatical editor: Jason Hambly, Paul Mercer

{mospagebreak title=Packaging and Construction}
Packaging:
RAM packaging is usually never flashy, it does not need to be. After all, its only purpose is to get you the product in one piece. The only reason we cover this is that there are still manufacturers that have very shoddy packaging even here. Kingston’s is not flashy, it is study and functional the only ornamentation (other than a utilitarian label) is a small sticker proclaiming support for the Core 2 Extreme. But we are not interested in the box, we are eager to find out more about what is inside partially visible under the clear plastic lid.

Construction:
The Kingston HyperX kits that I received were much flashier than I am used to. They still had the Neon Blue heat spreaders but these had the upper portion and one corner cut with diagonal grooves that gave it more visual appeal. The same cut had been used on the HyperX logo. This cut gave the look of transparency to this section of the heat spreader. I do not know if it will help with cooling or not, but it is still a nice touch. As to what chips Kingston is using under these attractive heat spreaders I do not know.
I was not going to run the risk of damaging them by removing the heat spreaders to look either.

Specifications:
The Kingston KHX13000D3LLK2/2GX kits have the following specs:
1.5V
533MHz fCK for 1066MB/Sec/pin
8 independent internal banks
Programmable CAS Latency: 5,6,7,8,9,10
Posted CAS
Programmable Additive Latency: 0, CL – 2, or CL – 1 clock
Programmable CAS Write Latency(CWL) = 7(DDR3-1066)
8-bit pre-fetch
Burst Length: 8 (Interleave without any limit, sequential with starting address “000” only), 4 with tCCD = 4 which does not allow seamless read or write [either on the fly using A12 or MRS]
Bi-directional Differential Data Strobe
Internal(self) calibration : Internal self calibration through ZQ pin (RZQ : 240 ohm ± 1%)
On Die Termination using ODT pin
Average Refresh Period 7.8us at lower then TCASE 85°C, 3.9us at 85°C < TCASE . 95°C
Asynchronous Reset
1066Mbps CL7 doesn’t have backward compatibility with 800Mbps CL5
PCB : Height 1.180” (30.00mm), single sided component

Performance Specs:
CL(IDD) 7 cycles
Row Cycle Time (tRCmin) 50.63ns (min.)
Refresh to Active/Refresh Command Time (tRFCmin) 110ns
Row Active Time (tRASmin) 37.5ns (min.)
Power TBD W (operating per module)
UL Rating 94 V – 0
Operating Temperature 0o C to 85o C
Storage Temperature -55o C to +100o C

These kits also support Intel’s XMP (Extreme Memory Profiles) and are certified to run at 1625MHz with timings of 7-7-7 all of this at 1.9v

{mospagebreak title=Performance}
Performance:
Test system
Intel Core 2 Quad Extreme 9770
Intel X48BT2 mainboard
Kingston KHX13000D3LLK2/2GX (2 Kits) 4GB RAM
2x Western Digital 74GB 10k RPM raptors (Stand Alone Drives)
Foxconn 8800GTX 768MB
16x Sony DVD-ROM
Generic 1.44 Floppy
Mushkin 1.1KW PSU
Corsair Nautilus 500 external water cooler
HSPC TechStation
Microsoft Windows Vista x64
nVidia 175.12 drivers

I will be following up this review with a full one on the X48BT2.
My testing was run using the XMP profile that is embedded in the KHX13000D3LLK2 modules.
This brought the RAM up to 1600MHz 1.9 vDIMM, 7-7-7-20.

Overclocking:
Overclocking was an interesting project with the Kingston KHX13000D3LLK2/2GX, not due to any failure of the memory, but due to issues with my choice of mainboard. The Intel X48BT2 did not like to overclock the memory and wanted to use dividers when I would push it beyond 410MHz. The really sticky part was that even under 460MHz it would show dividers but was not using them as the system BIOS and performance showed. I did not notice this until I saw that CPUz and Everest were reporting incorrect timings on the RAM. I started looking further into this and found that once the board thought it was using a divider it would keep displaying that divider. I contacted Intel and they sent me a new BIOS for the X48BT2 that corrected this issue. Also on this board you have to leave the RAM core speed set to “Default” this allows the memory to keep pace with the FSB. Otherwise it is going to try and keep it close to the speed you select using dividers.
Once I found the “sweet spot” I was able to get the KHX13000D3LLK2 running happily at 450 (1800) with timings of 8-8-8-22 and a voltage of 1.98, not a great OC for this RAM but since the majority of my crashes about this speed were related to the ICHR9 I think that with a more overclocking oriented board you will get better results

I backed the multiplier on the CPU to seven; this approximated the normal speed of the CPU.

Everest:
Everest is a good tool for basic benchmarking and for seeing what your system is doing.
For this I ran the Memory and Cache benchmark to get the potential read and write as well as latency for the Kingston memory. This was run with two and four memory modules to see what, if any, performance impact this would have.

My testing with Everest was not an overly exciting thing. The bandwidth was marginally better than what you would expect top end DDR2. With latency being an issue here, granted I am running a quad core CPU and 4 1GB sticks of memory and these two factors may account for the latency issues. With the increase to 1800MHz the bandwidth jumped quite a bit we saw and almost 30% increase with only a 12% overclock.

Sisoft Sandra:
Sandra from Sisoft is another catalog/benchmark application that allows you to get a good idea of what your system has in it and its potential.

Sandra confirms what we saw with Everest again the bandwidth is not much more than top end DDR2. Sandra also shows the large increase in bandwidth we get from the small 200MHz overclock.

PCMark Vantage:
FutureMark’s PCMark Vantage is another in our synthetic tests for memory. It is a series of tests that are designed to cover most of the things an average system would need to be able to do in today’s computer environment.

For the 32-bit suite I saw a little less than a 2% increase, although this still equals a 100 point difference and nothing to sneeze at. For the 64bit suite the performance gap was more pronounced with a difference of over 200 points this is a good indication of how much more you can get from running 64-bit optimized applications

3DMarkVantage:
Another synthetic from FutureMark 3DMark Vantage moves graphical testing into the DX10 realm.
By adding DX10 this test now increases the demand on a system including the memory in the system, so it makes for a good test of how efficient your memory sub-system is.

3DMark Vantage did not show much of an increase at all the 12% overclock here yielded a 3% increase in CPU score.
 
CineBench 10:
As a CPU ad GPU based rendering benchmark CineBench R10 also can show us performance of the memory sub-system.

With Cinebench R10 x64 we see very small gains in CPU rendering performance with only a 20 point difference in single CPU mode and 153 point in Multi CPU mode The place we see the memory performance difference is in the OpenGL rendering. Here we see a difference of 732 points this is a 15% performance increase. Not too bad for only a 12% OC.

HyperPi 0.99b
As I have mentioned in just about all of my evaluations, I like SuperPi and now this new version HyperPi. It is a very good test for stability of a system and isolates system performance between the RAM, CPU and HDD.  SuperPi seems to do this like no other application I have used. For my testing I ran the 16M run of HyperPi.

HyperPi is very memory dependent and a little extra bandwidth goes a long way. With our 12% overclock we managed to shave a full minute off of the 16M HyperPi run.

DiVX Encode (AutoGK)
Video Encoding is also a memory dependant function, buffering quite a bit of data for execution by the CPU then buffering it for writing to the HDD.
For my testing I use AutoGK 2.4 and a source DVD of my own personal footage that is roughly the length of a standard movie. The footage was edited in FinalCut HD and encoded to DVD using DVDStudio Pro HD.

{mospagebreak title=Conclusion}
Conclusion:
I was fairly impressed with the performance of the Kingston KHX13000D3LLK2/2GX kits that I tested. While they are not heart stopping by any means they do show a performance increase over the fastest stock DDR2 and also seem to improve overall system performance. The heatspreaders did a excellent job and even with over 2 volts pushed into them they were never hot nor burned my fingers. Unlike some of the FB-DIMMs I have used. Overclocking was a little of a letdown but I feel like the lackluster overclocking was more the fault of the Intel X48BT2 than the Kingston RAM. Price is a small issue but one that I feel will come down (as has all other memory in the past) with time.

Acknowledgments:
I would like to thank Peggy Kelly at Kingston for her help in putting this review together.