Kingston low-latency 1375MHz CL7 DDR3 Memory Review

 

As one of the largest memory manufacturers around, Kingston is usually one of the first companies to produce products based on the latest technologies and with the arrival of the P35 chipset based motherboards bringing DDR3 support, it wasn’t long before Kingston announced both ValueRAM and premium HyperX parts running at 1066MHz and 1375MHz respectively.

PlanetX64 have been fortunate enough that Kingston decided to send us one of their HyperX KHX11000D3LLK2/2G kits for testing, so let’s get on with it.

Product: Kingston HyperX KHX11000D3LLK2/2G
Author: Paul Mercer
Category: Enthusiast/Overclocker
Date: September 2007
Product cost: MSRP $ 518.00 US
Manufacturer: Kingston
Spelling and Grammatical Editor: PlanetX64 Staff

{mospagebreak title=Features}

 

KHX11000D3LLK2/2G 2GB (1GB 128M x 64-Bit x 2 pcs.) PC3-11000 CL7 240-Pin DIMM Kit

DESCRIPTION:

(shamelessly copied from the Kingston website) 

Kingston’s KHX11000D3LLK2/2G is a kit of two 128M x 64-bit 1GB (1024MB) DDR3-1375 CL7 SDRAM (Synchronous DRAM) memory modules, based on sixteen 64M x 8-bit DDR3 FBGA components per module. Each module pair has been tested to run at DDR3-1375MHz at a low latency timing of 7-7-7 at 1.7V. The SPDs are programmed to JEDEC standard latency DDR3-1066Mhz timing of 7-7-7 at 1.5V. Each 240-pin DIMM uses gold contact fingers and requires +1.5V.

The JEDEC standard electrical and mechanical specifications are as follows:

FEATURES:

JEDEC standard 1.5V ± 0.075V Power Supply

VDDQ = 1.5V ± 0.075V

533MHz fCK for 1066Mb/sec/pin

8 independent internal bank

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), double sided component

PERFORMANCE:

CL(IDD) 7 cycles

Row Cycle Time (tRCmin) 50.63ns (min.)

Refresh to Active/Refresh Command Time (tRFCmin) 90ns

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

{mospagebreak title= Packaging}

 

The packaging of the new HyperX DDR3 modules is what I’ve come to expect from Kingston, a no-nonsense plastic package, easy to open, but strong enough to survive most shipping companies.

    

I have been informed that these kits use Elpida chips, but the heatspreaders on this kit were firmly attached, so I decided to leave them on there. (For now…)

 

The introduction of DDR3 has meant a redesign of IC’s and PCB to maintain the signal integrity at higher frequencies.This redesign also has its drawbacks though with increased cycle latencies.

Because of these increased internal latencies, we can safely assume that if we were to run DDR3 and DDR2 at the same speed and latencies, DDR3 should always be slower, and it’s only when the DDR3 frequencies are increased (counteracting the slower latencies and increasing the theoretical bandwidth) that the benefits can be seen. 

There’s a very good article on the tech repository by freecableguy, with additions from Tony, who can explain the relationship between timings and frequency a lot better than myself:

How memory frequency affects latency – Tighter timings vs. Higher speeds?

http://www.thetechrepository.com/showthread.php?t=160

and another good post on the OCZ forums by Tony:

http://www.ocztechnologyforum.com/forum/showthread.php?t=30816

{mospagebreak title= Performance}

 

Test System: Intel Core 2 Duo E6750
Asus P5K3 Deluxe WiFi – BIOS:0604
2GB Kingston HyperX PC3-11000 (2x1024MB)
Foxconn 7950GT 512MB
2x Hitachi Deskstar 80GB 7200RPM SATA2 HDD (RAID 0)
Generic DVD Rom Drive
Tagan 900W TurboJet PSU
Windows XP Professional x86 & x64

As per usual with these reviews, a clean installation of Windows was performed before benchmarking began, all updates were applied and the latest official drivers used. A number of software applications were also installed to simulate a new system owned by an average user and then drives were defragmented.

Software installed included the following:
Avast Antivirus
Microsoft Office 2007
Diskeeper 10.0
Adobe Acrobat Reader
CyberLink Power DVD
Firefox

For standard, stock speed testing, I ran the Kingston DDR3 at 1333MHz, although this is slightly below the manufacturer specified rating, testing was done in this manor because the 1375MHz/PC3-11000 rating for the memory is not an official JEDEC standard and would require the FSB to be increased slightly and the CPU overclocked.

1333MHz, 7-7-7-20, 1.70V 

Lavalys Everest:

 

SiSoftware Sandra XII:

Memory Latency 

Memory Bandwidth 

Cinebench 10:

 

Super PI:

Single 32M 

Windows XP x86	Windows XP x64
18m 18.844s	18m 14.156s

Dual 32M

Windows XP x86

Windows XP x64

19m 48.391s

19m 48.250s

19m 34.890s

19m 34.828s

 

PCMark05:

 

3DMark06:

 

The performance at stock speeds appears to be similar to mid range DDR2. But without a DDR2 P35 board available to me at this time, I was unable to give a direct comparison with some high-speed DDR2.

 

{mospagebreak title= Overclocking}

 

I decided to begin the overclocking tests on the Kingston DDR3 by seeing if they would run at a lower CAS latency. I started by trying to get a range of overclocks before running benchmarks, and after having to reinstall the OS a couple of times after pushing too hard, I ended up with the following ranges:

All testing was done with the BIOS setting for “DRAM Command Rate” set to 1T
(although it’s debatable whether it is truly 1T).

5-5-5-15	1.70V	533MHz - DDR 1066
5-6-5-15	1.65V	536MHz – DDR 1072
5-6-5-15	1.65V	550MHz – DDR 1100
5-6-5-15	1.70V	575MHz – DDR 1150
5-6-5-15	1.75V	600MHz – DDR 1200
5-6-5-15	1.80V	630MHz – DDR 1260
5-7-5-15	1.70V	640MHz – DDR 1280
5-7-5-15	1.70V	650MHz – DDR 1300
5-7-5-15	1.70V	666MHz – DDR 1333
5-7-5-15	1.75V	676MHz – DDR 1352
5-7-5-15	1.80V	688MHz – DDR 1376
5-7-5-15	1.80V	700MHz – DDR 1400
5-7-5-15	1.85V	710MHz – DDR 1420
5-7-5-15	1.90V	716MHz – DDR 1432
5-8-5-15	1.95V	726MHz – DDR 1452
5-8-5-15	1.95V	730MHz – DDR 1460
5-8-6-15	1.95V	736MHz – DDR 1472
5-8-6-15	1.95V	740MHz – DDR 1480
6-8-6-15	1.95V	746MHz – DDR 1492
6-8-8-15	1.95V	750MHz – DDR 1500

 

During the time I spent testing this memory I quickly realised that (with this kit) RAS to CAS Delay was more important for stability than the CAS latency.  At lower speeds, more voltage did not really help very much and usually increasing would only produce errors sooner. This left me without much choice other than to increase tRCD (RAS to CAS Delay) and/or tRP (RAS Precharge)

I do have screenshots for all of the 12H HCI Memtest sessions, but rather than bore everyone and fill a page with over 20 screenshots of pretty much the same thing, I’ll leave them off. If for whatever reason you would like to see any 12hour HCI Memtest screenshots or you have any other tests you would like to see, please feel free to make requests on the forums and I’ll do my best to fill them. 

Unfortunately, without a truly unlinked memory controller to allow the overclocking of memory frequency without touching the FSB, it is difficult to separate the performance increases due to memory from any performance increase due to overclocking the CPU, so please bear this in mind when viewing the overclocked results. 

 

1333MHz, 5-7-5-15, 1.70V

Lavalys Everest:

 

SiSoftware Sandra XII:

Memory Latency 

Memory Bandwidth 

 

Cinebench 10:

 

Super PI:

Single 32M 

Windows XP x86	Windows XP x64
18m 13.000s	18m 04.343s

Dual 32M

Windows XP x86

Windows XP x64

Easy AdSense by Unreal

19m 36.141s

19m 35.547s

19m 27.672s

19m 27.235s

 

PCMark05:

 

3DMark06:

 

 

1375MHz, 7-7-7-20, 1.70V

Lavalys Everest:

SiSoftware Sandra XII:

Memory Latency 

Memory Bandwidth 

 

Cinebench 10:

 

Super PI:

Single 32M 

Windows XP x86	Windows XP x64
18m 01.156s	17m 43.609s

Dual 32M

Windows XP x86

Windows XP x64

19m 22.453s

19m 21.343s

19m 07.079s

19m 02.031s

 

PCMark05:

 

3DMark06:

 

 

1375MHz, 5-7-5-15, 1.80V

 

Lavalys Everest:

SiSoftware Sandra XII:

Memory Latency 

Memory Bandwidth 

Cinebench 10:

 

Super PI:

Single 32M 

Windows XP x86	Windows XP x64
17m 46.828s	17m 32.406s

Dual 32M

Windows XP x86

Windows XP x64

18m 57.531s

18m 58.000s

18m 48.625s

18m 48.562s

 

PCMark05:

3DMark06:

 

As you can see from the results, these particular sticks would have probably gone into the higher binned 1375MHz 5-7-5-15 HyperX kits from Kingston, but failed the voltage binning. They easily surpassed the manufacturer specified speeds and timings, but as always YMMV.

 

{mospagebreak title=Maximum Overclock}

Although these sticks were quite stable at 1500MHz, the memory refused to run any higher even with timings relaxed to 8-8-8. The kit would pass a few benchmarks, but displayed errors quickly in Memtest. Maybe with more tweaking a few more MHz may be attained, there may have even been a little more scaling left in the voltage. But personally, I would feel a little uncomfortable running more than 1.9-1.95V through this DDR3 for 24/7 operation, so I believe this to be pretty much the limit for the kit I have here.

 

1500MHz, 6-8-8-18, 1.95V

Lavalys Everest:

Memory Bandwidth 

Cinebench 10:

Super PI:

Single 32M 

Windows XP x86	Windows XP x64
16m 29.422s	16m 14.515s

Dual 32M

Windows XP x86

Windows XP x64

17m 48.484s

17m 48.656s

17m 31.328s

17m 30.484s

 

PCMark05:

3DMark06:

 

{mospagebreak title= Price and Warranty}

As with all DDR3 at the moment, this kit is rather expensive for 2GB. But comparing to the prices and speed ratings from some of the other vendors first DDR3 kits, the Kingston HyperX actually seems fairly reasonable. Available for $478.49 from Newegg.com in the US. or at around the £300 mark in the UK (although finding some in stock is another question),

As always Kingston offers a lifetime guarantee for their HyperX memory modules, full details of which can be found here:

http://www.kingston.com/company/warranty.asp

{mospagebreak title= Conclusion}

There’s no doubt that at this time, DDR3 is still rather expensive and as such, the pricing puts it out of the reach of a lot of people. It’s ideal for the enthusiasts and overclockers out there though, as it allows you to use some of the latest and best boards out there like the Asus P5K3 Deluxe or the Asus Blitz Extreme.

This kit was quite tweakable and ideal for those people who love to spend time adjusting latencies and voltages to gain those last few points in benchmarks or increase memory overclocks. But the extra bandwidth offered by DDR3 may not be of much benefit to the Core2Duo systems as these Intel chips are not exactly bandwidth limited to begin with. Quad core chips on the other hand may see larger benefits with the extra cores needing constant feeding.

I personally enjoyed working with this kit and although it seemed to take a little time to get the full range of stable overclocked settings down, it was a nice intro to DDR3. With such a wide range of speeds and timings available to you, this memory shouldn’t hold back those CPU and FSB overclocks.

Although the results for DDR3 are nothing spectacular at the moment, things can only improve. As manufacturing processes are refined, faster latencies and higher frequencies are achieved and prices lower, DDR3 will become a mainstream part; and when this happens, it will become the better choice for users, not only because of higher speeds than DDR2 could ever dream of, but also the lower power consumption.

 Pros:
New technology
Low voltage
High frequency
Extra bandwidth
Overclockable

Cons:
Very expensive
Limited motherboard support